MedVIC Lab
Medical Vision and Intelligent Computing
We build structured AI systems that make the geometry, variability, and uncertainty embedded in medical images computationally tractable — and clinically actionable.
Skip to content
Research Themes
Three interconnected pillars that define our approach to medical AI
Modeling Patient State
Geometry-aware representations of anatomy: how it is shaped, how it varies across populations, how it changes over time, and how it manifests disease. Anatomy is not just geometry — it is a structured signal we can model.
ShapeWorks · CranioRate · Point2SSM →Learning Under Clinical Constraints
AI systems that learn reliably from weak, sparse, heterogeneous, and domain-shifted clinical data without requiring dense expert annotation. Clinical constraints define how we design representations.
Multi-instance learning · Domain adaptation →Clinically Trustworthy AI
Embedding uncertainty, interpretability, and semantic grounding into AI systems so that clinicians know what models know, what they don't, and why. Trust is not accuracy — it is knowing what the model knows.
VIB · Uncertainty · Interpretability →Flagship Software
ShapeWorks
Open-source platform for automated construction of statistical shape models of anatomical structures. The complete analysis pipeline: preprocessing, correspondence optimization, shape representation, statistical analysis, and visualization.
8,000+ downloads100+ institutions163+ citing studies14 major releases
CranioRate
AI severity scoring for craniosynostosis. Converts a patient's 3D CT scan into an objective, population-grounded severity score — replacing subjective clinical assessment with reproducible, data-driven measurement.
2,900+ scans processed7 clinical sitesPrivacy by design
Latest News
June 2026
PaperPoint2SSM++ Published in Medical Image Analysis
Point2SSM++, a self-supervised framework for learning anatomical shape models directly from point clouds, has been accepted and published in Medical Image Analysis. The work extends Point2SSM with a self-supervised training paradigm that eliminates the need for ground-truth correspondences.
May 2026
PaperSix Papers Under Review
MedVIC has six papers under review at top venues, spanning cross-modal representation learning, multi-scale pathology supervision, MRI quality assessment, interpretability in medical vision-language models, normalizing flows for morphology, and accelerated 3D LGE-MRI reconstruction.
Spring 2026
PeopleTuan Le Joins MedVIC
Tuan Le has joined MedVIC as a PhD student, working on few-shot learning in a co-advising arrangement with Prof. Gianfranco Doretto.
Interested in working with MedVIC?
We recruit PhD students through the Kahlert School of Computing and welcome postdoctoral researchers, visiting scholars, and clinical collaborators.
Join the LabGet in Touch