Biography
My research interests are in Computer-aided diagnosis and interventions. These include: Machine learning techniques for in silico inference and prediction Analysis of ultrasound images and signals for enhancement of cancer detection Image-aided, computer-assisted diagnosis of disease Ultrasound-guided interventions Knowledge discovery from high throughput biological data Quantitative modeling and reverse engineering of gene regulatory networks Analysis, segmentation and classification of fluorescence microscopy images Chromosome and cell imaging.
Affiliations
- Professor, School of Computing
- Member, College of New Scholars, Artists and Scientists, Royal Society of Canada
- c/a Department of Electrical and Computer Engineering
- c/a Human Mobility Research Centre, Kingston General Hospital
Admissions
The Med-I Lab is always looking for the brightest students to join us. We continually have openings for high-school, undergraduate, graduate, and postdoctoral fellowship positions.
Publications
2026
Wilson, Paul FR; Harmanani, Mohamed; Guo, Zhuoxin; Dzikunu, Obed K; Cash, Hannes; Kinnaird, Adam; Wodlinger, Brian; Abolmaesumi, Purang; Mousavi, Parvin
Compass: Prostate Cancer Detection Needs Multi-View Context Journal Article
In: arXiv preprint arXiv:2607.06919, 2026.
@article{wilson2026,
title = {Compass: Prostate Cancer Detection Needs Multi-View Context},
author = {Paul FR Wilson and Mohamed Harmanani and Zhuoxin Guo and Obed K Dzikunu and Hannes Cash and Adam Kinnaird and Brian Wodlinger and Purang Abolmaesumi and Parvin Mousavi},
year = {2026},
date = {2026-01-01},
journal = {arXiv preprint arXiv:2607.06919},
abstract = {Artificial intelligence (AI) analysis of micro-ultrasound (US) has shown promise for prostate cancer (PCa) detection. However, most existing AI methods focus on the analysis of single US images in isolation. By contrast, expert US readers typically assess a full recorded video study, which provides three-dimensional context, to improve PCa detection compared to single-frame analysis. Inspired by this clinical workflow, we propose Compass, a novel AI methodology which models a US study as a stream of 2D images. Compass jointly integrates rotational sweep videos of the prostate with US frames acquired at the moment of biopsy, and performs evidence aggregation across the study using a transformer conditioned on the probe's rotational angle. Finally, a decoder head predicts frame-level and study-level risk scores for the patient. The model is trained and evaluated using a multi-center clinical trial dataset of US studies, including continuous rotational scans of the prostate and videos captured during biopsy acquisition. We compare the proposed method to baseline AI methods from the literature and to risk scores provided by clinical experts. Our framework shows strong performance, highlighting the value of multi-view context for US PCa detection, and providing a potentially powerful tool to complement human expertise in US-based PCa diagnosis. Our code is available at: https://github.com/mharmanani/Compass.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Artificial intelligence (AI) analysis of micro-ultrasound (US) has shown promise for prostate cancer (PCa) detection. However, most existing AI methods focus on the analysis of single US images in isolation. By contrast, expert US readers typically assess a full recorded video study, which provides three-dimensional context, to improve PCa detection compared to single-frame analysis. Inspired by this clinical workflow, we propose Compass, a novel AI methodology which models a US study as a stream of 2D images. Compass jointly integrates rotational sweep videos of the prostate with US frames acquired at the moment of biopsy, and performs evidence aggregation across the study using a transformer conditioned on the probe's rotational angle. Finally, a decoder head predicts frame-level and study-level risk scores for the patient. The model is trained and evaluated using a multi-center clinical trial dataset of US studies, including continuous rotational scans of the prostate and videos captured during biopsy acquisition. We compare the proposed method to baseline AI methods from the literature and to risk scores provided by clinical experts. Our framework shows strong performance, highlighting the value of multi-view context for US PCa detection, and providing a potentially powerful tool to complement human expertise in US-based PCa diagnosis. Our code is available at: https://github.com/mharmanani/Compass.
Wilson, Paul FR; Harmanani, Mohamed; To, Minh Nguyen Nhat; Jamzad, Amoon; Elghareb, Tarek; Guo, Zhuoxin; Kinnaird, Adam; Wodlinger, Brian; Abolmaesumi, Purang; Mousavi, Parvin
ProstNFound+: a prospective study using medical foundation models for prostate cancer detection Journal Article
In: International Journal of Computer Assisted Radiology and Surgery, vol. 21, no. 4, pp. 693-702, 2026.
@article{wilson2026a,
title = {ProstNFound+: a prospective study using medical foundation models for prostate cancer detection},
author = {Paul FR Wilson and Mohamed Harmanani and Minh Nguyen Nhat To and Amoon Jamzad and Tarek Elghareb and Zhuoxin Guo and Adam Kinnaird and Brian Wodlinger and Purang Abolmaesumi and Parvin Mousavi},
year = {2026},
date = {2026-01-01},
journal = {International Journal of Computer Assisted Radiology and Surgery},
volume = {21},
number = {4},
pages = {693-702},
publisher = {Springer International Publishing},
abstract = {Purpose
Medical foundation models (FMs) offer a path to build high-performance diagnostic systems. However, their application to prostate cancer (PCa) detection from micro-ultrasound (US) remains untested in clinical settings. We present ProstNFound+, an adaptation of FMs for PCa detection from US, along with its first prospective validation.
Methods
ProstNFound+ incorporates a medical FM, adapter tuning, and a custom prompt encoder that embeds PCa-specific clinical biomarkers. The model generates a cancer heatmap and a risk score for clinically significant PCa. Following training on multicenter retrospective data, the model is prospectively evaluated on data acquired five years later from a new clinical site. Model predictions are benchmarked against standard clinical scoring protocols (PRI-MUS and PI-RADS).
Results
ProstNFound+ shows strong generalization to the prospective data, with no …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Purpose
Medical foundation models (FMs) offer a path to build high-performance diagnostic systems. However, their application to prostate cancer (PCa) detection from micro-ultrasound (US) remains untested in clinical settings. We present ProstNFound+, an adaptation of FMs for PCa detection from US, along with its first prospective validation.
Methods
ProstNFound+ incorporates a medical FM, adapter tuning, and a custom prompt encoder that embeds PCa-specific clinical biomarkers. The model generates a cancer heatmap and a risk score for clinically significant PCa. Following training on multicenter retrospective data, the model is prospectively evaluated on data acquired five years later from a new clinical site. Model predictions are benchmarked against standard clinical scoring protocols (PRI-MUS and PI-RADS).
Results
ProstNFound+ shows strong generalization to the prospective data, with no …
Medical foundation models (FMs) offer a path to build high-performance diagnostic systems. However, their application to prostate cancer (PCa) detection from micro-ultrasound (US) remains untested in clinical settings. We present ProstNFound+, an adaptation of FMs for PCa detection from US, along with its first prospective validation.
Methods
ProstNFound+ incorporates a medical FM, adapter tuning, and a custom prompt encoder that embeds PCa-specific clinical biomarkers. The model generates a cancer heatmap and a risk score for clinically significant PCa. Following training on multicenter retrospective data, the model is prospectively evaluated on data acquired five years later from a new clinical site. Model predictions are benchmarked against standard clinical scoring protocols (PRI-MUS and PI-RADS).
Results
ProstNFound+ shows strong generalization to the prospective data, with no …
Abootorabi, Mohammad Mahdi; Saghafian, Armin; Bazshoushtari, Ali; Rezaei, Hamid; Hwang, EunJeong; Shwartz, Vered; Mousavi, Parvin; Abolmaesumi, Purang
Steering Geometry: Validating Human Value Geometry in LLM Steering Space Journal Article
In: arXiv preprint arXiv:2609.06289, 2026.
@article{abootorabi2026,
title = {Steering Geometry: Validating Human Value Geometry in LLM Steering Space},
author = {Mohammad Mahdi Abootorabi and Armin Saghafian and Ali Bazshoushtari and Hamid Rezaei and EunJeong Hwang and Vered Shwartz and Parvin Mousavi and Purang Abolmaesumi},
year = {2026},
date = {2026-01-01},
journal = {arXiv preprint arXiv:2609.06289},
abstract = {As large language models (LLMs) are increasingly deployed in alignment-sensitive contexts, activation steering has emerged as a lightweight, inference-time alternative to fine-tuning methods (e.g., RLHF, DPO) for behavioral control. However, existing work typically validates steering on isolated behaviors, leaving it unclear whether steering vectors encode coherent semantic structure or merely exploit behavior-specific shortcuts. We investigate whether the latent geometry of LLM steering vectors reflects theory-specified structure in human values and morality. Using Schwartz's Theory of Basic Human Values as our primary fine-grained framework, we introduce a 26K-sample benchmark covering 20 human values and analyze distribution-driven methods (e.g., CAA, SphericalSteer, ODESteer) and behavior-centric approaches (e.g., COLD-Steer, BiPO) across diverse model families and sizes. We find that distribution-driven methods recover human value topologies aligned with theoretical predictions (Spearman up to 0.51, ). In contrast, behavior-centric methods achieve comparable steering performance but show little correlation with the expected value geometry. Geometric fidelity improves with model scale but drops after instruction tuning. Finally, better geometric alignment also leads to more human-consistent transfer across values: steering one value correctly lifts compatible values and suppresses opposing ones. Code and data are available at: https://github.com/DeepRCL/Steering_Geometry.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
As large language models (LLMs) are increasingly deployed in alignment-sensitive contexts, activation steering has emerged as a lightweight, inference-time alternative to fine-tuning methods (e.g., RLHF, DPO) for behavioral control. However, existing work typically validates steering on isolated behaviors, leaving it unclear whether steering vectors encode coherent semantic structure or merely exploit behavior-specific shortcuts. We investigate whether the latent geometry of LLM steering vectors reflects theory-specified structure in human values and morality. Using Schwartz's Theory of Basic Human Values as our primary fine-grained framework, we introduce a 26K-sample benchmark covering 20 human values and analyze distribution-driven methods (e.g., CAA, SphericalSteer, ODESteer) and behavior-centric approaches (e.g., COLD-Steer, BiPO) across diverse model families and sizes. We find that distribution-driven methods recover human value topologies aligned with theoretical predictions (Spearman up to 0.51, ). In contrast, behavior-centric methods achieve comparable steering performance but show little correlation with the expected value geometry. Geometric fidelity improves with model scale but drops after instruction tuning. Finally, better geometric alignment also leads to more human-consistent transfer across values: steering one value correctly lifts compatible values and suppresses opposing ones. Code and data are available at: https://github.com/DeepRCL/Steering_Geometry.
Abootorabi, Mohammad Mahdi; Namazi, Sina; Saadat, Armin; Wang, Lyuyang; Dzikunu, Obed; Wilson, Paul FR; Guo, Zhuoxin; Wodlinger, Brian; Mousavi, Parvin; Abolmaesumi, Purang
Learning Where to Look: A Reinforcement Learning Framework for Robust Micro-Ultrasound Prostate Cancer Detection Journal Article
In: arXiv preprint arXiv:2606.30951, 2026.
@article{abootorabi2026a,
title = {Learning Where to Look: A Reinforcement Learning Framework for Robust Micro-Ultrasound Prostate Cancer Detection},
author = {Mohammad Mahdi Abootorabi and Sina Namazi and Armin Saadat and Lyuyang Wang and Obed Dzikunu and Paul FR Wilson and Zhuoxin Guo and Brian Wodlinger and Parvin Mousavi and Purang Abolmaesumi},
year = {2026},
date = {2026-01-01},
journal = {arXiv preprint arXiv:2606.30951},
abstract = {Micro-ultrasound (US) is a new, emerging, and promising imaging modality for prostate cancer (PCa) detection, but accurate identification of suspicious tissue remains highly dependent on clinical experience, leading to substantial inter-observer variability. Machine-learning assistance can reduce this variability; however, training reliable deep models is challenging because supervision is sparse and noisy – typically limited to core-level histopathology outcomes (e.g., cancer grade and its percentage in a biopsy core) without pixel-level lesion annotations and under severe class imbalance. We introduce Prost-RL, which reframes US PCa detection as a spatially aware, policy-driven inference problem by learning where to look before decoding. Prost-RL integrates a lightweight reinforcement-learning policy into a foundation-model encoder-decoder to generate interpretable spatial attention maps that act as soft prompts for both cancer-likelihood heatmap prediction and image-level classification. We further propose Adaptive Policy Optimization (APO) to stabilize hybrid supervised-RL training and a noise-robust objective combining symmetric cross-entropy with negative-entropy regularization to mitigate weak-label noise and encourage sharp localization. On a cohort of 6,607 biopsy cores from 693 patients across five clinical sites, Prost-RL achieves AUROC with % sensitivity at 80% specificity for core-level detection (+2.1 AUROC and +4.5 sensitivity points over the strongest baseline), and AUROC for clinically significant cancer classification. The learned policy highlights biopsy-aligned regions, providing …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Micro-ultrasound (US) is a new, emerging, and promising imaging modality for prostate cancer (PCa) detection, but accurate identification of suspicious tissue remains highly dependent on clinical experience, leading to substantial inter-observer variability. Machine-learning assistance can reduce this variability; however, training reliable deep models is challenging because supervision is sparse and noisy – typically limited to core-level histopathology outcomes (e.g., cancer grade and its percentage in a biopsy core) without pixel-level lesion annotations and under severe class imbalance. We introduce Prost-RL, which reframes US PCa detection as a spatially aware, policy-driven inference problem by learning where to look before decoding. Prost-RL integrates a lightweight reinforcement-learning policy into a foundation-model encoder-decoder to generate interpretable spatial attention maps that act as soft prompts for both cancer-likelihood heatmap prediction and image-level classification. We further propose Adaptive Policy Optimization (APO) to stabilize hybrid supervised-RL training and a noise-robust objective combining symmetric cross-entropy with negative-entropy regularization to mitigate weak-label noise and encourage sharp localization. On a cohort of 6,607 biopsy cores from 693 patients across five clinical sites, Prost-RL achieves AUROC with % sensitivity at 80% specificity for core-level detection (+2.1 AUROC and +4.5 sensitivity points over the strongest baseline), and AUROC for clinically significant cancer classification. The learned policy highlights biopsy-aligned regions, providing …
Abootorabi, Mohammad Mahdi; Mousavi, Parvin; Abolmaesumi, Purang; Shelhamer, Evan
ProtoTTA: Prototype-Guided Test-Time Adaptation Journal Article
In: arXiv preprint arXiv:2604.15494, 2026.
@article{abootorabi2026b,
title = {ProtoTTA: Prototype-Guided Test-Time Adaptation},
author = {Mohammad Mahdi Abootorabi and Parvin Mousavi and Purang Abolmaesumi and Evan Shelhamer},
year = {2026},
date = {2026-01-01},
journal = {arXiv preprint arXiv:2604.15494},
abstract = {Deep networks that rely on prototypes-interpretable representations that can be related to the model input-have gained significant attention for balancing high accuracy with inherent interpretability, which makes them suitable for critical domains such as healthcare. However, these models are limited by their reliance on training data, which hampers their robustness to distribution shifts. While test-time adaptation (TTA) improves the robustness of deep networks by updating parameters and statistics, the prototypes of interpretable models have not been explored for this purpose. We introduce ProtoTTA, a general framework for prototypical models that leverages intermediate prototype signals rather than relying solely on model outputs. ProtoTTA minimizes the entropy of the prototype-similarity distribution to encourage more confident and prototype-specific activations on shifted data. To maintain stability, we employ geometric filtering to restrict updates to samples with reliable prototype activations, regularized by prototype-importance weights and model-confidence scores. Experiments across four prototypical backbones on four diverse benchmarks spanning fine-grained vision, histopathology, and NLP demonstrate that ProtoTTA improves robustness over standard output entropy minimization while restoring correct semantic focus in prototype activations. We also introduce novel interpretability metrics and a vision-language model (VLM) evaluation framework to explain TTA dynamics, confirming ProtoTTA restores human-aligned semantic focus and correlates reliably with VLM-rated reasoning quality. Code is available at: https://github.com …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Deep networks that rely on prototypes-interpretable representations that can be related to the model input-have gained significant attention for balancing high accuracy with inherent interpretability, which makes them suitable for critical domains such as healthcare. However, these models are limited by their reliance on training data, which hampers their robustness to distribution shifts. While test-time adaptation (TTA) improves the robustness of deep networks by updating parameters and statistics, the prototypes of interpretable models have not been explored for this purpose. We introduce ProtoTTA, a general framework for prototypical models that leverages intermediate prototype signals rather than relying solely on model outputs. ProtoTTA minimizes the entropy of the prototype-similarity distribution to encourage more confident and prototype-specific activations on shifted data. To maintain stability, we employ geometric filtering to restrict updates to samples with reliable prototype activations, regularized by prototype-importance weights and model-confidence scores. Experiments across four prototypical backbones on four diverse benchmarks spanning fine-grained vision, histopathology, and NLP demonstrate that ProtoTTA improves robustness over standard output entropy minimization while restoring correct semantic focus in prototype activations. We also introduce novel interpretability metrics and a vision-language model (VLM) evaluation framework to explain TTA dynamics, confirming ProtoTTA restores human-aligned semantic focus and correlates reliably with VLM-rated reasoning quality. Code is available at: https://github.com …
Akbarifar, Faranak; Maghsoodi, Nooshin; Dukelow, Sean P; Scott, Stephen H; Mousavi, Parvin
Reducing robotic upper-limb assessment time while maintaining precision: a time series foundation model approach Journal Article
In: Journal of NeuroEngineering and Rehabilitation, 2026.
@article{akbarifar2026,
title = {Reducing robotic upper-limb assessment time while maintaining precision: a time series foundation model approach},
author = {Faranak Akbarifar and Nooshin Maghsoodi and Sean P Dukelow and Stephen H Scott and Parvin Mousavi},
year = {2026},
date = {2026-01-01},
journal = {Journal of NeuroEngineering and Rehabilitation},
publisher = {BioMed Central},
abstract = {Purpose
Visually Guided Reaching (VGR) on the Kinarm robot yields sensitive kinematic biomarkers but requires 40–64 reaches, imposing time and fatigue burdens. We evaluate whether time series foundation models can replace unrecorded trials from an early subset of reaches while preserving agreement with full-session estimates of standard Kinarm parameters.
Methods
We analyzed VGR speed signals from 461 stroke and 599 control participants across 4- and 8-target reaching protocols. We withheld all but the first 8 or 16 reaching trials and used ARIMA, MOMENT, and Chronos models, fine-tuned on 70% of participants, to forecast synthetic trials. We recomputed four kinematic features of reaching (reaction time, movement time, posture speed, max speed) on combined recorded plus forecasted trials and compared to full-length references using ICC(2,1).
Results
Chronos forecasts increased ICC values for …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Purpose
Visually Guided Reaching (VGR) on the Kinarm robot yields sensitive kinematic biomarkers but requires 40–64 reaches, imposing time and fatigue burdens. We evaluate whether time series foundation models can replace unrecorded trials from an early subset of reaches while preserving agreement with full-session estimates of standard Kinarm parameters.
Methods
We analyzed VGR speed signals from 461 stroke and 599 control participants across 4- and 8-target reaching protocols. We withheld all but the first 8 or 16 reaching trials and used ARIMA, MOMENT, and Chronos models, fine-tuned on 70% of participants, to forecast synthetic trials. We recomputed four kinematic features of reaching (reaction time, movement time, posture speed, max speed) on combined recorded plus forecasted trials and compared to full-length references using ICC(2,1).
Results
Chronos forecasts increased ICC values for …
Visually Guided Reaching (VGR) on the Kinarm robot yields sensitive kinematic biomarkers but requires 40–64 reaches, imposing time and fatigue burdens. We evaluate whether time series foundation models can replace unrecorded trials from an early subset of reaches while preserving agreement with full-session estimates of standard Kinarm parameters.
Methods
We analyzed VGR speed signals from 461 stroke and 599 control participants across 4- and 8-target reaching protocols. We withheld all but the first 8 or 16 reaching trials and used ARIMA, MOMENT, and Chronos models, fine-tuned on 70% of participants, to forecast synthetic trials. We recomputed four kinematic features of reaching (reaction time, movement time, posture speed, max speed) on combined recorded plus forecasted trials and compared to full-length references using ICC(2,1).
Results
Chronos forecasts increased ICC values for …
Maghsoodi, Nooshin; Nassar, Sarah; Wilson, Paul FR; To, Minh Nguyen Nhat; Mannina, Sophia; Addas, Shamel; Sibley, Stephanie; Pichora, David; Maslove, David; Abolmaesumi, Purang; Mousavi, Parvin
Domain Knowledge is Power: Leveraging Physiological Priors for Self-Supervised Representation Learning in Electrocardiography Journal Article
In: IEEE Transactions on Biomedical Engineering, 2026.
@article{maghsoodi2026a,
title = {Domain Knowledge is Power: Leveraging Physiological Priors for Self-Supervised Representation Learning in Electrocardiography},
author = {Nooshin Maghsoodi and Sarah Nassar and Paul FR Wilson and Minh Nguyen Nhat To and Sophia Mannina and Shamel Addas and Stephanie Sibley and David Pichora and David Maslove and Purang Abolmaesumi and Parvin Mousavi},
year = {2026},
date = {2026-01-01},
journal = {IEEE Transactions on Biomedical Engineering},
publisher = {IEEE},
abstract = {Objective
Electrocardiograms (ECGs) play a crucial role in diagnosing heart conditions; however, the effectiveness of artificial intelligence (AI)-based ECG analysis is often hindered by the limited availability of labeled data. Self-supervised learning (SSL) can address this by leveraging large-scale unlabeled data. We introduce PhysioCLR (Physiology-aware Contrastive Learning Representation for ECG), a physiology-aware contrastive learning framework that incorporates domain-specific priors to enhance the generalizability and clinical relevance of ECG-based arrhythmia classification.
Methods
During pretraining, PhysioCLR learns to bring together embeddings of samples that share similar clinically relevant features while pushing apart those that are dissimilar. Unlike existing methods, our method integrates ECG physiological similarity cues into contrastive learning, promoting the learning of clinically …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Objective
Electrocardiograms (ECGs) play a crucial role in diagnosing heart conditions; however, the effectiveness of artificial intelligence (AI)-based ECG analysis is often hindered by the limited availability of labeled data. Self-supervised learning (SSL) can address this by leveraging large-scale unlabeled data. We introduce PhysioCLR (Physiology-aware Contrastive Learning Representation for ECG), a physiology-aware contrastive learning framework that incorporates domain-specific priors to enhance the generalizability and clinical relevance of ECG-based arrhythmia classification.
Methods
During pretraining, PhysioCLR learns to bring together embeddings of samples that share similar clinically relevant features while pushing apart those that are dissimilar. Unlike existing methods, our method integrates ECG physiological similarity cues into contrastive learning, promoting the learning of clinically …
Electrocardiograms (ECGs) play a crucial role in diagnosing heart conditions; however, the effectiveness of artificial intelligence (AI)-based ECG analysis is often hindered by the limited availability of labeled data. Self-supervised learning (SSL) can address this by leveraging large-scale unlabeled data. We introduce PhysioCLR (Physiology-aware Contrastive Learning Representation for ECG), a physiology-aware contrastive learning framework that incorporates domain-specific priors to enhance the generalizability and clinical relevance of ECG-based arrhythmia classification.
Methods
During pretraining, PhysioCLR learns to bring together embeddings of samples that share similar clinically relevant features while pushing apart those that are dissimilar. Unlike existing methods, our method integrates ECG physiological similarity cues into contrastive learning, promoting the learning of clinically …
Willis, Emma; To, Minh Nguyen Nhat; Elghareb, Tarek; Abolmaesumi, Purang; Mousavi, Parvin
MRI–histopathology alignment for improved ISUP grading in prostate MRI Proceedings Article
In: pp. 65-70, SPIE, 2026.
@inproceedings{willis2026a,
title = {MRI–histopathology alignment for improved ISUP grading in prostate MRI},
author = {Emma Willis and Minh Nguyen Nhat To and Tarek Elghareb and Purang Abolmaesumi and Parvin Mousavi},
year = {2026},
date = {2026-01-01},
volume = {13929},
pages = {65-70},
publisher = {SPIE},
abstract = {Prostate cancer grading relies on histopathology, but this requires invasive biopsy and delays treatment decisions. MRI offers a non-invasive alternative, yet current MRI-based grading methods underperform compared to pathology. We present an unpaired teacher–student framework that aligns an MRI encoder to the embedding space of a histopathology ISUP classifier using triplet loss, without requiring patient-paired MRI–histopathology data. After alignment, we freeze the aligned MRI representation and train only a lightweight MRI classifier head for three-class ISUP bucket prediction (0–1 / 2–3 / 4–5). Using large public MRI and histopathology datasets, this strategy improves test AUROC from 70.3% to 81.9% and substantially improves discrimination of high-grade disease (ISUP 4–5). These results demonstrate the feasibility of unpaired MRI–histopathology alignment for earlier, pathology-informed prostate …},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Prostate cancer grading relies on histopathology, but this requires invasive biopsy and delays treatment decisions. MRI offers a non-invasive alternative, yet current MRI-based grading methods underperform compared to pathology. We present an unpaired teacher–student framework that aligns an MRI encoder to the embedding space of a histopathology ISUP classifier using triplet loss, without requiring patient-paired MRI–histopathology data. After alignment, we freeze the aligned MRI representation and train only a lightweight MRI classifier head for three-class ISUP bucket prediction (0–1 / 2–3 / 4–5). Using large public MRI and histopathology datasets, this strategy improves test AUROC from 70.3% to 81.9% and substantially improves discrimination of high-grade disease (ISUP 4–5). These results demonstrate the feasibility of unpaired MRI–histopathology alignment for earlier, pathology-informed prostate …
Willis, Emma; Elghareb, Tarek; Wilson, Paul FR; To, Minh Nguyen Nhat; Abootorabi, Mohammad Mahdi; Jamzad, Amoon; Wodlinger, Brian; Mousavi, Parvin; Abolmaesumi, Purang
GUIDE-US: grade-informed unpaired distillation of encoder knowledge from histopathology to micro-ultrasound Journal Article
In: arXiv preprint arXiv:2602.19005, 2026.
@article{willis2026,
title = {GUIDE-US: grade-informed unpaired distillation of encoder knowledge from histopathology to micro-ultrasound},
author = {Emma Willis and Tarek Elghareb and Paul FR Wilson and Minh Nguyen Nhat To and Mohammad Mahdi Abootorabi and Amoon Jamzad and Brian Wodlinger and Parvin Mousavi and Purang Abolmaesumi},
year = {2026},
date = {2026-01-01},
journal = {arXiv preprint arXiv:2602.19005},
abstract = {Purpose
Non-invasive grading of prostate cancer (PCa) from micro-ultrasound (micro-US) could expedite triage and guide biopsies toward the most aggressive regions, yet current models struggle to infer tissue micro-structure at coarse imaging resolutions.
Methods
We introduce an unpaired histopathology knowledge-distillation strategy that trains a micro-US encoder to emulate the embedding distribution of a pretrained histopathology foundation model, conditioned on International Society of Urological Pathology (ISUP) grades. Training requires no patient-level pairing or image registration, and histopathology inputs are not used at inference.
Results
Compared to the current state of the art, our approach increases sensitivity to clinically significant PCa (csPCa) at 60% specificity by 3.5% and improves overall sensitivity at 60% specificity by 1.2%.
Conclusion
By enabling earlier and more dependable cancer risk stratification solely from imaging, our method advances clinical feasibility. Source code will be publicly released upon publication.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Purpose
Non-invasive grading of prostate cancer (PCa) from micro-ultrasound (micro-US) could expedite triage and guide biopsies toward the most aggressive regions, yet current models struggle to infer tissue micro-structure at coarse imaging resolutions.
Methods
We introduce an unpaired histopathology knowledge-distillation strategy that trains a micro-US encoder to emulate the embedding distribution of a pretrained histopathology foundation model, conditioned on International Society of Urological Pathology (ISUP) grades. Training requires no patient-level pairing or image registration, and histopathology inputs are not used at inference.
Results
Compared to the current state of the art, our approach increases sensitivity to clinically significant PCa (csPCa) at 60% specificity by 3.5% and improves overall sensitivity at 60% specificity by 1.2%.
Conclusion
By enabling earlier and more dependable cancer risk stratification solely from imaging, our method advances clinical feasibility. Source code will be publicly released upon publication.
Non-invasive grading of prostate cancer (PCa) from micro-ultrasound (micro-US) could expedite triage and guide biopsies toward the most aggressive regions, yet current models struggle to infer tissue micro-structure at coarse imaging resolutions.
Methods
We introduce an unpaired histopathology knowledge-distillation strategy that trains a micro-US encoder to emulate the embedding distribution of a pretrained histopathology foundation model, conditioned on International Society of Urological Pathology (ISUP) grades. Training requires no patient-level pairing or image registration, and histopathology inputs are not used at inference.
Results
Compared to the current state of the art, our approach increases sensitivity to clinically significant PCa (csPCa) at 60% specificity by 3.5% and improves overall sensitivity at 60% specificity by 1.2%.
Conclusion
By enabling earlier and more dependable cancer risk stratification solely from imaging, our method advances clinical feasibility. Source code will be publicly released upon publication.
To, Minh Nguyen Nhat; Kim, Diane; Harmanani, Mohamed; Wilson, Paul FR; Fooladgar, Fahimeh; Sojoudi, Samira; Jamzad, Amoon; Abdalla, Sherif; Tsang, Teresa; Luong, Christina; Chang, Silvia; Black, Peter; Siemens, Robert; Leveridge, Michael; Krishnan, Rahul G; Mousavi, Parvin; Abolmaesumi, Purang
Shift happens: a fairness-oriented framework for medical classification under hidden bias Journal Article
In: International Journal of Computer Assisted Radiology and Surgery, pp. 1-8, 2026.
@article{to2026,
title = {Shift happens: a fairness-oriented framework for medical classification under hidden bias},
author = {Minh Nguyen Nhat To and Diane Kim and Mohamed Harmanani and Paul FR Wilson and Fahimeh Fooladgar and Samira Sojoudi and Amoon Jamzad and Sherif Abdalla and Teresa Tsang and Christina Luong and Silvia Chang and Peter Black and Robert Siemens and Michael Leveridge and Rahul G Krishnan and Parvin Mousavi and Purang Abolmaesumi},
year = {2026},
date = {2026-01-01},
journal = {International Journal of Computer Assisted Radiology and Surgery},
pages = {1-8},
publisher = {Springer International Publishing},
abstract = {Purpose
Many medical AI models perform unevenly across patient groups because they learn shortcuts from biased data. These hidden biases make models less reliable and less fair in real-world use. This work aims to develop a system that remains accurate and fair across different patient subpopulations, even when those groups are not explicitly labeled.
Methods
We introduce DPE-Former, a model that combines prototype-based learning with transformer attention. The system trains several complementary classifiers on balanced subsets of data, each capturing different aspects of the population. A transformer module then learns how to combine its outputs in an adaptive way, helping the model make more balanced decisions across unseen or minority groups.
Results
Across diverse datasets, including prostate ultrasound, skin lesion images, and cardiac patient records, DPE-Former achieved higher accuracy on …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Purpose
Many medical AI models perform unevenly across patient groups because they learn shortcuts from biased data. These hidden biases make models less reliable and less fair in real-world use. This work aims to develop a system that remains accurate and fair across different patient subpopulations, even when those groups are not explicitly labeled.
Methods
We introduce DPE-Former, a model that combines prototype-based learning with transformer attention. The system trains several complementary classifiers on balanced subsets of data, each capturing different aspects of the population. A transformer module then learns how to combine its outputs in an adaptive way, helping the model make more balanced decisions across unseen or minority groups.
Results
Across diverse datasets, including prostate ultrasound, skin lesion images, and cardiac patient records, DPE-Former achieved higher accuracy on …
Many medical AI models perform unevenly across patient groups because they learn shortcuts from biased data. These hidden biases make models less reliable and less fair in real-world use. This work aims to develop a system that remains accurate and fair across different patient subpopulations, even when those groups are not explicitly labeled.
Methods
We introduce DPE-Former, a model that combines prototype-based learning with transformer attention. The system trains several complementary classifiers on balanced subsets of data, each capturing different aspects of the population. A transformer module then learns how to combine its outputs in an adaptive way, helping the model make more balanced decisions across unseen or minority groups.
Results
Across diverse datasets, including prostate ultrasound, skin lesion images, and cardiac patient records, DPE-Former achieved higher accuracy on …
Srikanthan, Dilakshan; Jamzad, Amoon; Wilson, Paul; Maghsoodi, Nooshin; Policelli, Robert; Fichtinger, Gabor; Rudan, John F; Mousavi, Parvin
Do Foundation Models See Biology? Evaluating Attention Coherence with Spatial Transcriptomics in Glioblastoma Journal Article
In: arXiv preprint arXiv:2606.04764, 2026.
@article{srikanthan2026,
title = {Do Foundation Models See Biology? Evaluating Attention Coherence with Spatial Transcriptomics in Glioblastoma},
author = {Dilakshan Srikanthan and Amoon Jamzad and Paul Wilson and Nooshin Maghsoodi and Robert Policelli and Gabor Fichtinger and John F Rudan and Parvin Mousavi},
year = {2026},
date = {2026-01-01},
journal = {arXiv preprint arXiv:2606.04764},
abstract = {Whether attention maps from pathology foundation models capture genuine biology remains unknown, yet this question is critical for clinical trust and regulatory approval. We propose a spatial transcriptomics-based framework for orthogonal, hypothesis-free evaluation of attention and apply it to five pathology foundation models (CONCH v1.5, UNI v2, Virchow2, GigaPath, H-Optimus-1) and a ResNet50 baseline. Using attention-based multiple instance learning, we train single-task and multi-task models to predict five molecular alterations in glioblastoma on the CPTAC cohort, validate on an independent TCGA cohort, and evaluate biological coherence of attention maps against 87 transcriptional signatures using co-registered Visium spatial transcriptomics data from 18 samples. Internally, no single encoder dominates across all tasks, and external validation inverts internal performance rankings. Attention maps show a five-fold enrichment gradient from pathways (Cohen's d=0.329) to individual genes (d=0.055), indicating that attention captures emergent multi-gene transcriptional programs rather than individual molecular events. Spatially smooth attention maps do not imply biological coherence, and different encoders attend to distinct biological compartments. Our framework provides objective, quantitative assessment of what foundation models learn from histopathology, moving the field beyond qualitative saliency map review.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Whether attention maps from pathology foundation models capture genuine biology remains unknown, yet this question is critical for clinical trust and regulatory approval. We propose a spatial transcriptomics-based framework for orthogonal, hypothesis-free evaluation of attention and apply it to five pathology foundation models (CONCH v1.5, UNI v2, Virchow2, GigaPath, H-Optimus-1) and a ResNet50 baseline. Using attention-based multiple instance learning, we train single-task and multi-task models to predict five molecular alterations in glioblastoma on the CPTAC cohort, validate on an independent TCGA cohort, and evaluate biological coherence of attention maps against 87 transcriptional signatures using co-registered Visium spatial transcriptomics data from 18 samples. Internally, no single encoder dominates across all tasks, and external validation inverts internal performance rankings. Attention maps show a five-fold enrichment gradient from pathways (Cohen's d=0.329) to individual genes (d=0.055), indicating that attention captures emergent multi-gene transcriptional programs rather than individual molecular events. Spatially smooth attention maps do not imply biological coherence, and different encoders attend to distinct biological compartments. Our framework provides objective, quantitative assessment of what foundation models learn from histopathology, moving the field beyond qualitative saliency map review.
Radcliffe, Olivia; Tun, Aung Tin; Aung, Nyein Chan; Thwe, Wunn Lei; Connolly, Laura; Ungi, Tamas; Thornton, Kanchana; Davison, Colleen; Purkey, Eva; Mousavi, Parvin; Fichtinger, Gabor
Advancing Prosthetic Care Access on the Thailand–Burma Border Through Open-Source Technology Journal Article
In: IEEE pulse, vol. 16, no. 6, pp. 64-70, 2026.
@article{radcliffe2026,
title = {Advancing Prosthetic Care Access on the Thailand–Burma Border Through Open-Source Technology},
author = {Olivia Radcliffe and Aung Tin Tun and Nyein Chan Aung and Wunn Lei Thwe and Laura Connolly and Tamas Ungi and Kanchana Thornton and Colleen Davison and Eva Purkey and Parvin Mousavi and Gabor Fichtinger},
year = {2026},
date = {2026-01-01},
journal = {IEEE pulse},
volume = {16},
number = {6},
pages = {64-70},
publisher = {IEEE},
abstract = {The ongoing civil war in Myanmar (Burma) has severely disrupted the country’s health care system, forcing widespread displacement and creating a critical shortage of medical care and prosthetic devices for refugees and migrants along the Thai border. Prohibitive costs, resource shortages, and movement constraints severely limit access to essential health care and functional prosthetics in this region. This article details a collaboration between Queen’s University and the Burma Children Medical Fund (BCMF), a nongovernmental organization (NGO) that established a 3-D-printed prosthetic program in 2019 using open-source technology. Since 2023, this collaboration has facilitated a three-year student mobility program, sending researchers to Thailand to enhance BCMF’s existing initiative by addressing technical limitations and improving design autonomy. Using the development of a short transradial (below …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The ongoing civil war in Myanmar (Burma) has severely disrupted the country’s health care system, forcing widespread displacement and creating a critical shortage of medical care and prosthetic devices for refugees and migrants along the Thai border. Prohibitive costs, resource shortages, and movement constraints severely limit access to essential health care and functional prosthetics in this region. This article details a collaboration between Queen’s University and the Burma Children Medical Fund (BCMF), a nongovernmental organization (NGO) that established a 3-D-printed prosthetic program in 2019 using open-source technology. Since 2023, this collaboration has facilitated a three-year student mobility program, sending researchers to Thailand to enhance BCMF’s existing initiative by addressing technical limitations and improving design autonomy. Using the development of a short transradial (below …
Nassar, Sarah; Maghsoodi, Nooshin; Mannina, Sophia; Addas, Shamel; Sibley, Stephanie; Fichtinger, Gabor; Pichora, David; Maslove, David; Abolmaesumi, Purang; Mousavi, Parvin
A Dataset and Benchmarks for Atrial Fibrillation Detection from Electrocardiograms of Intensive Care Unit Patients Journal Article
In: IEEE Transactions on Biomedical Engineering, 2026.
@article{nassar2026,
title = {A Dataset and Benchmarks for Atrial Fibrillation Detection from Electrocardiograms of Intensive Care Unit Patients},
author = {Sarah Nassar and Nooshin Maghsoodi and Sophia Mannina and Shamel Addas and Stephanie Sibley and Gabor Fichtinger and David Pichora and David Maslove and Purang Abolmaesumi and Parvin Mousavi},
year = {2026},
date = {2026-01-01},
journal = {IEEE Transactions on Biomedical Engineering},
publisher = {IEEE},
abstract = {Objective
Atrial fibrillation (AF) is the most common cardiac arrhythmia experienced by intensive care unit (ICU) patients and can cause adverse health effects. In this study, we publish a labelled ICU dataset and bench marks for AF detection.
Methods
We compared machine learning models across three data-driven artificial intelligence (AI) approaches: feature-based classifiers, deep learning (DL), and ECG foundation models (FMs). This comparison addresses a critical gap in the literature and aims to pinpoint which AI approach is best for high-performing AF detection. Electrocardiograms (ECGs) from a Canadian ICU and the 2021 PhysioNet/Computing in Cardiology Challenge were used to conduct the experiments. Multiple training configurations were tested, ranging from zero-shot in ference to transfer learning.
Results
Across both datasets, ECG FMs generally performed best, followed by DL, then feature …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Objective
Atrial fibrillation (AF) is the most common cardiac arrhythmia experienced by intensive care unit (ICU) patients and can cause adverse health effects. In this study, we publish a labelled ICU dataset and bench marks for AF detection.
Methods
We compared machine learning models across three data-driven artificial intelligence (AI) approaches: feature-based classifiers, deep learning (DL), and ECG foundation models (FMs). This comparison addresses a critical gap in the literature and aims to pinpoint which AI approach is best for high-performing AF detection. Electrocardiograms (ECGs) from a Canadian ICU and the 2021 PhysioNet/Computing in Cardiology Challenge were used to conduct the experiments. Multiple training configurations were tested, ranging from zero-shot in ference to transfer learning.
Results
Across both datasets, ECG FMs generally performed best, followed by DL, then feature …
Atrial fibrillation (AF) is the most common cardiac arrhythmia experienced by intensive care unit (ICU) patients and can cause adverse health effects. In this study, we publish a labelled ICU dataset and bench marks for AF detection.
Methods
We compared machine learning models across three data-driven artificial intelligence (AI) approaches: feature-based classifiers, deep learning (DL), and ECG foundation models (FMs). This comparison addresses a critical gap in the literature and aims to pinpoint which AI approach is best for high-performing AF detection. Electrocardiograms (ECGs) from a Canadian ICU and the 2021 PhysioNet/Computing in Cardiology Challenge were used to conduct the experiments. Multiple training configurations were tested, ranging from zero-shot in ference to transfer learning.
Results
Across both datasets, ECG FMs generally performed best, followed by DL, then feature …
Maghsoodi, Nooshin; Jamzad, Amoon; Policelli, Robert; Farahmand, Mohammad; Srikanthan, Dilakshan; Kaufmann, Martin; Ren, Kevin YM; Merchant, Shaila; Varma, Sonal; Walker, Ross; McKay, Doug; Rudan, John; Fichtinger, Gabor; Mousavi, Parvin
Agent-Guided Relational Concept Discovery: Toward Interpretable Surgical Margin Assessment Journal Article
In: arXiv preprint arXiv:2607.21437, 2026.
@article{maghsoodi2026,
title = {Agent-Guided Relational Concept Discovery: Toward Interpretable Surgical Margin Assessment},
author = {Nooshin Maghsoodi and Amoon Jamzad and Robert Policelli and Mohammad Farahmand and Dilakshan Srikanthan and Martin Kaufmann and Kevin YM Ren and Shaila Merchant and Sonal Varma and Ross Walker and Doug McKay and John Rudan and Gabor Fichtinger and Parvin Mousavi},
year = {2026},
date = {2026-01-01},
journal = {arXiv preprint arXiv:2607.21437},
abstract = {Deep learning models can effectively use Rapid Evaporative Ionization Mass Spectrometry (REIMS) data for surgical margin assessment. However, their clinical adoption remains challenging due to limited generalization to operating room conditions. This difficulty arises because models are typically trained on labeled spectra collected from resected tissue samples, while they must operate on noisy, unlabeled data acquired directly during surgery. In addition, the black-box nature of deep learning models makes it difficult to understand and systematically improve their behavior. Concept-based learning offers a promising way to address these challenges by mapping raw measurements to human-understandable concepts. However, supervised concept-based approaches rely on concept annotations, which are difficult to obtain in complex mass spectrometry workflows. We propose Agent-Guided Concept Discovery, a framework that learns meaningful concepts directly from data without requiring predefined concept labels. During training, a reasoning agent refines semantic descriptions of the learned concepts and adaptively adjusts their weight based on diagnostic relevance. These concepts are further grounded using a biochemical knowledge graph to ensure consistency with known metabolic relationships. Across Skin and Breast Cancer datasets, our model improves balanced accuracy and sensitivity over the baseline. In a representative intraoperative case, it shows fewer false positives, indicating better generalization to surgical conditions.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Deep learning models can effectively use Rapid Evaporative Ionization Mass Spectrometry (REIMS) data for surgical margin assessment. However, their clinical adoption remains challenging due to limited generalization to operating room conditions. This difficulty arises because models are typically trained on labeled spectra collected from resected tissue samples, while they must operate on noisy, unlabeled data acquired directly during surgery. In addition, the black-box nature of deep learning models makes it difficult to understand and systematically improve their behavior. Concept-based learning offers a promising way to address these challenges by mapping raw measurements to human-understandable concepts. However, supervised concept-based approaches rely on concept annotations, which are difficult to obtain in complex mass spectrometry workflows. We propose Agent-Guided Concept Discovery, a framework that learns meaningful concepts directly from data without requiring predefined concept labels. During training, a reasoning agent refines semantic descriptions of the learned concepts and adaptively adjusts their weight based on diagnostic relevance. These concepts are further grounded using a biochemical knowledge graph to ensure consistency with known metabolic relationships. Across Skin and Breast Cancer datasets, our model improves balanced accuracy and sensitivity over the baseline. In a representative intraoperative case, it shows fewer false positives, indicating better generalization to surgical conditions.
Logan, Stuart; Connolly, Laura; Barr, Colton; Mousavi, Parvin; Fichtinger, Gabor; Hashtrudi-Zaad, Keyvan
Evaluating vibrotactile feedback for electromagnetic surgical navigation Proceedings Article
In: pp. 316-323, SPIE, 2026.
@inproceedings{logan2026,
title = {Evaluating vibrotactile feedback for electromagnetic surgical navigation},
author = {Stuart Logan and Laura Connolly and Colton Barr and Parvin Mousavi and Gabor Fichtinger and Keyvan Hashtrudi-Zaad},
year = {2026},
date = {2026-01-01},
volume = {13927},
pages = {316-323},
publisher = {SPIE},
abstract = {INTRODUCTION
Breast-conserving surgery (BCS) requires precise excision of tumours while preserving healthy tissue, yet positive margin rates remain high due to challenges in intraoperative tumour localization. We investigate a guidance system that integrates tracked ultrasound, electromagnetic navigation, and a vibrotactile tactor to provide haptic feedback to the surgeon.
METHODS
Our proposed system continuously measures the distance between the cautery tool and the tumour boundary, triggering vibrotactile signals as the instrument approaches the margin. In this paper, we evaluate system performance through electromagnetic interference characterization to determine the optimal tactor placement; latency testing to quantify end-to-end response time; and root-mean-square displacement analysis to verify that tactor vibration does not perturb the surgical instrument.
RESULTS
We demonstrate that the …},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
INTRODUCTION
Breast-conserving surgery (BCS) requires precise excision of tumours while preserving healthy tissue, yet positive margin rates remain high due to challenges in intraoperative tumour localization. We investigate a guidance system that integrates tracked ultrasound, electromagnetic navigation, and a vibrotactile tactor to provide haptic feedback to the surgeon.
METHODS
Our proposed system continuously measures the distance between the cautery tool and the tumour boundary, triggering vibrotactile signals as the instrument approaches the margin. In this paper, we evaluate system performance through electromagnetic interference characterization to determine the optimal tactor placement; latency testing to quantify end-to-end response time; and root-mean-square displacement analysis to verify that tactor vibration does not perturb the surgical instrument.
RESULTS
We demonstrate that the …
Breast-conserving surgery (BCS) requires precise excision of tumours while preserving healthy tissue, yet positive margin rates remain high due to challenges in intraoperative tumour localization. We investigate a guidance system that integrates tracked ultrasound, electromagnetic navigation, and a vibrotactile tactor to provide haptic feedback to the surgeon.
METHODS
Our proposed system continuously measures the distance between the cautery tool and the tumour boundary, triggering vibrotactile signals as the instrument approaches the margin. In this paper, we evaluate system performance through electromagnetic interference characterization to determine the optimal tactor placement; latency testing to quantify end-to-end response time; and root-mean-square displacement analysis to verify that tactor vibration does not perturb the surgical instrument.
RESULTS
We demonstrate that the …
Liu, Sophia; Jamzad, Amoon; Farahmand, Mohammad; Mousavi, Parvin
Integrating Foundation Models into MassVision for Mass Spectrometry Imaging Journal Article
In: Inquiry@ Queen's Undergraduate Research Conference Proceedings, vol. 20, no. 2, 2026.
@article{liu2026,
title = {Integrating Foundation Models into MassVision for Mass Spectrometry Imaging},
author = {Sophia Liu and Amoon Jamzad and Mohammad Farahmand and Parvin Mousavi},
year = {2026},
date = {2026-01-01},
journal = {Inquiry@ Queen's Undergraduate Research Conference Proceedings},
volume = {20},
number = {2},
abstract = {Mass spectrometry imaging (MSI) combines mass spectrometry with spatial information to map the distribution of molecules across tissue samples. This provides insight into biochemical changes associated with diseases such as cancer. MassVision, a 3D Slicer module developed in the Med-i Lab, offers tools for the visualization and analysis of MSI datasets. The high dimensionality and complexity of MSI data, however, can make interpretation challenging. Foundation models offer a potential way to address this challenge by learning from large-scale datasets and converting complex data into simplified representations to identify meaningful patterns. This project investigates the introduction of foundation models into the MassVision workflow.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Mass spectrometry imaging (MSI) combines mass spectrometry with spatial information to map the distribution of molecules across tissue samples. This provides insight into biochemical changes associated with diseases such as cancer. MassVision, a 3D Slicer module developed in the Med-i Lab, offers tools for the visualization and analysis of MSI datasets. The high dimensionality and complexity of MSI data, however, can make interpretation challenging. Foundation models offer a potential way to address this challenge by learning from large-scale datasets and converting complex data into simplified representations to identify meaningful patterns. This project investigates the introduction of foundation models into the MassVision workflow.
Jamzad, Amoon; Srikanthan, Dilakshan; Akbarifar, Faranak; Maghsoodi, Nooshin; Mousavi, Parvin
P3CA: Encoder-Agnostic Interpretation of Vision Foundation Model Embeddings via Spatial Probing Journal Article
In: arXiv preprint arXiv:2608.10131, 2026.
@article{jamzad2026,
title = {P3CA: Encoder-Agnostic Interpretation of Vision Foundation Model Embeddings via Spatial Probing},
author = {Amoon Jamzad and Dilakshan Srikanthan and Faranak Akbarifar and Nooshin Maghsoodi and Parvin Mousavi},
year = {2026},
date = {2026-01-01},
journal = {arXiv preprint arXiv:2608.10131},
abstract = {Vision foundation models are increasingly used as reusable encoders in medical image computing, yet their high-dimensional spatial embeddings are difficult to inspect beyond downstream task performance or global dimensionality reduction. We propose position-prompted PCA (P3CA), an encoder-agnostic method for local probing of channel-rich spatial tensors. Given a user-selected spatial prompt, P3CA estimates the feature normalization and dominant covariance directions within that region, then applies the resulting projection to the full tensor to visualize where locally informative directions are expressed. This produces a region-conditioned representation lens without modifying the encoder, retraining, or requiring task-specific labels. We implement P3CA in EmbedVision, an interactive 3D Slicer-based workflow, and evaluate it across natural images, colorectal pathology foundation-model embeddings, and spatial transcriptomic tensors. Across these settings, prompted projections reveal local structure suppressed by global PCA, improve prompt-matched pathology discrimination from frozen three-dimensional projections, and support comparison between learned and measured spatial representations.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Vision foundation models are increasingly used as reusable encoders in medical image computing, yet their high-dimensional spatial embeddings are difficult to inspect beyond downstream task performance or global dimensionality reduction. We propose position-prompted PCA (P3CA), an encoder-agnostic method for local probing of channel-rich spatial tensors. Given a user-selected spatial prompt, P3CA estimates the feature normalization and dominant covariance directions within that region, then applies the resulting projection to the full tensor to visualize where locally informative directions are expressed. This produces a region-conditioned representation lens without modifying the encoder, retraining, or requiring task-specific labels. We implement P3CA in EmbedVision, an interactive 3D Slicer-based workflow, and evaluate it across natural images, colorectal pathology foundation-model embeddings, and spatial transcriptomic tensors. Across these settings, prompted projections reveal local structure suppressed by global PCA, improve prompt-matched pathology discrimination from frozen three-dimensional projections, and support comparison between learned and measured spatial representations.
Iaboni, Natasha; Kooster, Teaghan; Kaufmann, Martin; Carew, Madeleine; Jamzad, Amoon; Abdulhameed, Abdulhameed; Rubino, Rachel; Ren, Kevin; Rudan, John F; Mousavi, Parvin; Varma, Sonal; Nicol, Christopher JB
Investigating the metabolic differences between in situ and invasive ductal carcinoma using spatial metabolomics Proceedings Article
In: CANADIAN SCIENCE PUBLISHING, 2026.
@inproceedings{iaboni2026,
title = {Investigating the metabolic differences between in situ and invasive ductal carcinoma using spatial metabolomics},
author = {Natasha Iaboni and Teaghan Kooster and Martin Kaufmann and Madeleine Carew and Amoon Jamzad and Abdulhameed Abdulhameed and Rachel Rubino and Kevin Ren and John F Rudan and Parvin Mousavi and Sonal Varma and Christopher JB Nicol},
year = {2026},
date = {2026-01-01},
volume = {104},
publisher = {CANADIAN SCIENCE PUBLISHING},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Harmanani, Mohamed; Long, Bining; Guo, Zhuoxin; Wilson, Paul FR; Sabour, Amirhossein; To, Minh Nguyen Nhat; Fichtinger, Gabor; Abolmaesumi, Purang; Mousavi, Parvin
Vision-Language Models Encode Clinical Guidelines for Concept-Based Medical Reasoning Journal Article
In: arXiv preprint arXiv:2603.08921, 2026.
@article{harmanani2026,
title = {Vision-Language Models Encode Clinical Guidelines for Concept-Based Medical Reasoning},
author = {Mohamed Harmanani and Bining Long and Zhuoxin Guo and Paul FR Wilson and Amirhossein Sabour and Minh Nguyen Nhat To and Gabor Fichtinger and Purang Abolmaesumi and Parvin Mousavi},
year = {2026},
date = {2026-01-01},
journal = {arXiv preprint arXiv:2603.08921},
abstract = {Concept Bottleneck Models (CBMs) are a prominent framework for interpretable AI that map learned visual features to a set of meaningful concepts for task-specific downstream predictions. Their sequential structure enhances transparency by connecting model predictions to the underlying concepts that support them. In medical imaging, where transparency is essential, CBMs offer an appealing foundation for explainable model design. However, discrete concept representations often overlook broader clinical context such as diagnostic guidelines and expert heuristics, reducing reliability in complex cases. We propose MedCBR, a concept-based reasoning framework that integrates clinical guidelines with vision-language and reasoning models. Labeled clinical descriptors are transformed into guideline-conformant text, and a concept-based model is trained with a multitask objective combining multimodal contrastive alignment, concept supervision, and diagnostic classification to jointly ground image features, concepts, and pathology. A reasoning model then converts these predictions into structured clinical narratives that explain the diagnosis, emulating expert reasoning based on established guidelines. MedCBR achieves superior diagnostic and concept-level performance, with AUROCs of 94.2% on ultrasound and 84.0% on mammography. Further experiments on non-medical datasets achieve 86.1% accuracy. Our framework enhances interpretability and forms an end-to-end bridge from medical image analysis to decision-making.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Concept Bottleneck Models (CBMs) are a prominent framework for interpretable AI that map learned visual features to a set of meaningful concepts for task-specific downstream predictions. Their sequential structure enhances transparency by connecting model predictions to the underlying concepts that support them. In medical imaging, where transparency is essential, CBMs offer an appealing foundation for explainable model design. However, discrete concept representations often overlook broader clinical context such as diagnostic guidelines and expert heuristics, reducing reliability in complex cases. We propose MedCBR, a concept-based reasoning framework that integrates clinical guidelines with vision-language and reasoning models. Labeled clinical descriptors are transformed into guideline-conformant text, and a concept-based model is trained with a multitask objective combining multimodal contrastive alignment, concept supervision, and diagnostic classification to jointly ground image features, concepts, and pathology. A reasoning model then converts these predictions into structured clinical narratives that explain the diagnosis, emulating expert reasoning based on established guidelines. MedCBR achieves superior diagnostic and concept-level performance, with AUROCs of 94.2% on ultrasound and 84.0% on mammography. Further experiments on non-medical datasets achieve 86.1% accuracy. Our framework enhances interpretability and forms an end-to-end bridge from medical image analysis to decision-making.
Guo, Zhuoxin; Harmanani, Mohamed; Wilson, Paul FR; To, Minh Nguyen Nhat; Elghareb, Tarek; Dzikunu, Obed Korshie; Maghsoodi, Nooshin; Fichtinger, Gabor; Abolmaesumi, Purang; Mousavi, Parvin
Pathology-guided contrastive pretraining enriches preoperative CT representations for prognosis Proceedings Article
In: pp. 258-266, SPIE, 2026.
@inproceedings{guo2026,
title = {Pathology-guided contrastive pretraining enriches preoperative CT representations for prognosis},
author = {Zhuoxin Guo and Mohamed Harmanani and Paul FR Wilson and Minh Nguyen Nhat To and Tarek Elghareb and Obed Korshie Dzikunu and Nooshin Maghsoodi and Gabor Fichtinger and Purang Abolmaesumi and Parvin Mousavi},
year = {2026},
date = {2026-01-01},
volume = {13927},
pages = {258-266},
publisher = {SPIE},
abstract = {PURPOSE
Prognosis prediction is important for personalized cancer treatment. Histopathology imaging can provide key prognostic information but it relies on tissue obtained during surgery, which restricts its use for preoperative decision-making. Preoperative imaging modalities such as computed tomography (CT) are more accessible but often lack the pathological context needed for accurate prognosis. Multi-modal approaches that combine imaging and pathology aim to bridge this gap, yet they typically require both modalities at inference time, making them impractical for preoperative decision-making.
METHODS
To address this, we adopt a multi-modal contrastive learning (MMCL) framework that uses histopathology embeddings to guide CT feature learning. This approach enables the CT encoder to capture pathology-informed patterns during pretraining. At inference, MMCL uses only CT and routine clinical …},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
PURPOSE
Prognosis prediction is important for personalized cancer treatment. Histopathology imaging can provide key prognostic information but it relies on tissue obtained during surgery, which restricts its use for preoperative decision-making. Preoperative imaging modalities such as computed tomography (CT) are more accessible but often lack the pathological context needed for accurate prognosis. Multi-modal approaches that combine imaging and pathology aim to bridge this gap, yet they typically require both modalities at inference time, making them impractical for preoperative decision-making.
METHODS
To address this, we adopt a multi-modal contrastive learning (MMCL) framework that uses histopathology embeddings to guide CT feature learning. This approach enables the CT encoder to capture pathology-informed patterns during pretraining. At inference, MMCL uses only CT and routine clinical …
Prognosis prediction is important for personalized cancer treatment. Histopathology imaging can provide key prognostic information but it relies on tissue obtained during surgery, which restricts its use for preoperative decision-making. Preoperative imaging modalities such as computed tomography (CT) are more accessible but often lack the pathological context needed for accurate prognosis. Multi-modal approaches that combine imaging and pathology aim to bridge this gap, yet they typically require both modalities at inference time, making them impractical for preoperative decision-making.
METHODS
To address this, we adopt a multi-modal contrastive learning (MMCL) framework that uses histopathology embeddings to guide CT feature learning. This approach enables the CT encoder to capture pathology-informed patterns during pretraining. At inference, MMCL uses only CT and routine clinical …
Dzikunu, Obed Korshie; Abootorabi, Mohammad Mahdi; Harmanani, Mohamed; Wilson, Paul FR; Willis, Emma; Luger, Ferdinand; Kinnaird, Adam; Wodlinger, Brian; Mousavi, Parvin; Abolmaesumi, Purang
Learning Prostate Anatomy at Test Time for Cancer Detection in Micro-Ultrasound Journal Article
In: arXiv preprint arXiv:2608.20557, 2026.
@article{dzikunu2026,
title = {Learning Prostate Anatomy at Test Time for Cancer Detection in Micro-Ultrasound},
author = {Obed Korshie Dzikunu and Mohammad Mahdi Abootorabi and Mohamed Harmanani and Paul FR Wilson and Emma Willis and Ferdinand Luger and Adam Kinnaird and Brian Wodlinger and Parvin Mousavi and Purang Abolmaesumi},
year = {2026},
date = {2026-01-01},
journal = {arXiv preprint arXiv:2608.20557},
abstract = {Domain shift across clinical centers using different imaging hardware or acquisition protocols remains a fundamental barrier to deploying deep learning models for prostate cancer (PCa) detection. Existing test-time adaptation (TTA) methods address distribution shift through entropy minimization or augmentation-based self-supervision, correcting for statistical differences in image appearance but ignoring the anatomical structure of the target domain. We propose ANT, a segmentation-guided TTA framework that adapts a pretrained cancer detection encoder to the target domain by solving an auxiliary prostate segmentation task at test time, supervised by pseudo-masks from a frozen pretrained segmentation network. By aligning encoder representations to prostate anatomy in the target domain, ANT corrects domain-specific feature drift while preserving cancer-discriminative structure. The model was trained on 693 patients imaged with an earlier-generation micro-ultrasound scanner in a multi-center clinical trial, and evaluated on 118 patients acquired with a newer-generation system across two centers in another clinical trial. Under a leave-one-center-out protocol with identical evaluation conditions across all methods, ANT improves mean AUC by 2.9% and 3.6% at the biopsy-core and patient levels, respectively, over no adaptation, outperforming TTA baselines. Code is available at: https://github.com/ObedDzik/ant.git.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Domain shift across clinical centers using different imaging hardware or acquisition protocols remains a fundamental barrier to deploying deep learning models for prostate cancer (PCa) detection. Existing test-time adaptation (TTA) methods address distribution shift through entropy minimization or augmentation-based self-supervision, correcting for statistical differences in image appearance but ignoring the anatomical structure of the target domain. We propose ANT, a segmentation-guided TTA framework that adapts a pretrained cancer detection encoder to the target domain by solving an auxiliary prostate segmentation task at test time, supervised by pseudo-masks from a frozen pretrained segmentation network. By aligning encoder representations to prostate anatomy in the target domain, ANT corrects domain-specific feature drift while preserving cancer-discriminative structure. The model was trained on 693 patients imaged with an earlier-generation micro-ultrasound scanner in a multi-center clinical trial, and evaluated on 118 patients acquired with a newer-generation system across two centers in another clinical trial. Under a leave-one-center-out protocol with identical evaluation conditions across all methods, ANT improves mean AUC by 2.9% and 3.6% at the biopsy-core and patient levels, respectively, over no adaptation, outperforming TTA baselines. Code is available at: https://github.com/ObedDzik/ant.git.
Connolly, Laura; Song, Hyunwoo; Xu, Keshuai; Deguet, Anton; Leonard, Simon; Fichtinger, Gabor; Mousavi, Parvin; Taylor, Russell H; Boctor, Emad
No cancer left behind: a testbed and demonstration of concept for photoacoustic tumor bed inspection Journal Article
In: Computer Assisted Surgery, vol. 31, no. 1, pp. 2604123, 2026.
@article{connolly2026,
title = {No cancer left behind: a testbed and demonstration of concept for photoacoustic tumor bed inspection},
author = {Laura Connolly and Hyunwoo Song and Keshuai Xu and Anton Deguet and Simon Leonard and Gabor Fichtinger and Parvin Mousavi and Russell H Taylor and Emad Boctor},
year = {2026},
date = {2026-01-01},
journal = {Computer Assisted Surgery},
volume = {31},
number = {1},
pages = {2604123},
publisher = {Taylor & Francis},
abstract = {Cancer resection surgery is unsuccessful if tumor tissue is left behind in the surgical cavity. Identifying the residual cancer requires additional imaging or postoperative histological analysis. Photoacoustic imaging can be used to image both the surface and depths of the resection cavity; however, its performance hinges on consistent probe placement and stable acoustic and optical coupling. As intra-cavity deployment of photoacoustic imaging is largely uncharted, several potential embodiments warrant rigorous investigation. We address this need with an open-source robotic testbed for intraoperative tumor-bed inspection using photoacoustic imaging. The platform integrates the da Vinci Research Kit, depth imaging, and electromagnetic tracking to automate cavity scanning and maintain repeatable probe trajectories. Using tissue-mimicking phantoms, we (i) demonstrate a novel imaging embodiment for …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Cancer resection surgery is unsuccessful if tumor tissue is left behind in the surgical cavity. Identifying the residual cancer requires additional imaging or postoperative histological analysis. Photoacoustic imaging can be used to image both the surface and depths of the resection cavity; however, its performance hinges on consistent probe placement and stable acoustic and optical coupling. As intra-cavity deployment of photoacoustic imaging is largely uncharted, several potential embodiments warrant rigorous investigation. We address this need with an open-source robotic testbed for intraoperative tumor-bed inspection using photoacoustic imaging. The platform integrates the da Vinci Research Kit, depth imaging, and electromagnetic tracking to automate cavity scanning and maintain repeatable probe trajectories. Using tissue-mimicking phantoms, we (i) demonstrate a novel imaging embodiment for …
Barr, Colton; Galvin, Colin; Azimi, Amirali; Frisken, Sarah; Pieper, Steve; Fichtinger, Gabor; Golby, Alexandra; Mousavi, Parvin
Local LLMs as cooperative agents for low-cost surgical navigation support Journal Article
In: International Journal of Computer Assisted Radiology and Surgery, pp. 1-8, 2026.
@article{barr2026,
title = {Local LLMs as cooperative agents for low-cost surgical navigation support},
author = {Colton Barr and Colin Galvin and Amirali Azimi and Sarah Frisken and Steve Pieper and Gabor Fichtinger and Alexandra Golby and Parvin Mousavi},
year = {2026},
date = {2026-01-01},
journal = {International Journal of Computer Assisted Radiology and Surgery},
pages = {1-8},
publisher = {Springer International Publishing},
abstract = {Purpose
Computer-integrated surgical navigation systems are often run in the OR with the assistance of a technician for controlling the user interface and advising on technical details of the system. In lower-resource healthcare settings, limited access to additional OR staff and technical training for operating navigation systems can represent a barrier to sustainably deploying a low-cost surgical navigation solution. Recent advancements in locally deployable large language models (LLMs) have improved their ability to answer technical questions based on source materials and safely perform limited tasks on behalf of a user.
Methods
The objective of this paper is to explore the feasibility of using a network of local LLM-based agents to act as a natural language interface for a low-cost surgical navigation system, facilitating hands-free manipulation of the user interface and providing documentation-grounded technical …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Purpose
Computer-integrated surgical navigation systems are often run in the OR with the assistance of a technician for controlling the user interface and advising on technical details of the system. In lower-resource healthcare settings, limited access to additional OR staff and technical training for operating navigation systems can represent a barrier to sustainably deploying a low-cost surgical navigation solution. Recent advancements in locally deployable large language models (LLMs) have improved their ability to answer technical questions based on source materials and safely perform limited tasks on behalf of a user.
Methods
The objective of this paper is to explore the feasibility of using a network of local LLM-based agents to act as a natural language interface for a low-cost surgical navigation system, facilitating hands-free manipulation of the user interface and providing documentation-grounded technical …
Computer-integrated surgical navigation systems are often run in the OR with the assistance of a technician for controlling the user interface and advising on technical details of the system. In lower-resource healthcare settings, limited access to additional OR staff and technical training for operating navigation systems can represent a barrier to sustainably deploying a low-cost surgical navigation solution. Recent advancements in locally deployable large language models (LLMs) have improved their ability to answer technical questions based on source materials and safely perform limited tasks on behalf of a user.
Methods
The objective of this paper is to explore the feasibility of using a network of local LLM-based agents to act as a natural language interface for a low-cost surgical navigation system, facilitating hands-free manipulation of the user interface and providing documentation-grounded technical …
2025
Farahmand, Mohammad; Jamzad, Amoon; Fooladgar, Fahimeh; Connolly, Laura; Kaufmann, Martin; Ren, Kevin Yi Mi; Rudan, John; McKay, Doug; Fichtinger, Gabor; Mousavi, Parvin
FACT: Foundation Model for Assessing Cancer Tissue Margins with Mass Spectrometry Best Paper Journal Article
In: International Journal of Computer Assisted Radiology and Surgery, vol. 20, no. 6, pp. 1097-1104, 2025, ISSN: 1861-6429.
@article{Farahmand2025,
title = {FACT: Foundation Model for Assessing Cancer Tissue Margins with Mass Spectrometry},
author = {Mohammad Farahmand and Amoon Jamzad and Fahimeh Fooladgar and Laura Connolly and Martin Kaufmann and Kevin Yi Mi Ren and John Rudan and Doug McKay and Gabor Fichtinger and Parvin Mousavi},
url = {https://arxiv.org/pdf/2504.11519},
doi = {10.1007/s11548-025-03355-8},
issn = {1861-6429},
year = {2025},
date = {2025-06-01},
urldate = {2025-06-01},
journal = {International Journal of Computer Assisted Radiology and Surgery},
volume = {20},
number = {6},
pages = {1097-1104},
abstract = {Accurately classifying tissue margins during cancer surgeries is crucial for ensuring complete tumor removal. Rapid Evaporative Ionization Mass Spectrometry (REIMS), a tool for real-time intraoperative margin assessment, generates spectra that require machine learning models to support clinical decision-making. However, the scarcity of labeled data in surgical contexts presents a significant challenge. This study is the first to develop a foundation model tailored specifically for REIMS data, addressing this limitation and advancing real-time surgical margin assessment.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Accurately classifying tissue margins during cancer surgeries is crucial for ensuring complete tumor removal. Rapid Evaporative Ionization Mass Spectrometry (REIMS), a tool for real-time intraoperative margin assessment, generates spectra that require machine learning models to support clinical decision-making. However, the scarcity of labeled data in surgical contexts presents a significant challenge. This study is the first to develop a foundation model tailored specifically for REIMS data, addressing this limitation and advancing real-time surgical margin assessment.
Akbarifar, Faranak; Dukelow, Sean P; Jin, Albert; Mousavi, Parvin; Scott, Stephen H
Optimizing stroke detection using evidential networks and uncertainty-based refinement Journal Article
In: IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 33, pp. 566-576, 2025.
@article{akbarifar2025,
title = {Optimizing stroke detection using evidential networks and uncertainty-based refinement},
author = {Faranak Akbarifar and Sean P Dukelow and Albert Jin and Parvin Mousavi and Stephen H Scott},
year = {2025},
date = {2025-01-01},
journal = {IEEE Transactions on Neural Systems and Rehabilitation Engineering},
volume = {33},
pages = {566-576},
publisher = {IEEE},
abstract = {Evaluating neurological impairments post-stroke is essential for assessing treatment efficacy and managing subsequent disabilities. Conventional clinical assessment methods depend largely on clinicians’ visual and physical evaluations, resulting in coarse rating systems that frequently miss subtle impairments or improvements. Interactive robotic devices, like the Kinarm Exoskeleton system, are transforming the assessment of motor impairments by offering precise and objective movement measurements. In this study, we analyzed kinematic data from 337 stroke patients and 368 healthy controls performing three Kinarm tasks. Using deep learning methods, particularly an evidential network, we distinguished impaired participants from healthy controls while generating measures of prediction uncertainty. By retraining the network with the least uncertain samples and refining the test set by excluding the top 10 …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Evaluating neurological impairments post-stroke is essential for assessing treatment efficacy and managing subsequent disabilities. Conventional clinical assessment methods depend largely on clinicians’ visual and physical evaluations, resulting in coarse rating systems that frequently miss subtle impairments or improvements. Interactive robotic devices, like the Kinarm Exoskeleton system, are transforming the assessment of motor impairments by offering precise and objective movement measurements. In this study, we analyzed kinematic data from 337 stroke patients and 368 healthy controls performing three Kinarm tasks. Using deep learning methods, particularly an evidential network, we distinguished impaired participants from healthy controls while generating measures of prediction uncertainty. By retraining the network with the least uncertain samples and refining the test set by excluding the top 10 …
Wilson, Paul FR; Ronchetti, Matteo; Göbl, Rüdiger; Markova, Viktoria; Rosenzweig, Sebastian; Prevost, Raphael; Mousavi, Parvin; Zettinig, Oliver
Dualtrack: Sensorless 3d ultrasound needs local and global context Journal Article
In: pp. 3-12, 2025.
@article{wilson2025,
title = {Dualtrack: Sensorless 3d ultrasound needs local and global context},
author = {Paul FR Wilson and Matteo Ronchetti and Rüdiger Göbl and Viktoria Markova and Sebastian Rosenzweig and Raphael Prevost and Parvin Mousavi and Oliver Zettinig},
year = {2025},
date = {2025-01-01},
pages = {3-12},
publisher = {Springer Nature Switzerland},
abstract = {Three-dimensional ultrasound (US) offers many clinical advantages over conventional 2D imaging, yet its widespread adoption is limited by the cost and complexity of traditional 3D systems. Sensorless 3D US, using deep learning to estimate a 3D probe trajectory from a sequence of 2D US images, is a promising alternative.Local features such as speckle patterns can help predict frame-to-frame motions, while global features, such as coarse shapes and anatomical structures, can situate the scan relative to anatomy and help predict its general shape. In prior approaches, global features are either ignored or tightly coupled with local feature extraction, restricting the ability to robustly model these two complementary aspects. We propose DualTrack, a novel dual encoder architecture leveraging decoupled local and global encoders specializing in their respective scale of feature extraction. The local encoder uses …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Three-dimensional ultrasound (US) offers many clinical advantages over conventional 2D imaging, yet its widespread adoption is limited by the cost and complexity of traditional 3D systems. Sensorless 3D US, using deep learning to estimate a 3D probe trajectory from a sequence of 2D US images, is a promising alternative.Local features such as speckle patterns can help predict frame-to-frame motions, while global features, such as coarse shapes and anatomical structures, can situate the scan relative to anatomy and help predict its general shape. In prior approaches, global features are either ignored or tightly coupled with local feature extraction, restricting the ability to robustly model these two complementary aspects. We propose DualTrack, a novel dual encoder architecture leveraging decoupled local and global encoders specializing in their respective scale of feature extraction. The local encoder uses …
Jamzad, Amoon; Warren, Jade; Syeda, Ayesha; Kaufmann, Martin; Iaboni, Natasha; Nicol, Christopher JB; Rudan, John F; Ren, Kevin YM; Hurlbut, David; Varma, Sonal; Fichtinger, Gabor; Mousavi, Parvin
MassVision: An Open-Source End-to-End Platform for AI-Driven Mass Spectrometry Imaging Analysis Journal Article
In: Analytical Chemistry, vol. 97, no. 39, pp. 21588-21597, 2025.
@article{jamzad2025a,
title = {MassVision: An Open-Source End-to-End Platform for AI-Driven Mass Spectrometry Imaging Analysis},
author = {Amoon Jamzad and Jade Warren and Ayesha Syeda and Martin Kaufmann and Natasha Iaboni and Christopher JB Nicol and John F Rudan and Kevin YM Ren and David Hurlbut and Sonal Varma and Gabor Fichtinger and Parvin Mousavi},
year = {2025},
date = {2025-01-01},
journal = {Analytical Chemistry},
volume = {97},
number = {39},
pages = {21588-21597},
publisher = {American Chemical Society},
abstract = {Mass spectrometry imaging (MSI) combines spatial and spectral data to reveal detailed molecular compositions within biological samples. Despite their immense potential, MSI workflows are hindered by the complexity and high dimensionality of the data, making their analysis computationally intensive and often requiring expertise in coding. Existing tools frequently lack the integration needed for seamless, scalable, and end-to-end workflows, forcing researchers to rely on local solutions or multiple platforms, which hinders efficiency and accessibility. We introduce MassVision, a comprehensive software platform for MSI analysis. Built on the 3D Slicer ecosystem, MassVision integrates MSI-specific functionalities while addressing general user requirements for accessibility and usability. Its intuitive interface lowers barriers for researchers with varying levels of computational expertise, while its scalability supports high …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Mass spectrometry imaging (MSI) combines spatial and spectral data to reveal detailed molecular compositions within biological samples. Despite their immense potential, MSI workflows are hindered by the complexity and high dimensionality of the data, making their analysis computationally intensive and often requiring expertise in coding. Existing tools frequently lack the integration needed for seamless, scalable, and end-to-end workflows, forcing researchers to rely on local solutions or multiple platforms, which hinders efficiency and accessibility. We introduce MassVision, a comprehensive software platform for MSI analysis. Built on the 3D Slicer ecosystem, MassVision integrates MSI-specific functionalities while addressing general user requirements for accessibility and usability. Its intuitive interface lowers barriers for researchers with varying levels of computational expertise, while its scalability supports high …
Connolly, Laura; Ungi, Tamas; Munawar, Adnan; Deguet, Anton; Yeung, Chris; Taylor, Russell H; Mousavi, Parvin; Fichtinger, Gabor; Hashtrudi-Zaad, Keyvan
Touching the tumor boundary: a pilot study on ultrasound-based virtual fixtures for breast-conserving surgery Journal Article
In: International Journal of Computer Assisted Radiology and Surgery, vol. 20, no. 6, pp. 1105-1113, 2025.
@article{connolly2025,
title = {Touching the tumor boundary: a pilot study on ultrasound-based virtual fixtures for breast-conserving surgery},
author = {Laura Connolly and Tamas Ungi and Adnan Munawar and Anton Deguet and Chris Yeung and Russell H Taylor and Parvin Mousavi and Gabor Fichtinger and Keyvan Hashtrudi-Zaad},
year = {2025},
date = {2025-01-01},
journal = {International Journal of Computer Assisted Radiology and Surgery},
volume = {20},
number = {6},
pages = {1105-1113},
publisher = {Springer International Publishing},
abstract = {Purpose
Delineating tumor boundaries during breast-conserving surgery is challenging as tumors are often highly mobile, non-palpable, and have irregularly shaped borders. To address these challenges, we introduce a cooperative robotic guidance system that applies haptic feedback for tumor localization. In this pilot study, we aim to assess if and how this system can be successfully integrated into breast cancer care.
Methods
A small haptic robot is retrofitted with an electrocautery blade to operate as a cooperatively controlled surgical tool. Ultrasound and electromagnetic navigation are used to identify the tumor boundaries and position. A forbidden region virtual fixture is imposed when the surgical tool collides with the tumor boundary. We conducted a study where users were asked to resect tumors from breast simulants both with and without the haptic guidance. We then assess the results of these simulated …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Purpose
Delineating tumor boundaries during breast-conserving surgery is challenging as tumors are often highly mobile, non-palpable, and have irregularly shaped borders. To address these challenges, we introduce a cooperative robotic guidance system that applies haptic feedback for tumor localization. In this pilot study, we aim to assess if and how this system can be successfully integrated into breast cancer care.
Methods
A small haptic robot is retrofitted with an electrocautery blade to operate as a cooperatively controlled surgical tool. Ultrasound and electromagnetic navigation are used to identify the tumor boundaries and position. A forbidden region virtual fixture is imposed when the surgical tool collides with the tumor boundary. We conducted a study where users were asked to resect tumors from breast simulants both with and without the haptic guidance. We then assess the results of these simulated …
Delineating tumor boundaries during breast-conserving surgery is challenging as tumors are often highly mobile, non-palpable, and have irregularly shaped borders. To address these challenges, we introduce a cooperative robotic guidance system that applies haptic feedback for tumor localization. In this pilot study, we aim to assess if and how this system can be successfully integrated into breast cancer care.
Methods
A small haptic robot is retrofitted with an electrocautery blade to operate as a cooperatively controlled surgical tool. Ultrasound and electromagnetic navigation are used to identify the tumor boundaries and position. A forbidden region virtual fixture is imposed when the surgical tool collides with the tumor boundary. We conducted a study where users were asked to resect tumors from breast simulants both with and without the haptic guidance. We then assess the results of these simulated …
Harmanani, Mohamed; Jamzad, Amoon; To, Minh Nguyen Nhat; Wilson, Paul FR; Guo, Zhuoxin; Fooladgar, Fahimeh; Sojoudi, Samira; Gilany, Mahdi; Chang, Silvia; Black, Peter; Leveridge, Michael; Siemens, Robert; Abolmaesumi, Purang; Mousavi, Parvin
Cinepro: Robust training of foundation models for cancer detection in prostate ultrasound cineloops Proceedings Article
In: pp. 1-5, IEEE, 2025.
@inproceedings{harmanani2025a,
title = {Cinepro: Robust training of foundation models for cancer detection in prostate ultrasound cineloops},
author = {Mohamed Harmanani and Amoon Jamzad and Minh Nguyen Nhat To and Paul FR Wilson and Zhuoxin Guo and Fahimeh Fooladgar and Samira Sojoudi and Mahdi Gilany and Silvia Chang and Peter Black and Michael Leveridge and Robert Siemens and Purang Abolmaesumi and Parvin Mousavi},
year = {2025},
date = {2025-01-01},
pages = {1-5},
publisher = {IEEE},
abstract = {Prostate cancer (PCa) detection using deep learning (DL) models has shown potential for enhancing real-time guidance during biopsies. However, prostate ultrasound images lack pixel-level cancer annotations, introducing label noise. Current approaches often focus on limited regions of interest (ROIs), disregarding anatomical context necessary for accurate diagnosis. Foundation models can overcome this limitation by analyzing entire images to capture global spatial relationships; however, they still encounter challenges stemming from the weak labels associated with coarse pathology annotations in ultrasound data. We introduce Cinepro, a novel framework that strengthens foundation models' ability to localize PCa in ultrasound cineloops. Cinepro adapts robust training by integrating the proportion of cancer tissue reported by pathology in a biopsy core into its loss function to address label noise, providing a …},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Prostate cancer (PCa) detection using deep learning (DL) models has shown potential for enhancing real-time guidance during biopsies. However, prostate ultrasound images lack pixel-level cancer annotations, introducing label noise. Current approaches often focus on limited regions of interest (ROIs), disregarding anatomical context necessary for accurate diagnosis. Foundation models can overcome this limitation by analyzing entire images to capture global spatial relationships; however, they still encounter challenges stemming from the weak labels associated with coarse pathology annotations in ultrasound data. We introduce Cinepro, a novel framework that strengthens foundation models' ability to localize PCa in ultrasound cineloops. Cinepro adapts robust training by integrating the proportion of cancer tissue reported by pathology in a biopsy core into its loss function to address label noise, providing a …
Rubino, Rachel E; Kaufmann, Martin; Jamzad, Amoon; Iaboni, Natasha; Ren, Kevin Yi Mi; Yu, Jian; Metwally, Haidy; Kunz, Manuela; Rudan, John F; Mousavi, Parvin; Fichtinger, Gabor; Oleschuk, Richard; Nicol, Christopher JB
Evaluating rapid evaporative ionization mass spectrometry profiles of human breast cancer cells using a 3D in vitro assay: Rubino et al. Journal Article
In: Analytical and Bioanalytical Chemistry, vol. 417, no. 22, pp. 5115-5130, 2025.
@article{rubino2025,
title = {Evaluating rapid evaporative ionization mass spectrometry profiles of human breast cancer cells using a 3D in vitro assay: Rubino et al.},
author = {Rachel E Rubino and Martin Kaufmann and Amoon Jamzad and Natasha Iaboni and Kevin Yi Mi Ren and Jian Yu and Haidy Metwally and Manuela Kunz and John F Rudan and Parvin Mousavi and Gabor Fichtinger and Richard Oleschuk and Christopher JB Nicol},
year = {2025},
date = {2025-01-01},
journal = {Analytical and Bioanalytical Chemistry},
volume = {417},
number = {22},
pages = {5115-5130},
publisher = {Springer Berlin Heidelberg},
abstract = {Rapid evaporative ionization mass spectrometry (REIMS) enables near real-time sampling and classification of tissues on the basis of lipid and small molecule profiles. Because samples are ablated during REIMS, validation of mass spectra by assignment of class labels using gold-standard methods remains challenging. Further, determining the number of abnormal cells within a mixture of normal cells by REIMS remains an elusive but critical parameter when evaluating the utility of REIMS for use in clinical settings. To address this, we developed a three-dimensional assay based on fluorescently labelled human breast cell lines, MCF-10A (pseudo-normal) and MDA-MB-231 (malignant), embedded in matrigel cubes that enables calculation of cell numbers ablated by REIMS by measuring fluorescence before and after REIMS sampling. We observed a reasonably uniform distribution of cells throughout the matrigel …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Rapid evaporative ionization mass spectrometry (REIMS) enables near real-time sampling and classification of tissues on the basis of lipid and small molecule profiles. Because samples are ablated during REIMS, validation of mass spectra by assignment of class labels using gold-standard methods remains challenging. Further, determining the number of abnormal cells within a mixture of normal cells by REIMS remains an elusive but critical parameter when evaluating the utility of REIMS for use in clinical settings. To address this, we developed a three-dimensional assay based on fluorescently labelled human breast cell lines, MCF-10A (pseudo-normal) and MDA-MB-231 (malignant), embedded in matrigel cubes that enables calculation of cell numbers ablated by REIMS by measuring fluorescence before and after REIMS sampling. We observed a reasonably uniform distribution of cells throughout the matrigel …
Radcliffe, Olivia; Lawford-Wickham, Imogen A; Wilson, Paul FR; Abolmaesumi, Purang; Mousavi, Parvin
Leveraging SAM for automatic prostate segmentation on micro-ultrasound images Proceedings Article
In: pp. 188-194, SPIE, 2025.
@inproceedings{radcliffe2025a,
title = {Leveraging SAM for automatic prostate segmentation on micro-ultrasound images},
author = {Olivia Radcliffe and Imogen A Lawford-Wickham and Paul FR Wilson and Purang Abolmaesumi and Parvin Mousavi},
year = {2025},
date = {2025-01-01},
volume = {13412},
pages = {188-194},
publisher = {SPIE},
abstract = {PURPOSE
Deep learning is promising for enabling accurate and automatic prostate segmentation. Existing deep learning segmentation model approaches often rely on large training datasets for good generalization. We aim to adapt pre-trained foundation models to enable accurate, data-efficient, and robust prostate segmentation.
METHODS
We adapt the promptable foundation model SAM1 for prostate segmentation on rotational micro-ultrasound scans2 of the prostate. We design novel prompting strategies to provide the model with enhanced 3D context awareness to improve segmentation.
RESULTS
Our model, which we call SliceTrack-SAM, outperforms prior state-of-the-art in micro-ultrasound prostate segmentation. We achieve a Dice coefficient of 94.0% and Hausdorff distance 1.76mm on the MicroSegNet dataset. Qualitative analysis and ablation studies further validate the success of our approach …},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
PURPOSE
Deep learning is promising for enabling accurate and automatic prostate segmentation. Existing deep learning segmentation model approaches often rely on large training datasets for good generalization. We aim to adapt pre-trained foundation models to enable accurate, data-efficient, and robust prostate segmentation.
METHODS
We adapt the promptable foundation model SAM1 for prostate segmentation on rotational micro-ultrasound scans2 of the prostate. We design novel prompting strategies to provide the model with enhanced 3D context awareness to improve segmentation.
RESULTS
Our model, which we call SliceTrack-SAM, outperforms prior state-of-the-art in micro-ultrasound prostate segmentation. We achieve a Dice coefficient of 94.0% and Hausdorff distance 1.76mm on the MicroSegNet dataset. Qualitative analysis and ablation studies further validate the success of our approach …
Deep learning is promising for enabling accurate and automatic prostate segmentation. Existing deep learning segmentation model approaches often rely on large training datasets for good generalization. We aim to adapt pre-trained foundation models to enable accurate, data-efficient, and robust prostate segmentation.
METHODS
We adapt the promptable foundation model SAM1 for prostate segmentation on rotational micro-ultrasound scans2 of the prostate. We design novel prompting strategies to provide the model with enhanced 3D context awareness to improve segmentation.
RESULTS
Our model, which we call SliceTrack-SAM, outperforms prior state-of-the-art in micro-ultrasound prostate segmentation. We achieve a Dice coefficient of 94.0% and Hausdorff distance 1.76mm on the MicroSegNet dataset. Qualitative analysis and ablation studies further validate the success of our approach …
Radcliffe, Olivia; Connolly, Laura; Jamzad, Amoon; Kaufmann, Martin; Merchant, Shaila; Engel, Jay; Walker, Ross; Varma, Sonal; Fichtinger, Gabor; Rudan, John; Mousavi, Parvin
Anomaly detection using intraoperative iKnife data: a comparative analysis in breast cancer surgery Journal Article
In: International Journal of Computer Assisted Radiology and Surgery, vol. 20, no. 9, pp. 1953-1963, 2025.
@article{radcliffe2025,
title = {Anomaly detection using intraoperative iKnife data: a comparative analysis in breast cancer surgery},
author = {Olivia Radcliffe and Laura Connolly and Amoon Jamzad and Martin Kaufmann and Shaila Merchant and Jay Engel and Ross Walker and Sonal Varma and Gabor Fichtinger and John Rudan and Parvin Mousavi},
year = {2025},
date = {2025-01-01},
journal = {International Journal of Computer Assisted Radiology and Surgery},
volume = {20},
number = {9},
pages = {1953-1963},
publisher = {Springer International Publishing},
abstract = {Purpose
Intraoperative margin assessment is crucial to ensure complete tumor removal and minimize the risk of cancer recurrence during breast-conserving surgery. The Intelligent Knife (iKnife), a mass spectrometry device that analyzes surgical smoke, shows promise in near-real-time margin evaluation. However, current AI models depend on labeled ex-vivo datasets, which are costly and time-consuming to produce. This research explores the potential of machine learning anomaly detection models to reduce reliance on labeled ex-vivo datasets by utilizing unlabeled intraoperative spectra.
Methods
iKnife spectra were collected intraoperatively from 15 breast cancer surgeries. Ex-vivo samples were recorded from the resected specimen by a pathologist. Healthy samples were from the margin, and tumor samples were from the cross-section. We trained four anomaly detection methods, Isolation Forest (iForest), One …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Purpose
Intraoperative margin assessment is crucial to ensure complete tumor removal and minimize the risk of cancer recurrence during breast-conserving surgery. The Intelligent Knife (iKnife), a mass spectrometry device that analyzes surgical smoke, shows promise in near-real-time margin evaluation. However, current AI models depend on labeled ex-vivo datasets, which are costly and time-consuming to produce. This research explores the potential of machine learning anomaly detection models to reduce reliance on labeled ex-vivo datasets by utilizing unlabeled intraoperative spectra.
Methods
iKnife spectra were collected intraoperatively from 15 breast cancer surgeries. Ex-vivo samples were recorded from the resected specimen by a pathologist. Healthy samples were from the margin, and tumor samples were from the cross-section. We trained four anomaly detection methods, Isolation Forest (iForest), One …
Intraoperative margin assessment is crucial to ensure complete tumor removal and minimize the risk of cancer recurrence during breast-conserving surgery. The Intelligent Knife (iKnife), a mass spectrometry device that analyzes surgical smoke, shows promise in near-real-time margin evaluation. However, current AI models depend on labeled ex-vivo datasets, which are costly and time-consuming to produce. This research explores the potential of machine learning anomaly detection models to reduce reliance on labeled ex-vivo datasets by utilizing unlabeled intraoperative spectra.
Methods
iKnife spectra were collected intraoperatively from 15 breast cancer surgeries. Ex-vivo samples were recorded from the resected specimen by a pathologist. Healthy samples were from the margin, and tumor samples were from the cross-section. We trained four anomaly detection methods, Isolation Forest (iForest), One …
Elghareb, Tarek; Harmanani, Mohamed; To, Minh Nguyen Nhat; Wilson, Paul; Jamzad, Amoon; Fooladgar, Fahimeh; Abdelsamad, Baraa; Dzikunu, Obed; Sojoudi, Samira; Reznik, Gabrielle; Leveridge, Michael; Siemens, Robert; Chang, Silvia; Black, Peter; Mousavi, Parvin; Abolmaesumi, Purang
Proteus: a spatio-temporal enhanced ultrasound-based framework for prostate cancer detection Journal Article
In: pp. 405-414, 2025.
@article{elghareb2025,
title = {Proteus: a spatio-temporal enhanced ultrasound-based framework for prostate cancer detection},
author = {Tarek Elghareb and Mohamed Harmanani and Minh Nguyen Nhat To and Paul Wilson and Amoon Jamzad and Fahimeh Fooladgar and Baraa Abdelsamad and Obed Dzikunu and Samira Sojoudi and Gabrielle Reznik and Michael Leveridge and Robert Siemens and Silvia Chang and Peter Black and Parvin Mousavi and Purang Abolmaesumi},
year = {2025},
date = {2025-01-01},
pages = {405-414},
publisher = {Springer Nature Switzerland},
abstract = {Deep learning holds significant promise for enhancing real-time ultrasound-based prostate biopsy guidance through precise and effective tissue characterization. Despite recent advancements, prostate cancer (PCa) detection using ultrasound imaging still faces two critical challenges: (i) limited sensitivity to subtle tissue variations essential for detecting clinically significant disease, and (ii) weak and noisy labeling resulting from reliance on coarse annotations in histopathological reports. To address these issues, we introduce ProTeUS, an innovative spatio-temporal framework that integrates clinical metadata with comprehensive spatial and temporal ultrasound features extracted by a foundation model. Our method includes a novel hybrid, cancer involvement-aware loss function designed to enhance resilience against label noise and effectively learn distinct PCa signatures. Furthermore, we employ a progressive …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Deep learning holds significant promise for enhancing real-time ultrasound-based prostate biopsy guidance through precise and effective tissue characterization. Despite recent advancements, prostate cancer (PCa) detection using ultrasound imaging still faces two critical challenges: (i) limited sensitivity to subtle tissue variations essential for detecting clinically significant disease, and (ii) weak and noisy labeling resulting from reliance on coarse annotations in histopathological reports. To address these issues, we introduce ProTeUS, an innovative spatio-temporal framework that integrates clinical metadata with comprehensive spatial and temporal ultrasound features extracted by a foundation model. Our method includes a novel hybrid, cancer involvement-aware loss function designed to enhance resilience against label noise and effectively learn distinct PCa signatures. Furthermore, we employ a progressive …
Elghareb, Tarek; Jamzad, Amoon; To, Minh Nguyen Nhat; Fooladgar, Fahimeh; Wilson, Paul FR; Sojoudi, Samira; Reznik, Gabrielle; Leveridge, Michael; Siemens, Robert; Chang, Silvia; Black, Peter; Mousavi, Parvin; Abolmaesumi, Purang
Self-supervised prototype learning for spatio-temporal enhanced ultrasound-based prostate cancer detection Proceedings Article
In: pp. 1-5, IEEE, 2025.
@inproceedings{elghareb2025a,
title = {Self-supervised prototype learning for spatio-temporal enhanced ultrasound-based prostate cancer detection},
author = {Tarek Elghareb and Amoon Jamzad and Minh Nguyen Nhat To and Fahimeh Fooladgar and Paul FR Wilson and Samira Sojoudi and Gabrielle Reznik and Michael Leveridge and Robert Siemens and Silvia Chang and Peter Black and Parvin Mousavi and Purang Abolmaesumi},
year = {2025},
date = {2025-01-01},
pages = {1-5},
publisher = {IEEE},
abstract = {Prostate cancer detection in ultrasound data presents significant challenges due to the highly heterogeneous nature of the cancer and its appearance in ultrasound images. In this work, we introduce a novel multi-modal framework that integrates spatial information in ultrasound images with temporal time-series signals across a sequence of ultrasound images for robust prostate cancer detection. Our framework employs a clustering-based self-supervised learning approach, which allows the model to learn shared, complementary representations from combined spatio-temporal data. This approach not only captures the unique properties of spatial and temporal information, but also enhances the model's ability to handle noisy labels due to imperfect alignment between imaging data and histopathology reports, resulting in improved classification performance. Our method achieves a 7% improvement in AUC and a 10 …},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Prostate cancer detection in ultrasound data presents significant challenges due to the highly heterogeneous nature of the cancer and its appearance in ultrasound images. In this work, we introduce a novel multi-modal framework that integrates spatial information in ultrasound images with temporal time-series signals across a sequence of ultrasound images for robust prostate cancer detection. Our framework employs a clustering-based self-supervised learning approach, which allows the model to learn shared, complementary representations from combined spatio-temporal data. This approach not only captures the unique properties of spatial and temporal information, but also enhances the model's ability to handle noisy labels due to imperfect alignment between imaging data and histopathology reports, resulting in improved classification performance. Our method achieves a 7% improvement in AUC and a 10 …
Farahmand, Mohammad; Jamzad, Amoon; Fooladgar, Fahimeh; Connolly, Laura; Kaufmann, Martin; Ren, Kevin Yi Mi; Rudan, John; McKay, Doug; Fichtinger, Gabor; Mousavi, Parvin
FACT: foundation model for assessing cancer tissue margins with mass spectrometry Journal Article
In: International Journal of Computer Assisted Radiology and Surgery, vol. 20, no. 6, pp. 1097-1104, 2025.
@article{farahmand2025b,
title = {FACT: foundation model for assessing cancer tissue margins with mass spectrometry},
author = {Mohammad Farahmand and Amoon Jamzad and Fahimeh Fooladgar and Laura Connolly and Martin Kaufmann and Kevin Yi Mi Ren and John Rudan and Doug McKay and Gabor Fichtinger and Parvin Mousavi},
year = {2025},
date = {2025-01-01},
journal = {International Journal of Computer Assisted Radiology and Surgery},
volume = {20},
number = {6},
pages = {1097-1104},
publisher = {Springer International Publishing},
abstract = {Purpose
Accurately classifying tissue margins during cancer surgeries is crucial for ensuring complete tumor removal. Rapid Evaporative Ionization Mass Spectrometry (REIMS), a tool for real-time intraoperative margin assessment, generates spectra that require machine learning models to support clinical decision-making. However, the scarcity of labeled data in surgical contexts presents a significant challenge. This study is the first to develop a foundation model tailored specifically for REIMS data, addressing this limitation and advancing real-time surgical margin assessment.
Methods
We propose FACT, a Foundation model for Assessing Cancer Tissue margins. FACT is an adaptation of a foundation model originally designed for text-audio association, pretrained using our proposed supervised contrastive approach based on triplet loss. An ablation study is performed to compare our proposed model against other …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Purpose
Accurately classifying tissue margins during cancer surgeries is crucial for ensuring complete tumor removal. Rapid Evaporative Ionization Mass Spectrometry (REIMS), a tool for real-time intraoperative margin assessment, generates spectra that require machine learning models to support clinical decision-making. However, the scarcity of labeled data in surgical contexts presents a significant challenge. This study is the first to develop a foundation model tailored specifically for REIMS data, addressing this limitation and advancing real-time surgical margin assessment.
Methods
We propose FACT, a Foundation model for Assessing Cancer Tissue margins. FACT is an adaptation of a foundation model originally designed for text-audio association, pretrained using our proposed supervised contrastive approach based on triplet loss. An ablation study is performed to compare our proposed model against other …
Accurately classifying tissue margins during cancer surgeries is crucial for ensuring complete tumor removal. Rapid Evaporative Ionization Mass Spectrometry (REIMS), a tool for real-time intraoperative margin assessment, generates spectra that require machine learning models to support clinical decision-making. However, the scarcity of labeled data in surgical contexts presents a significant challenge. This study is the first to develop a foundation model tailored specifically for REIMS data, addressing this limitation and advancing real-time surgical margin assessment.
Methods
We propose FACT, a Foundation model for Assessing Cancer Tissue margins. FACT is an adaptation of a foundation model originally designed for text-audio association, pretrained using our proposed supervised contrastive approach based on triplet loss. An ablation study is performed to compare our proposed model against other …
Harmanani, Mohamed; Wilson, Paul FR; To, Minh Nguyen Nhat; Gilany, Mahdi; Jamzad, Amoon; Fooladgar, Fahimeh; Wodlinger, Brian; Abolmaesumi, Purang; Mousavi, Parvin
TRUSWorthy: toward clinically applicable deep learning for confident detection of prostate cancer in micro-ultrasound Journal Article
In: International journal of computer assisted radiology and surgery, vol. 20, no. 5, pp. 981-989, 2025.
@article{harmanani2025,
title = {TRUSWorthy: toward clinically applicable deep learning for confident detection of prostate cancer in micro-ultrasound},
author = {Mohamed Harmanani and Paul FR Wilson and Minh Nguyen Nhat To and Mahdi Gilany and Amoon Jamzad and Fahimeh Fooladgar and Brian Wodlinger and Purang Abolmaesumi and Parvin Mousavi},
year = {2025},
date = {2025-01-01},
journal = {International journal of computer assisted radiology and surgery},
volume = {20},
number = {5},
pages = {981-989},
publisher = {Springer International Publishing},
abstract = {Purpose
While deep learning methods have shown great promise in improving the effectiveness of prostate cancer (PCa) diagnosis by detecting suspicious lesions from trans-rectal ultrasound (TRUS), they must overcome multiple simultaneous challenges. There is high heterogeneity in tissue appearance, significant class imbalance in favor of benign examples, and scarcity in the number and quality of ground truth annotations available to train models. Failure to address even a single one of these problems can result in unacceptable clinical outcomes.
Methods
We propose TRUSWorthy, a carefully designed, tuned, and integrated system for reliable PCa detection. Our pipeline integrates self-supervised learning, multiple-instance learning aggregation using transformers, random-undersampled boosting and ensembling: These address label scarcity, weak labels, class imbalance, and overconfidence, respectively …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Purpose
While deep learning methods have shown great promise in improving the effectiveness of prostate cancer (PCa) diagnosis by detecting suspicious lesions from trans-rectal ultrasound (TRUS), they must overcome multiple simultaneous challenges. There is high heterogeneity in tissue appearance, significant class imbalance in favor of benign examples, and scarcity in the number and quality of ground truth annotations available to train models. Failure to address even a single one of these problems can result in unacceptable clinical outcomes.
Methods
We propose TRUSWorthy, a carefully designed, tuned, and integrated system for reliable PCa detection. Our pipeline integrates self-supervised learning, multiple-instance learning aggregation using transformers, random-undersampled boosting and ensembling: These address label scarcity, weak labels, class imbalance, and overconfidence, respectively …
While deep learning methods have shown great promise in improving the effectiveness of prostate cancer (PCa) diagnosis by detecting suspicious lesions from trans-rectal ultrasound (TRUS), they must overcome multiple simultaneous challenges. There is high heterogeneity in tissue appearance, significant class imbalance in favor of benign examples, and scarcity in the number and quality of ground truth annotations available to train models. Failure to address even a single one of these problems can result in unacceptable clinical outcomes.
Methods
We propose TRUSWorthy, a carefully designed, tuned, and integrated system for reliable PCa detection. Our pipeline integrates self-supervised learning, multiple-instance learning aggregation using transformers, random-undersampled boosting and ensembling: These address label scarcity, weak labels, class imbalance, and overconfidence, respectively …
Gabriel, Alon; Jamzad, Amoon; Farahmand, Mohammad; Kaufmann, Martin; Iaboni, Natasha; Hurlbut, David; Ren, Kevin Yi Mi; Nicol, Christopher JB; Rudan, John F; Varma, Sonal; Fichtinger, Gabor; Mousavi, Parvin
Application of foundation models for colorectal cancer tissue classification in mass spectrometry imaging Journal Article
In: Technologies, vol. 13, no. 10, pp. 434, 2025.
@article{gabriel2025,
title = {Application of foundation models for colorectal cancer tissue classification in mass spectrometry imaging},
author = {Alon Gabriel and Amoon Jamzad and Mohammad Farahmand and Martin Kaufmann and Natasha Iaboni and David Hurlbut and Kevin Yi Mi Ren and Christopher JB Nicol and John F Rudan and Sonal Varma and Gabor Fichtinger and Parvin Mousavi},
year = {2025},
date = {2025-01-01},
journal = {Technologies},
volume = {13},
number = {10},
pages = {434},
publisher = {MDPI},
abstract = {Colorectal cancer (CRC) remains a leading global health challenge, with early and accurate diagnosis crucial for effective treatment. Histopathological evaluation, the current diagnostic gold standard, faces limitations including subjectivity, delayed results, and reliance on well-prepared tissue slides. Mass spectrometry imaging (MSI) offers a complementary approach by providing molecular-level information, but its high dimensionality and the scarcity of labeled data present unique challenges for traditional supervised learning. In this study, we present the first implementation of foundation models for MSI-based cancer classification using desorption electrospray ionization (DESI) data. We evaluate multiple architectures adapted from other domains, including a spectral classification model known as FACT, which leverages audio–language pretraining. Compared to conventional machine learning approaches, these foundation models achieved superior performance, with FACT achieving the highest cross-validated balanced accuracy (93.27%±3.25%) and AUROC (98.4%±0.7%). Ablation studies demonstrate that these models retain strong performance even under reduced data conditions, highlighting their potential for generalizable and scalable MSI-based cancer diagnostics. Future work will explore the integration of spatial and multi-modal data to enhance clinical utility.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Colorectal cancer (CRC) remains a leading global health challenge, with early and accurate diagnosis crucial for effective treatment. Histopathological evaluation, the current diagnostic gold standard, faces limitations including subjectivity, delayed results, and reliance on well-prepared tissue slides. Mass spectrometry imaging (MSI) offers a complementary approach by providing molecular-level information, but its high dimensionality and the scarcity of labeled data present unique challenges for traditional supervised learning. In this study, we present the first implementation of foundation models for MSI-based cancer classification using desorption electrospray ionization (DESI) data. We evaluate multiple architectures adapted from other domains, including a spectral classification model known as FACT, which leverages audio–language pretraining. Compared to conventional machine learning approaches, these foundation models achieved superior performance, with FACT achieving the highest cross-validated balanced accuracy (93.27%±3.25%) and AUROC (98.4%±0.7%). Ablation studies demonstrate that these models retain strong performance even under reduced data conditions, highlighting their potential for generalizable and scalable MSI-based cancer diagnostics. Future work will explore the integration of spatial and multi-modal data to enhance clinical utility.
Nezamabadi, Kasra; Sivalokanathan, Sanjay; Lee, Jiwon; Tanriverdi, Talha; Chen, Meiling; Lu, Daiyin; Abraham, Jadyn; Sardaripour, Neda; Li, Pengyuan; Mousavi, Parvin; Abraham, M Roselle
Explainable artificial intelligence identifies and localizes left ventricular scar in hypertrophic cardiomyopathy using 12-Lead electrocardiogram Journal Article
In: Scientific Reports, vol. 15, no. 1, pp. 33918, 2025.
@article{nezamabadi2025,
title = {Explainable artificial intelligence identifies and localizes left ventricular scar in hypertrophic cardiomyopathy using 12-Lead electrocardiogram},
author = {Kasra Nezamabadi and Sanjay Sivalokanathan and Jiwon Lee and Talha Tanriverdi and Meiling Chen and Daiyin Lu and Jadyn Abraham and Neda Sardaripour and Pengyuan Li and Parvin Mousavi and M Roselle Abraham},
year = {2025},
date = {2025-01-01},
journal = {Scientific Reports},
volume = {15},
number = {1},
pages = {33918},
publisher = {Nature Publishing Group UK},
abstract = {Left ventricular (LV) scar is a major risk factor for sudden death and heart failure in hypertrophic cardiomyopathy (HCM). LV scar evolves over time and needs longitudinal assessment. Currently, LV scar detection relies on late gadolinium enhancement MRI, which is limited by high cost and artifacts from implanted cardiac devices. To address this, we developed XplainScar, an explainable machine learning model that identifies LV scar using 12-lead electrocardiogram (ECG) data. XplainScar was trained and validated on retrospective data from 748 HCM patients across two centers (500 from Johns Hopkins hospital for model development, and 248 from UCSF for validation). XplainScar employs a combination of unsupervised and self-supervised representation learning to effectively predict scar presence, and discover ECG features associated with LV scar. XplainScar rapidly analyzes ECG data (< 1 min for 10 …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Left ventricular (LV) scar is a major risk factor for sudden death and heart failure in hypertrophic cardiomyopathy (HCM). LV scar evolves over time and needs longitudinal assessment. Currently, LV scar detection relies on late gadolinium enhancement MRI, which is limited by high cost and artifacts from implanted cardiac devices. To address this, we developed XplainScar, an explainable machine learning model that identifies LV scar using 12-lead electrocardiogram (ECG) data. XplainScar was trained and validated on retrospective data from 748 HCM patients across two centers (500 from Johns Hopkins hospital for model development, and 248 from UCSF for validation). XplainScar employs a combination of unsupervised and self-supervised representation learning to effectively predict scar presence, and discover ECG features associated with LV scar. XplainScar rapidly analyzes ECG data (< 1 min for 10 …
To, Minh Nguyen Nhat; RWilson, Paul F; Nguyen, Viet; Harmanani, Mohamed; Cooper, Michael; Fooladgar, Fahimeh; Abolmaesumi, Purang; Mousavi, Parvin; Krishnan, Rahul G
Diverse prototypical ensembles improve robustness to subpopulation shift Journal Article
In: arXiv e-prints, pp. arXiv: 2505.23027, 2025.
@article{to2025,
title = {Diverse prototypical ensembles improve robustness to subpopulation shift},
author = {Minh Nguyen Nhat To and Paul F RWilson and Viet Nguyen and Mohamed Harmanani and Michael Cooper and Fahimeh Fooladgar and Purang Abolmaesumi and Parvin Mousavi and Rahul G Krishnan},
year = {2025},
date = {2025-01-01},
journal = {arXiv e-prints},
pages = {arXiv: 2505.23027},
abstract = {The subpopulationtion shift, characterized by a disparity in subpopulation distributibetween theween the training and target datasets, can significantly degrade the performance of machine learning models. Current solutions to subpopulation shift involve modifying empirical risk minimization with re-weighting strategies to improve generalization. This strategy relies on assumptions about the number and nature of subpopulations and annotations on group membership, which are unavailable for many real-world datasets. Instead, we propose using an ensemble of diverse classifiers to adaptively capture risk associated with subpopulations. Given a feature extractor network, we replace its standard linear classification layer with a mixture of prototypical classifiers, where each member is trained to classify the data while focusing on different features and samples from other members. In empirical evaluation on nine real …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The subpopulationtion shift, characterized by a disparity in subpopulation distributibetween theween the training and target datasets, can significantly degrade the performance of machine learning models. Current solutions to subpopulation shift involve modifying empirical risk minimization with re-weighting strategies to improve generalization. This strategy relies on assumptions about the number and nature of subpopulations and annotations on group membership, which are unavailable for many real-world datasets. Instead, we propose using an ensemble of diverse classifiers to adaptively capture risk associated with subpopulations. Given a feature extractor network, we replace its standard linear classification layer with a mixture of prototypical classifiers, where each member is trained to classify the data while focusing on different features and samples from other members. In empirical evaluation on nine real …
2024
To, Minh Nguyen Nhat; Fooladgar, Fahimeh; Wilson, Paul; Harmanani, Mohamed; Gilany, Mahdi; Sojoudi, Samira; Jamzad, Amoon; Chang, Silvia; Black, Peter; Mousavi, Parvin; Abolmaesumi, Purang
LensePro: Label noise-tolerant prototype-based network for improving cancer detection in prostate ultrasound with limited annotations Journal Article
In: International Journal of Computer Assisted Radiology and Surgery, vol. 19, no. 6, pp. 1121-1128, 2024.
@article{to2024,
title = {LensePro: Label noise-tolerant prototype-based network for improving cancer detection in prostate ultrasound with limited annotations},
author = {Minh Nguyen Nhat To and Fahimeh Fooladgar and Paul Wilson and Mohamed Harmanani and Mahdi Gilany and Samira Sojoudi and Amoon Jamzad and Silvia Chang and Peter Black and Parvin Mousavi and Purang Abolmaesumi},
year = {2024},
date = {2024-01-01},
journal = {International Journal of Computer Assisted Radiology and Surgery},
volume = {19},
number = {6},
pages = {1121-1128},
publisher = {Springer International Publishing},
abstract = {Purpose
The standard of care for prostate cancer (PCa) diagnosis is the histopathological analysis of tissue samples obtained via transrectal ultrasound (TRUS) guided biopsy. Models built with deep neural networks (DNNs) hold the potential for direct PCa detection from TRUS, which allows targeted biopsy and subsequently enhances outcomes. Yet, there are ongoing challenges with training robust models, stemming from issues such as noisy labels, out-of-distribution (OOD) data, and limited labeled data.
Methods
This study presents LensePro, a unified method that not only excels in label efficiency but also demonstrates robustness against label noise and OOD data. LensePro comprises two key stages: first, self-supervised learning to extract high-quality feature representations from abundant unlabeled TRUS data and, second, label noise-tolerant prototype-based learning to classify the extracted features …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Purpose
The standard of care for prostate cancer (PCa) diagnosis is the histopathological analysis of tissue samples obtained via transrectal ultrasound (TRUS) guided biopsy. Models built with deep neural networks (DNNs) hold the potential for direct PCa detection from TRUS, which allows targeted biopsy and subsequently enhances outcomes. Yet, there are ongoing challenges with training robust models, stemming from issues such as noisy labels, out-of-distribution (OOD) data, and limited labeled data.
Methods
This study presents LensePro, a unified method that not only excels in label efficiency but also demonstrates robustness against label noise and OOD data. LensePro comprises two key stages: first, self-supervised learning to extract high-quality feature representations from abundant unlabeled TRUS data and, second, label noise-tolerant prototype-based learning to classify the extracted features …
The standard of care for prostate cancer (PCa) diagnosis is the histopathological analysis of tissue samples obtained via transrectal ultrasound (TRUS) guided biopsy. Models built with deep neural networks (DNNs) hold the potential for direct PCa detection from TRUS, which allows targeted biopsy and subsequently enhances outcomes. Yet, there are ongoing challenges with training robust models, stemming from issues such as noisy labels, out-of-distribution (OOD) data, and limited labeled data.
Methods
This study presents LensePro, a unified method that not only excels in label efficiency but also demonstrates robustness against label noise and OOD data. LensePro comprises two key stages: first, self-supervised learning to extract high-quality feature representations from abundant unlabeled TRUS data and, second, label noise-tolerant prototype-based learning to classify the extracted features …
Nguyen, Vivian; Jamzad, Amoon; Wilson, Paul; Gilany, Mahdi; Abolmaesumi, Purang; Leveridge, Michael; Siemens, Robert; Mousavi, Parvin
Assessing the performance of foundation models in prostate segmentation across different ultrasound modalities Proceedings Article
In: pp. 876-880, IEEE, 2024.
@inproceedings{nguyen2024,
title = {Assessing the performance of foundation models in prostate segmentation across different ultrasound modalities},
author = {Vivian Nguyen and Amoon Jamzad and Paul Wilson and Mahdi Gilany and Purang Abolmaesumi and Michael Leveridge and Robert Siemens and Parvin Mousavi},
year = {2024},
date = {2024-01-01},
pages = {876-880},
publisher = {IEEE},
abstract = {Prostate cancer (PCa) remains a major health issue for men and relies on early diagnostic procedures such as transrectal ultrasound-guided (TRUS) biopsy for effective treatment. Since this procedure is prone to high rates of false negatives, advancements are being made in ultrasound (US) imaging and artificial intelligence to improve the detection of cancerous lesions. Segmentation is necessary for establishing the area for analysis however, manual segmentation can be laborious and time-intensive. Recent studies have focused on leveraging pre-trained foundation models that can be applied to various downstream tasks such as prostate segmentation in US imaging. However, only the processed B-mode US images were used in these studies and other imaging modes like raw RF images, although proved superior in tissue characterization, have not been explored. This study evaluates the segmentation …},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Prostate cancer (PCa) remains a major health issue for men and relies on early diagnostic procedures such as transrectal ultrasound-guided (TRUS) biopsy for effective treatment. Since this procedure is prone to high rates of false negatives, advancements are being made in ultrasound (US) imaging and artificial intelligence to improve the detection of cancerous lesions. Segmentation is necessary for establishing the area for analysis however, manual segmentation can be laborious and time-intensive. Recent studies have focused on leveraging pre-trained foundation models that can be applied to various downstream tasks such as prostate segmentation in US imaging. However, only the processed B-mode US images were used in these studies and other imaging modes like raw RF images, although proved superior in tissue characterization, have not been explored. This study evaluates the segmentation …
Harmanani, Mohamed; Wilson, Paul FR; Fooladgar, Fahimeh; Jamzad, Amoon; Gilany, Mahdi; To, Minh Nguyen Nhat; Wodlinger, Brian; Abolmaesumi, Purang; Mousavi, Parvin
Benchmarking image transformers for prostate cancer detection from ultrasound data Proceedings Article
In: pp. 245-251, SPIE, 2024.
@inproceedings{harmanani2024,
title = {Benchmarking image transformers for prostate cancer detection from ultrasound data},
author = {Mohamed Harmanani and Paul FR Wilson and Fahimeh Fooladgar and Amoon Jamzad and Mahdi Gilany and Minh Nguyen Nhat To and Brian Wodlinger and Purang Abolmaesumi and Parvin Mousavi},
year = {2024},
date = {2024-01-01},
volume = {12928},
pages = {245-251},
publisher = {SPIE},
abstract = {PURPOSE
Deep learning methods for classifying prostate cancer (PCa) in ultrasound images typically employ convolutional neural networks (CNN) to detect cancer in small regions of interest (ROI) along a needle trace region. However, this approach suffers from weak labelling, since the ground-truth histopathology labels do not describe the properties of individual ROIs. Recently, multi-scale approaches have sought to mitigate this issue by combining the context awareness of transformers with a convolutional feature extractor to detect cancer from multiple ROIs using multiple-instance learning (MIL). In this work, we present a detailed study of several image transformer architectures for both ROI-scale and multi-scale classification, and a comparison of the performance of CNNs and transformers for ultrasound-based prostate cancer classification. We also design a novel multi-objective learning strategy that …},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
PURPOSE
Deep learning methods for classifying prostate cancer (PCa) in ultrasound images typically employ convolutional neural networks (CNN) to detect cancer in small regions of interest (ROI) along a needle trace region. However, this approach suffers from weak labelling, since the ground-truth histopathology labels do not describe the properties of individual ROIs. Recently, multi-scale approaches have sought to mitigate this issue by combining the context awareness of transformers with a convolutional feature extractor to detect cancer from multiple ROIs using multiple-instance learning (MIL). In this work, we present a detailed study of several image transformer architectures for both ROI-scale and multi-scale classification, and a comparison of the performance of CNNs and transformers for ultrasound-based prostate cancer classification. We also design a novel multi-objective learning strategy that …
Deep learning methods for classifying prostate cancer (PCa) in ultrasound images typically employ convolutional neural networks (CNN) to detect cancer in small regions of interest (ROI) along a needle trace region. However, this approach suffers from weak labelling, since the ground-truth histopathology labels do not describe the properties of individual ROIs. Recently, multi-scale approaches have sought to mitigate this issue by combining the context awareness of transformers with a convolutional feature extractor to detect cancer from multiple ROIs using multiple-instance learning (MIL). In this work, we present a detailed study of several image transformer architectures for both ROI-scale and multi-scale classification, and a comparison of the performance of CNNs and transformers for ultrasound-based prostate cancer classification. We also design a novel multi-objective learning strategy that …
Fooladgar, Fahimeh; To, Minh Nguyen Nhat; Mousavi, Parvin; Abolmaesumi, Purang
Manifold DivideMix: A semi-supervised contrastive learning framework for severe label noise Proceedings Article
In: pp. 4012-4021, IEEE, 2024.
@inproceedings{fooladgar2024b,
title = {Manifold DivideMix: A semi-supervised contrastive learning framework for severe label noise},
author = {Fahimeh Fooladgar and Minh Nguyen Nhat To and Parvin Mousavi and Purang Abolmaesumi},
year = {2024},
date = {2024-01-01},
pages = {4012-4021},
publisher = {IEEE},
abstract = {Deep neural networks have proven to be highly effective when large amounts of data with clean labels are available. However, their performance degrades when training data contains noisy labels, leading to poor generalization on the test set. Real-world datasets contain noisy label samples that either have similar visual semantics to other classes (in-distribution) or have no semantic relevance to any class (out-of-distribution) in the dataset. Most state-of-the-art methods leverage ID labeled noisy samples as unlabeled data for semi-supervised learning, but OOD labeled noisy samples cannot be used in this way because they do not belong to any class within the dataset. Hence, in this paper, we propose incorporating the information from all the training data by leveraging the benefits of self-supervised training. Our method aims to extract a meaningful and generalizable embedding space for each sample …},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Deep neural networks have proven to be highly effective when large amounts of data with clean labels are available. However, their performance degrades when training data contains noisy labels, leading to poor generalization on the test set. Real-world datasets contain noisy label samples that either have similar visual semantics to other classes (in-distribution) or have no semantic relevance to any class (out-of-distribution) in the dataset. Most state-of-the-art methods leverage ID labeled noisy samples as unlabeled data for semi-supervised learning, but OOD labeled noisy samples cannot be used in this way because they do not belong to any class within the dataset. Hence, in this paper, we propose incorporating the information from all the training data by leveraging the benefits of self-supervised training. Our method aims to extract a meaningful and generalizable embedding space for each sample …
Warren, Jade; Jamzad, Amoon; Jamaspishvili, Tamara; Iseman, Rachael; Syeda, Ayesha; Kaufmann, Martin; Rudan, John; Fichtinger, Gabor; Berman, David M; Mousavi, Parvin
Towards Improving Surgical Margins in Tumour Resection Using Mass Spectrometry Imaging Proceedings Article
In: pp. 554-558, IEEE, 2024.
@inproceedings{warren2024,
title = {Towards Improving Surgical Margins in Tumour Resection Using Mass Spectrometry Imaging},
author = {Jade Warren and Amoon Jamzad and Tamara Jamaspishvili and Rachael Iseman and Ayesha Syeda and Martin Kaufmann and John Rudan and Gabor Fichtinger and David M Berman and Parvin Mousavi},
year = {2024},
date = {2024-01-01},
pages = {554-558},
publisher = {IEEE},
abstract = {Successful cancer resection is limited by the inability to differentiate between cancer and normal tissue intraoperatively. Desorption electrospray ionization mass spectrometry imaging (DESI-MSI) is an emerging and powerful analytical technique that offers a rapid and low cost approach for assessing surgical margins by generating detailed metabolic profiles. However, exploiting this data for tissue characterization based on molecular signals requires machine learning methods to handle its complexity. In this work, we utilize machine learning models for the characterization of tissue using DESI-MSI data obtained from prostate tissue samples. We use ViPRE, a novel open-source software, to annotate a large DESI-MSI dataset. We explore various machine learning models and train test schemes for cancer classification. Cross-validation of our models result in high balanced accuracy, sensitivity and specificity for …},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Successful cancer resection is limited by the inability to differentiate between cancer and normal tissue intraoperatively. Desorption electrospray ionization mass spectrometry imaging (DESI-MSI) is an emerging and powerful analytical technique that offers a rapid and low cost approach for assessing surgical margins by generating detailed metabolic profiles. However, exploiting this data for tissue characterization based on molecular signals requires machine learning methods to handle its complexity. In this work, we utilize machine learning models for the characterization of tissue using DESI-MSI data obtained from prostate tissue samples. We use ViPRE, a novel open-source software, to annotate a large DESI-MSI dataset. We explore various machine learning models and train test schemes for cancer classification. Cross-validation of our models result in high balanced accuracy, sensitivity and specificity for …
Gilany, Mahdi; Harmanani, Mohamed; Wilson, Paul; To, Minh Nguyen Nhat; Jamzad, Amoon; Fooladgar, Fahimeh; Wodlinger, Brian; Abolmaesumi, Purang; Mousavi, Parvin
Calibrated diverse ensemble entropy minimization for robust test-time adaptation in prostate cancer detection Journal Article
In: pp. 361-371, 2024.
@article{gilany2024,
title = {Calibrated diverse ensemble entropy minimization for robust test-time adaptation in prostate cancer detection},
author = {Mahdi Gilany and Mohamed Harmanani and Paul Wilson and Minh Nguyen Nhat To and Amoon Jamzad and Fahimeh Fooladgar and Brian Wodlinger and Purang Abolmaesumi and Parvin Mousavi},
year = {2024},
date = {2024-01-01},
pages = {361-371},
publisher = {Springer Nature Switzerland},
abstract = {High resolution micro-ultrasound has demonstrated promise in real-time prostate cancer detection, with deep learning becoming a prominent tool for learning complex tissue properties reflected on ultrasound. However, a significant roadblock to real-world deployment remains, which prior works often overlook: model performance suffers when applied to data from different clinical centers due to variations in data distribution. This distribution shift significantly impacts the model’s robustness, posing major challenge to clinical deployment. Domain adaptation and specifically its test-time adaption (TTA) variant offer a promising solution to address this challenge. In a setting designed to reflect real-world conditions, we compare existing methods to state-of-the-art TTA approaches adopted for cancer detection, demonstrating the lack of robustness to distribution shifts in the former. We then propose Diverse Ensemble Entropy …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
High resolution micro-ultrasound has demonstrated promise in real-time prostate cancer detection, with deep learning becoming a prominent tool for learning complex tissue properties reflected on ultrasound. However, a significant roadblock to real-world deployment remains, which prior works often overlook: model performance suffers when applied to data from different clinical centers due to variations in data distribution. This distribution shift significantly impacts the model’s robustness, posing major challenge to clinical deployment. Domain adaptation and specifically its test-time adaption (TTA) variant offer a promising solution to address this challenge. In a setting designed to reflect real-world conditions, we compare existing methods to state-of-the-art TTA approaches adopted for cancer detection, demonstrating the lack of robustness to distribution shifts in the former. We then propose Diverse Ensemble Entropy …
Mannina, Sophia; Addas, Shamel; Abolmaesumi, Purang; Mousavi, Parvin; Maghsoodi, Nooshin; Nassar, Sarah; Maslove, David; Sibley, Stephanie
Digital nudges in healthcare contexts: An information systems perspective Journal Article
In: 2024.
@article{mannina2024,
title = {Digital nudges in healthcare contexts: An information systems perspective},
author = {Sophia Mannina and Shamel Addas and Purang Abolmaesumi and Parvin Mousavi and Nooshin Maghsoodi and Sarah Nassar and David Maslove and Stephanie Sibley},
year = {2024},
date = {2024-01-01},
abstract = {Digital transformation has presented healthcare providers with new tools, roles, and challenges related to patient care. Although digital technologies like electronic health records can offer valuable information concerning patients' needs, the growing volume of data that healthcare providers receive through these tools can contribute to information overload and alert fatigue. Nudging is a behavioural economics technique that can be applied to guide healthcare providers toward optimal care decisions while limiting information overload. To better understand the application of this technique, we perform a systematic literature review that explores digital nudges oriented toward healthcare providers from an information systems perspective. This review identifies positive and negative outcomes of digital nudges and presents design principles that can guide development of nudges directed toward healthcare providers. Opportunities are discussed to further assess digital nudges through the information systems lens.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Digital transformation has presented healthcare providers with new tools, roles, and challenges related to patient care. Although digital technologies like electronic health records can offer valuable information concerning patients' needs, the growing volume of data that healthcare providers receive through these tools can contribute to information overload and alert fatigue. Nudging is a behavioural economics technique that can be applied to guide healthcare providers toward optimal care decisions while limiting information overload. To better understand the application of this technique, we perform a systematic literature review that explores digital nudges oriented toward healthcare providers from an information systems perspective. This review identifies positive and negative outcomes of digital nudges and presents design principles that can guide development of nudges directed toward healthcare providers. Opportunities are discussed to further assess digital nudges through the information systems lens.
Wilson, Paul FR; Harmanani, Mohamed; To, Minh Nguyen Nhat; Gilany, Mahdi; Jamzad, Amoon; Fooladgar, Fahimeh; Wodlinger, Brian; Abolmaesumi, Purang; Mousavi, Parvin
Toward confident prostate cancer detection using ultrasound: a multi-center study Journal Article
In: International Journal of Computer Assisted Radiology and Surgery, pp. 1-9, 2024.
@article{wilson2024,
title = {Toward confident prostate cancer detection using ultrasound: a multi-center study},
author = {Paul FR Wilson and Mohamed Harmanani and Minh Nguyen Nhat To and Mahdi Gilany and Amoon Jamzad and Fahimeh Fooladgar and Brian Wodlinger and Purang Abolmaesumi and Parvin Mousavi},
year = {2024},
date = {2024-01-01},
journal = {International Journal of Computer Assisted Radiology and Surgery},
pages = {1-9},
publisher = {Springer International Publishing},
abstract = {Purpose
Deep learning-based analysis of micro-ultrasound images to detect cancerous lesions is a promising tool for improving prostate cancer (PCa) diagnosis. An ideal model should confidently identify cancer while responding with appropriate uncertainty when presented with out-of-distribution inputs that arise during deployment due to imaging artifacts and the biological heterogeneity of patients and prostatic tissue.
Methods
Using micro-ultrasound data from 693 patients across 5 clinical centers who underwent micro-ultrasound guided prostate biopsy, we train and evaluate convolutional neural network models for PCa detection. To improve robustness to out-of-distribution inputs, we employ and comprehensively benchmark several state-of-the-art uncertainty estimation methods.
Results
PCa detection models achieve performance scores up to average AUROC with a 10-fold cross validation setup. Models with …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Purpose
Deep learning-based analysis of micro-ultrasound images to detect cancerous lesions is a promising tool for improving prostate cancer (PCa) diagnosis. An ideal model should confidently identify cancer while responding with appropriate uncertainty when presented with out-of-distribution inputs that arise during deployment due to imaging artifacts and the biological heterogeneity of patients and prostatic tissue.
Methods
Using micro-ultrasound data from 693 patients across 5 clinical centers who underwent micro-ultrasound guided prostate biopsy, we train and evaluate convolutional neural network models for PCa detection. To improve robustness to out-of-distribution inputs, we employ and comprehensively benchmark several state-of-the-art uncertainty estimation methods.
Results
PCa detection models achieve performance scores up to average AUROC with a 10-fold cross validation setup. Models with …
Deep learning-based analysis of micro-ultrasound images to detect cancerous lesions is a promising tool for improving prostate cancer (PCa) diagnosis. An ideal model should confidently identify cancer while responding with appropriate uncertainty when presented with out-of-distribution inputs that arise during deployment due to imaging artifacts and the biological heterogeneity of patients and prostatic tissue.
Methods
Using micro-ultrasound data from 693 patients across 5 clinical centers who underwent micro-ultrasound guided prostate biopsy, we train and evaluate convolutional neural network models for PCa detection. To improve robustness to out-of-distribution inputs, we employ and comprehensively benchmark several state-of-the-art uncertainty estimation methods.
Results
PCa detection models achieve performance scores up to average AUROC with a 10-fold cross validation setup. Models with …
Akbarifar, Faranak; Dukelow, Sean P; Jin, Albert; Mousavi, Parvin; Scott, Stephen H
Optimizing Stroke Detection Using Evidential Networks and Uncertainty-Based Refinement Journal Article
In: 2024.
@article{akbarifar2024,
title = {Optimizing Stroke Detection Using Evidential Networks and Uncertainty-Based Refinement},
author = {Faranak Akbarifar and Sean P Dukelow and Albert Jin and Parvin Mousavi and Stephen H Scott},
year = {2024},
date = {2024-01-01},
abstract = {Background:
Technologies such as interactive robotics and motion capture systems permit the development of kinematic-based approaches to assess motor impairments in stroke survivors. Here we utilise the Kinarm Exoskeleton robotic system and deep learning techniques to explore differences in motor performance between healthy controls, individuals with stroke and transient ischemic attacks (TIA).
Methods:
Building upon previous research that employed deep learning methods to distinguish between minimally impaired stroke patients and healthy controls using Kinarm data, this study introduces a novel dimension by estimating the confidence or uncertainty of the model's predictions. An evidential network is employed to measure this confidence, which subsequently aids in the refinement of training and testing datasets.
Results:
The application of deep learning techniques in this context proves to be promising …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Background:
Technologies such as interactive robotics and motion capture systems permit the development of kinematic-based approaches to assess motor impairments in stroke survivors. Here we utilise the Kinarm Exoskeleton robotic system and deep learning techniques to explore differences in motor performance between healthy controls, individuals with stroke and transient ischemic attacks (TIA).
Methods:
Building upon previous research that employed deep learning methods to distinguish between minimally impaired stroke patients and healthy controls using Kinarm data, this study introduces a novel dimension by estimating the confidence or uncertainty of the model's predictions. An evidential network is employed to measure this confidence, which subsequently aids in the refinement of training and testing datasets.
Results:
The application of deep learning techniques in this context proves to be promising …
Technologies such as interactive robotics and motion capture systems permit the development of kinematic-based approaches to assess motor impairments in stroke survivors. Here we utilise the Kinarm Exoskeleton robotic system and deep learning techniques to explore differences in motor performance between healthy controls, individuals with stroke and transient ischemic attacks (TIA).
Methods:
Building upon previous research that employed deep learning methods to distinguish between minimally impaired stroke patients and healthy controls using Kinarm data, this study introduces a novel dimension by estimating the confidence or uncertainty of the model's predictions. An evidential network is employed to measure this confidence, which subsequently aids in the refinement of training and testing datasets.
Results:
The application of deep learning techniques in this context proves to be promising …
Kim, Andrew; Yeung, Chris; Szabo, Robert; Sunderland, Kyle; Hisey, Rebecca; Morton, David; Kikinis, Ron; Diao, Babacar; Mousavi, Parvin; Ungi, Tamas; Fichtinger, Gabor
Percutaneous nephrostomy needle guidance using real-time 3D anatomical visualization with live ultrasound segmentation Proceedings Article
In: pp. 163-168, SPIE, 2024.
@inproceedings{kim2024,
title = {Percutaneous nephrostomy needle guidance using real-time 3D anatomical visualization with live ultrasound segmentation},
author = {Andrew Kim and Chris Yeung and Robert Szabo and Kyle Sunderland and Rebecca Hisey and David Morton and Ron Kikinis and Babacar Diao and Parvin Mousavi and Tamas Ungi and Gabor Fichtinger},
year = {2024},
date = {2024-01-01},
volume = {12928},
pages = {163-168},
publisher = {SPIE},
abstract = {PURPOSE
Percutaneous nephrostomy is a commonly performed procedure to drain urine to provide relief in patients with hydronephrosis. Conventional percutaneous nephrostomy needle guidance methods can be difficult, expensive, or not portable. We propose an open-source real-time 3D anatomical visualization aid for needle guidance with live ultrasound segmentation and 3D volume reconstruction using free, open-source software.
METHODS
Basic hydronephrotic kidney phantoms were created, and recordings of these models were manually segmented and used to train a deep learning model that makes live segmentation predictions to perform live 3D volume reconstruction of the fluid-filled cavity. Participants performed 5 needle insertions with the visualization aid and 5 insertions with ultrasound needle guidance on a kidney phantom in randomized order, and these were recorded. Recordings of the …},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
PURPOSE
Percutaneous nephrostomy is a commonly performed procedure to drain urine to provide relief in patients with hydronephrosis. Conventional percutaneous nephrostomy needle guidance methods can be difficult, expensive, or not portable. We propose an open-source real-time 3D anatomical visualization aid for needle guidance with live ultrasound segmentation and 3D volume reconstruction using free, open-source software.
METHODS
Basic hydronephrotic kidney phantoms were created, and recordings of these models were manually segmented and used to train a deep learning model that makes live segmentation predictions to perform live 3D volume reconstruction of the fluid-filled cavity. Participants performed 5 needle insertions with the visualization aid and 5 insertions with ultrasound needle guidance on a kidney phantom in randomized order, and these were recorded. Recordings of the …
Percutaneous nephrostomy is a commonly performed procedure to drain urine to provide relief in patients with hydronephrosis. Conventional percutaneous nephrostomy needle guidance methods can be difficult, expensive, or not portable. We propose an open-source real-time 3D anatomical visualization aid for needle guidance with live ultrasound segmentation and 3D volume reconstruction using free, open-source software.
METHODS
Basic hydronephrotic kidney phantoms were created, and recordings of these models were manually segmented and used to train a deep learning model that makes live segmentation predictions to perform live 3D volume reconstruction of the fluid-filled cavity. Participants performed 5 needle insertions with the visualization aid and 5 insertions with ultrasound needle guidance on a kidney phantom in randomized order, and these were recorded. Recordings of the …
Kaufmann, Martin; Jamzad, Amoon; Ungi, Tamas; Rodgers, Jessica; Koster, Teaghan; Chris, Yeung; Janssen, Natasja; McMullen, Julie; Solberg, Kathryn; Cheesman, Joanna; Ren, Kevin Ti Mi; Varma, Sonal; Merchant, Shaila; Engel, Cecil Jay; Walker, G Ross; Gallo, Andrea; Jabs, Doris; Mousavi, Parvin; Fichtinger, Gabor; Rudan, John
Three-dimensional navigated mass spectrometry for intraoperative margin assessment during breast cancer surgery Journal Article
In: vol. 31, no. 1, pp. S10-S10, 2024.
@article{kaufmann2024b,
title = {Three-dimensional navigated mass spectrometry for intraoperative margin assessment during breast cancer surgery},
author = {Martin Kaufmann and Amoon Jamzad and Tamas Ungi and Jessica Rodgers and Teaghan Koster and Yeung Chris and Natasja Janssen and Julie McMullen and Kathryn Solberg and Joanna Cheesman and Kevin Ti Mi Ren and Sonal Varma and Shaila Merchant and Cecil Jay Engel and G Ross Walker and Andrea Gallo and Doris Jabs and Parvin Mousavi and Gabor Fichtinger and John Rudan},
year = {2024},
date = {2024-01-01},
volume = {31},
number = {1},
pages = {S10-S10},
publisher = {SPRINGER},
keywords = {},
pubstate = {published},
tppubtype = {article}
}