Publications

@article{zachmann2026,

title = {Extended reality for enhancing surgery: Research directions to transform the operating environment},

author = {Gabriel Zachmann and Peter Haddawy and Ron Kikinis and Anja Hennemuth and Philippe C. Cattin and Tanja Döring and Simon Drouin and Gábor Fichtinger and Tabea Flügge and Joaquim A. Jorge and Luv Kohli and Mario Lorenz and Rainer Malaka and Javier Pascau and Dirk Reiners and Anke V. Reinschluessel and Andrea Schenk and Falko Schmid and Siriwan Suebnukarn and Verena Uslar and Gregory F. Welch and René Weller and Dirk Weyhe and Myat Su Yin and Sudanthi Wijewickrema},

year  = {2026},

date = {2026-01-01},

journal = {Journal of Healthcare Informatics Research},

abstract = {Abstract In the past decade, virtual reality (VR) and augmented reality (AR) have seen a second wave of activity due to consumer-level availability of devices. In the medical realm, use of VR and AR has been investigated extensively and successfully for training, teaching, rehabilitation, and therapy. However, extended reality (XR) and its intraoperative use in the operating room has yet been underexplored and underdeveloped. Intelligent XR environments that integrate advanced visualization, intuitive interaction, and AI-assisted decision support have the potential to transform clinical care. Responsive, context-aware spaces could enhance medical teams’ capabilities by enabling real-time access to critical information, enhancing situation awareness, and supporting remote collaboration and access to clinical expertise across geographical boundaries—all while preserving the natural flow of procedures. But to realize the potential benefits of XR in the operating room requires a significant and focused effort by an interdisciplinary community. It is the intention of this paper to help to catalyze such an effort. In this paper, we identify pressing issues in clinical care, the potential of XR to address them, and the further research that is needed to realize that potential. We look at a broad range of areas where XR can play an important role, including direct support for the surgeon, support for the surgical team, support for the patient, and remote collaboration. We also consider ways in which XR can be integrated with AI to provide enhanced information and interaction. Discussions are underpinned by considerations of methods to evaluate contributions and progress. This paper is the result of a Dagstuhl Seminar on Extended Reality for the Operating Room (XR4OR) , held at the Leibniz Center for Informatics, Schloss Dagstuhl, Germany, in February 2025. It gathered twenty-four researchers working in the areas of virtual reality, augmented reality, medical informatics, human-computer interaction, and surgery.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@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 Gábor Fichtinger and John F. Rudan and Parvin Mousavi},

year  = {2026},

date = {2026-01-01},

journal = {arXiv (Cornell University)},

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}

}

@article{siddiqui2026,

title = {Registration of clinical MRI volumes to low-cost manikin heads for neuronavigation training},

author = {Shahbaaz Siddiqui and Colton Barr and András Lassó and Gábor Fichtinger},

year  = {2026},

date = {2026-01-01},

pages = {43},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{rosenfeld2026,

title = {Validating RGB-D object detection for skill assessment in central venous catheterization using expert benchmarks},

author = {Josh Rosenfeld and Rebecca Hisey and Denise R. Wong and Ryan Varma and Gábor Fichtinger},

year  = {2026},

date = {2026-01-01},

pages = {41},

abstract = {Purpose: Computer-assisted surgical skill assessment has traditionally used physical sensors to track tool motion, which adds cost, setup time and can affect instrument handling. End-to-end video classifiers predict skill from video but offer limited interpretability. This study tests a low-cost, camera-based alternative for central venous catheterization that detects tools and computes depth-derived motion metrics. We compare the correlation with expert-rated OSATS scores for video-derived and EM-tracked path length to identify the better-aligned method. Methods: Five novices and five experts each performed ten trials on a venous access phantom, yielding 100 recordings. An RGB-D camera captured synchronized colour and depth video, and EM sensors recorded motion for the ultrasound probe and syringe. Videos were anonymized and scored with OSATS by four expert raters, then averaged per trial. A YOLOv8 network produced bounding boxes on the colour stream. Path length in millimetres was computed from depth for all annotated tools, and from EM tracking for the ultrasound probe and syringe using the Perk Tutor extension of 3D Slicer. Spearman rank correlations were computed between OSATS and path length per tool for both methods. Results: Ultrasound probe correlations with OSATS were −0.45 (p<0.001) for video and −0.35 (p<0.001) for EM. Syringe correlations were −0.58 (p<0.001) for video and −0.45 (p<0.001) for EM. No other tools showed a significant correlation. Conclusions: Object detection-derived path length aligned more strongly with expert ratings than EM-tracked path length, indicating a practical, low-cost, camera-based alternative.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{orosz2026,

title = {Explainable transfer learning ensemble AI model for lung ultrasound pneumothorax detection with expert benchmark},

author = {Gábor Orosz and Róbert Zsolt Szabó and Marcell Szabó and Pál Gyombolai and József T Tóth and Tamás Ruttkay and Tamás Ferenci and Tamás Ungi and Gábor Fichtinger and Tamás Haidegger},

year  = {2026},

date = {2026-01-01},

journal = {Scandinavian Journal of Trauma, Resuscitation and Emergency Medicine},

volume = {34},

number = {1},

pages = {95},

publisher = {Biomed Central},

abstract = {Background 

Lung ultrasound is essential for rapid, radiation-free bedside pneumothorax diagnosis but limited by variability in human interpretation. Key gaps include insufficiently large and diverse human datasets, inconsistent image acquisition, lack of rigorous expert benchmarking, and inadequate clinical interpretability of existing artificial intelligence models. We aimed to develop and validate a robust, explainable artificial intelligence (AI) ensemble model addressing these critical gaps. 

Methods 

With our multidisciplinary team, we developed an explainable soft-voting ensemble model trained on 1,856 diverse ultrasound clips from critically ill patients, healthy volunteers, and tailored cadaver models. Model interpretability was ensured using visualization, with heatmaps validated by expert clinicians. The model’s diagnostic performance was rigorously benchmarked against 11 experienced clinicians using an …},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@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 Gábor 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},

volume = {PP},

pages = {1-11},

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-based classifiers. However, the difference between DL and feature-based classifiers is minimal and highly dependent on the model selected, with feature-based classifiers obtaining a very slightly higher average performance on our ICU test set but DL achieving a higher overall maximum performance. The models that achieved the top F1 score on our ICU test set were InceptionV3 with recurrence plots and a fine-tuned ECGFounder (F1 = 0.88). CONCLUSION: This study demonstrates promising potential for using AI to build an automatic patient monitoring system. SIGNIFICANCE: By publishing our labelled ICU dataset 1 and performance benchmarks, this work enables the research community to continue advancing the state-of-the-art in AF detection in the ICU environment.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@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 Y. M. Ren and Shaila Merchant and Sonal Varma and Ross Walker and Doug McKay and JOHN RUDAN and Gábor Fichtinger and Parvin Mousavi},

year  = {2026},

date = {2026-01-01},

journal = {arXiv (Cornell University)},

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}

}

@article{logan2026,

title = {Evaluating vibrotactile feedback for electromagnetic surgical navigation},

author = {Stuart Logan and Laura Connolly and Colton Barr and Parvin Mousavi and Gábor Fichtinger and Keyvan Hashtrudi-Zaad},

year  = {2026},

date = {2026-01-01},

pages = {73},

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 hardware and software architecture are feasible for use, with minimal field distortion when the tactor is positioned greater than 10 cm from the electromagnetic (EM) sensor. End-to-end latency of 156 ± 45 ms for the wireless configuration, near the upper perceptual threshold for tactile feedback, and 126 ± 10 ms when using a wired connection. Finally, tactor activation induced negligible instrument motion, increasing cautery tip displacement by only 0.2 mm compared to the non-activated condition. CONCLUSIONS: These findings suggest that with proper tactor placement and communication configuration, the system can provide intraoperative tactile feedback without compromising navigation accuracy, supporting safer tumour margin identification and with further development, potentially reduce reoperation rates in BCS.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{hisey2026,

title = {Feasibility of Learner-Led Model Building for Simulation-Based Training in Open Inguinal Hernia Repair: A Randomized Controlled Trial},

author = {Rebecca Hisey and Denesh Peramakumar and Vanessa Wiseman and Farah Zaza and Iona Abolmaesumi and Elizabeth Klosa and Aden Wong and Gabor Fichtinger and Boris Zevin},

year  = {2026},

date = {2026-01-01},

journal = {Journal of Surgical Education},

volume = {83},

number = {9},

pages = {104054},

publisher = {Elsevier},

abstract = {Objectives (1) To establish the feasibility of novice learners building inguinal hernia models for simulation-based training in open inguinal hernia repairs (IHR). (2) To examine the associations between participation in model building and learners’ self-reported confidence in understanding of the inguinal canal anatomy, the anatomic accuracy of learner-constructed models, and technical performance during a simulated open IHR. We hypothesized that learner-led model construction is feasible, supports acquisition of relevant anatomy knowledge, and improves performance in some steps of a simulated open IHR. Design Prospective randomized controlled trial. Setting Queen’s University, Kingston, Ontario, Canada; institutional academic training environment. Participants Forty-eight novice learners (second-year medical students) with minimal prior exposure to open or laparoscopic IHR were enrolled and randomly …},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@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 Wilson and Amirhossein Sabour and Minh Nguyen Nhat To and Gábor Fichtinger and Purang Abolmaesumi and Parvin Mousavi},

year  = {2026},

date = {2026-01-01},

journal = {arXiv (Cornell University)},

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}

}

@article{guo2026,

title = {Pathology-guided contrastive pretraining enriches preoperative CT representations for prognosis},

author = {Zhuoxin Guo and Mohamed Harmanani and Paul Wilson and Minh Nguyen Nhat To and Tarek Elghareb and Obed Korshie Dzikunu and Nooshin Maghsoodi and Gábor Fichtinger and Purang Abolmaesumi and Parvin Mousavi},

year  = {2026},

date = {2026-01-01},

pages = {36},

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 factors, allowing practical preoperative application. We evaluate MMCL on the MMIST-ccRCC dataset using cross-validation for predicting 12-month survival status. RESULTS: MMCL achieves an AUROC of 76% using only preoperative CT and clinical factors during inference, surpassing state-of-the-art multi-modal models that leverage both preoperative data and postoperative pathology. Even with CT alone, MMCL matches the performance of state-of-the-art multi-modal baselines, underscoring its robustness. CONCLUSION: Our findings demonstrate that contrastive learning allows CT encoders to extract richer and prognostically relevant features. Importantly, at the testing stage, more accurate and non-invasive survival prediction is achieved using only preoperative CT and general clinical features. This approach provides a promising strategy to improve preoperative prognostic assessment, supporting early treatment planning and guiding personalized cancer care.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@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}

}

@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 Gábor Fichtinger and Alexandra Golby and Parvin Mousavi},

year  = {2026},

date = {2026-01-01},

journal = {International Journal of Computer Assisted Radiology and Surgery},

volume = {21},

number = {7},

pages = {1613-1620},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{Policelli2025,

title = {Prediction of brain metastasis progression after stereotactic radiosurgery: sensitivity to changing the definition of progression},

author = {Robert Policelli and David DeVries and Joanna Laba and Andrew Leung and Terence Tang and Ali Albweady and Ghada Alqaidy and Aaron D. Ward},

doi = {10.1117/1.jmi.12.2.024504},

issn = {2329-4302},

year  = {2025},

date = {2025-03-01},

journal = {J. Med. Imag.},

volume = {12},

number = {02},

publisher = {SPIE-Intl Soc Optical Eng},

keywords = {},

pubstate = {published},

tppubtype = {article}

}