Probabilistic Vision Group


The Probabilistic Vision Group (PVG) and Medical Imaging Labs are led by Prof. Tal Arbel, CIFAR AI Chair, MILA, and are located within the centre for Intelligent machines, Department of Electrical & Computer Engineering, McGill University. The research group lies at the juxtaposition of the fields of computer vision, machine learning and medical image analysis. Established in 2001, the PVG is an internationally-recognized, interdisciplinary research lab focused on developing probabilistic machine learning frameworks in computer vision developed for a wide range of real-world applications in neurology and neurosurgery. Recent work is focused on the development of modern deep learning models for inference in medical image analysis in the presence of pathological structures (e.g. lesions, tumours), including: estimating and propagating uncertainties, knowledge distillation, interpretability/explainability, domain adaptation, learning and adapting to cohort biases in aggregated datasets, self-supervision, active learning, multi-modal predictions (e.g. merging clinical and imaging), etc. for problems ranging from segmentation, detection and probabilistic lesion count estimation to precision medicine based on patient brain images.

In addition to theoretical advances, the outcome of this research has already led to concrete improvements in patient care. For example, probabilistic graphical machine learning algorithms developed by her team for Multiple Sclerosis (MS) lesion detection and segmentation have been used in the clinical trial analysis of most of the new multiple sclerosis drugs currently used worldwide. These advances have been made possible through strong collaborations with medical imaging, machine learning and computer vision researchers worldwide, as well as clinicians and several industrial partners.

News

Gian Favero presented his latest work entitled “Conditional Diffusion Models are Medical Image Classifiers that Provide Explainability and Uncertainty for Free” at the 8th Conference on Medical Imaging with Deep Learning (MIDL 2025) held in Salt Lake City, USA.[07/10/2025]

Amar Kumar presented his latest work entitled “PRISM: High-Resolution & Precise Counterfactual Medical Image Generation using Language-guided Stable Diffusion” at the 8th Conference on Medical Imaging with Deep Learning (MIDL 2025) held in Salt Lake City, USA.[07/09/2025]

Professor Tal Arbel presented her keynote talk at the 8th Conference on Medical Imaging with Deep Learning (MIDL 2025) held in Salt Lake City, USA.[07/09/2025]

Amar Kumar presented his latest work entitled “Language-Guided Trajectory Traversal in Disentangled Stable Diffusion Latent Space for Factorized Medical Image Generation” at the Mechanistic Interpretability for Vision (MIV) Workshop at CVPR 2025 held in Nashville, USA.[06/12/2025]

Anita Kriz presented her latest work entitled “Leveraging Vision-Language Foundation Models to Reveal Hidden Image-Attribute Relationships in Medical Imaging” at the Mechanistic Interpretability for Vision (MIV) Workshop at CVPR 2025 held in Nashville, USA.[06/12/2025]

Parham Saremi presented his latest work in the Montreal Medical Imaging Workshop 2025 at Polytechnique Montreal. [05/02/2025]

Professor Tal Arbel presented a tech talk on “Building Trustworthy Medical Imaging AI Systems for Safe Clinical Deployment" at the 7th World Summit AI (WSAI 2025), held in Montreal, Canada. [04/16/2025]

Congratulations to Dr. Joshua Durso-Finley for completing his PhD.[03/14/2025]

Gallery