Pinned Tweet
🔥 The 12th CVPR Workshop on Medical Computer Vision.
11 keynotes. One day. The entire frontier of medical AI in a single room.
📍 June 3, 2026 · Room 605 · Colorado Convention Center, Denver
🎙️ Session I (9:15 am):
→ Dimitris Metaxas (@RutgersU)
→ Daguang Xu (@nvidia)
→ Sharon X. Huang (@penn_state)
→ Mathias Unberath (@JohnsHopkins)
→ Archana Venkataraman (@BU_Tweets)
🎙️ Session II (1:00 pm):
→ Maddie Traverse (@GoogleResearch)
→ Hoifung Poon (@MSFTResearch)
→ Jeremias Sulam (@JohnsHopkins)
→ Kayhan Batmanghelich (@BU_Tweets)
→ Ehsan Adeli (@Stanford)
→ Mert Sabuncu (@Cornell)
If you're at #CVPR2026, you cannot miss this.
#MedicalAI #ComputerVision #FoundationModels #HealthcareAI #MedicalImaging
At its center is BodyMaps: an open medical imaging library, built at @JohnsHopkins, that anyone in the world can use.
🧭 Explore 32,768 CT volumes from 145 centers
🩻 Upload your own scan and run our AI models on it
📝 Turn scans into plain-language reports patients can actually read
Try it: bodymaps.wse.jhu.edu/
Our hope is that BodyMaps becomes for cancer what the JHU COVID-19 dashboard was for the pandemic.
We're LIVE. 🏔️
The 12th Medical Computer Vision Workshop just kicked off at #CVPR2026 — Room 605, Denver, right now.
11 keynotes. All day. The full frontier of medical AI.
Starting in minutes: Dimitris Metaxas (@RutgersU), CVPR 2026 General Chair himself, opens the floor.
If you're in the building — get to 605. Now.
@MathiasUnberath @ArchanaVen @hoifungpoon @Jere_je_je @kbatmang @eadeli @mertrory · Dimitris Metaxas · Daguang Xu · Sharon X. Huang · Maddie Traverse · Rene Vidal
#MedicalAI #MedicalComputerVision #HealthcareAI #ComputerVision @CVPR
That's a wrap on Day 1 of MCV 2026 @JHU_BodyMaps 🎉
From physics-based deformable models → open-source MONAI infrastructure
→ ambient intelligence in senior homes → mathematical foundations of
trustworthy AI —
Medical computer vision has never been this broad, or this urgent.
Thanks to all 12 speakers. The clinic is watching. 👁️🏥
The 13th workshop on medical computer vision will be back next year in Seattle.🔬
#CVPR2026 #MedCV #BodyMaps @CVPRConf @CVPR
12/ Closing keynote: Prof. René Vidal (@Penn), Rachleff PIK University Professor, IEEE/ACM/AIMBE Fellow, Director of IDEAS Center.
Talk: "Trustworthy AI in Health: Foundation Models for Radiology, Cardiology and Autism Diagnosis"
Currently building a foundation model for cardiology (in collaboration with 2026 Lifetime Achievement Award winner Julio Chirinos). Leads THEORINET — the NSF-Simons collaboration on the mathematical foundations of deep learning.
The math has to be right before the medicine can be safe. ✅🔬
#CVPR2026 #TrustworthyAI #HealthAI @CVPR
11/ Prof. Mert Sabuncu (@Cornell + Weill Cornell Medicine), VP of AI & Engineering Research in Radiology.
Talk: "Prevalence Adjustment as a Way to Handle Distribution Shift in Medical Vision"
His CoPA/GPA framework explicitly models population-level label shifts across hospital sites — rather than suppressing them. VoxelMorph (CVPR) for deformable registration is a landmark in the field.
You can't just train on one hospital and hope. He's building the math to bridge them. 🏥➡️🏥
#CVPR2026 #DistributionShift #MedicalAI @Cornell & @cornell_tech @CVPR
Coming up next at #CVPR2026 MCV Workshop — Prof. Mert Sabuncu @mertrory
(@Cornell + Weill Cornell Medicine, Vice Chair of AI & Engineering Research in Radiology)
Talk: "Prevalence Adjustment as a Way to Handle Distribution Shift in Medical Vision"
Here's the brutal truth no one wants to say out loud:
Your model trained at Hospital A will quietly fail at Hospital B.
Not because the architecture is wrong.
Not because you didn't train long enough.
Because the *disease prevalence* changed — and your model never knew.
Mert's work asks the question every clinical AI team is avoiding:
What if we stopped pretending our training distribution is the deployment distribution?
His lab builds the math to bridge hospitals, scanners, and populations — without retraining from scratch.
This is the deployment science clinical AI desperately needs.
Room 605 | 3:45pm sharp 👇
#MedicalAI #DistributionShift #TrustworthyAI #BodyMaps @CVPR
10/ Prof. Ehsan Adeli (@Stanford Psychiatry + CS), Director of STAI Lab.
Talk: "From Bedside to Living Room: Reimagining Care Through Ambient Intelligence"
His lab deploys tiny cameras in senior homes to detect early signals of cognitive decline — catching Alzheimer's before a clinical visit ever would. Computer vision that lives quietly in the background of everyday life.
The clinic can't be everywhere. Ambient AI can. 🏠🧠
#CVPR2026 #AmbientAI #CognitiveCare @CVPR @Stanford
9/ Prof. Kayhan Batmanghelich (@BostonUniversity), NSF CAREER & Google Faculty Research Award.
Talk: "Image-Text Foundations for Volumetric Images: Challenges and Opportunities"
His lab's latest work (arXiv 2604.10437): fine-grained spatial grounding in 3D CT report generation via discriminative guidance — making text reports spatially precise, not just descriptive.
Volumetric grounding is the next frontier for radiology AI. 📐🫁
#CVPR2026 #3DImaging #RadiologyAI @CVPR
8/ Prof. Jeremias Sulam (@JHUBME @JohnsHopkins), William R. Brody Faculty Scholar.
Talk: "Flexible Methods for Uncertainty Quantification in Medical Imaging"
His work bridges sparse representations, conformal prediction, and learned proximal networks — giving clinicians not just a diagnosis, but a confidence interval they can trust.
A radiologist shouldn't just know *what* the model says. They should know *how sure* it is. 📏🔬
#CVPR2026 #UncertaintyQuantification #TrustworthyAI @CVPR
7/ Hoifung Poon, GM of Health AI @MSFTResearch and affiliate faculty @UW Medicine.
Talk: "Learning the Language of Patients: Multimodal Generative AI for Precision Health"
His lab structures the chaos of clinical data — imaging, notes, labs, genomics — into machine-readable representations for precision oncology and drug discovery.
The patient record is the world's hardest language model problem. He's solving it. 💊📊
#CVPR2026 #PrecisionHealth #ClinicalAI @CVPR @Microsoft
6/ Maddie Traverse (@Google Research) — key contributor to Google's CT Foundation Model and lead ML engineer on MedGemma.
Talk: "MedGemma: An Open Vision-Language Model for Diverse Medical Applications"
MedGemma 1.5 (arXiv 2604.05081) benchmarks across radiology (2D, longitudinal, 3D), dermatology, histopathology, ophthalmology, and multimodal clinical reasoning — in an open, fine-tunable package.
Open-source medical VLM infrastructure is finally here. 🧬🤖
#CVPR2026 #MedGemma #MedicalVLM @CVPR @Google
Afternoon session just kicked off at the 12th CVPR Workshop on Medical Computer Vision 🏥
Up first: Maddie Traverse (@Google Research) presenting MedGemma — Google's open medical vision-language foundation model.
MedGemma 1.5 is the first open VLM that natively handles:
→ Full 3D CT & MRI volumes (not just slices)
→ Gigapixel whole-slide pathology images
→ Anatomical bbox localization on X-rays
→ Multi-timepoint longitudinal radiology
→ Lab reports + EHR document understanding
All in a single 4B-parameter architecture you can run offline.
+11% on 3D MRI classification. +47% macro F1 on whole-slide pathology.
Open weights. Free for research and commercial use.
The open-source infrastructure for clinical AI is finally here. 🔬
#CVPR2026 #MedGemma #MedicalAI #BodyMaps @CVPR
5/ Prof. Archana Venkataraman (@BostonUniversity), NSF CAREER awardee in ML for brain disorders.
Talk: "Lightweight and Interpretable AI as a New Window into Brain Dysfunction"
She builds data-driven methods for brain connectivity, fusing multimodal biomedical data to understand mental health — models that are small enough to explain and powerful enough to matter.
Interpretability isn't a constraint. It's the scientific instrument. 🧠
#CVPR2026 #Neuroscience #XAI @CVPR
4/ Prof. Mathias Unberath (@JohnsHopkins), John C. Malone Associate Professor & Director of the ARCADE Lab.
Talk: "Digital Twins for Ambient and Embodied Surgical AI"
NSF CAREER + NIH Trailblazer awardee. His lab sits at the intersection of CV, robotics, AR, and clinical decision-making.
Digital twins that model the OR in real time — surgical AI that doesn't just observe, but anticipates. 🤖🏥
#CVPR2026 #SurgicalAI #DigitalTwins @CVPR
3/ Prof. Sharon X. Huang (@PennState), David Reese Professor and Department Head at Penn State IST.
Talk: "Advancing Diagnostic Robustness and Privacy-Preserving Model Training with Medical Image and Video Generation"
200+ publications. Her group develops generative approaches for medical images AND video — addressing data scarcity while never leaking patient identity.
Privacy-preserving generation is not a trade-off. It's the only path to scale. 🔐
#CVPR2026 #MedicalImaging #Privacy @CVPR
2/ Next: Daguang Xu (@NVIDIA) — PhD from Johns Hopkins, now leading healthcare AI at NVIDIA Research.
His talk: "Driving AI Innovation in Healthcare through Open Data and Foundation Models"
He built MONAI (the open-source framework powering most medical AI pipelines worldwide) and MAISI — a diffusion model that synthesizes high-resolution 3D CT volumes to fight data scarcity.
Open data + open infrastructure = the picks-and-shovels of clinical AI. 🏗️
#CVPR2026 #MONAI #MedAI @nvidia @CVPR
1/ Kicking off: Prof. Dimitris Metaxas (@Rutgers), CVPR 2026 General Chair and one of the field's most prolific builders — 800+ papers, 80 PhD students.
His talk: "Explainability, Generation, Physics and Dynamics in ML for Biomedical Applications"
His lab fuses physics-based deformable models with generative AI for organ shape estimation from MRI/CT, cardiac analytics, and cancer histopathology.
The rare researcher who makes the physics of tissue talk to the math of deep learning. 🧬
#CVPR2026 #MedCV @CVPR
Closing out #CVPR2026 MCV Workshop with a heavy hitter. 🎙️
Rene Vidal — Rachleff University Professor at Penn, Amazon Scholar, Affiliated Chief Scientist at NORCE — presents:
"Trustworthy AI in Health: Foundation Models for Radiology, Cardiology and Autism Diagnosis"
Radiology. Cardiology. Autism. Three of the hardest clinical domains in medical AI, one talk.
This is what trustworthy AI actually looks like in practice.
Room 605, June 3. Don't leave early. ⚡
@Penn_Engineering #TrustworthyAI #FoundationModels #MedicalAI #CVPR2026 @CVPR
Today's the day. 🎙️
We're kicking off #CVPR2026 with the man himself — Dimitris Metaxas, Rutgers Distinguished Professor and CVPR 2026 General Chair, opening the 12th Medical Computer Vision Workshop at 9:15 AM sharp.
Talk: "Explainability, Generation, Physics and Dynamics in ML for Biomedical Applications"
When the General Chair of the entire conference chooses to open your workshop — that tells you everything about where medical computer vision stands right now.
Room 605. Denver. See you there. 🏔️
@RutgersU @CVPR #MedicalComputerVision #CVPR2026
📊 Real-world medical AI fails when distributions shift. Here's a rigorous solution.
Mert Sabuncu (@Cornell & @cornell_tech, Vice Chair AI @ Weill Cornell Medicine, @NSF CAREER + @NIH Early Career Awards) closes @CVPR Medical Computer Vision Workshop:
"Prevalence Adjustment as a Way to Handle Distribution Shift in Medical Vision"
From Princeton → MIT CSAIL → Harvard/MGH → Cornell. A career of solving the hardest problems in biomedical imaging.
#MedicalAI #DistributionShift #MedicalImaging #CVPR2026
