Kostas Kamnitsas

🤖 Interested in #Out_of_Distribution_detection?
Read about a very interesting model behaviour we found in our new #CVPR2026 work!👇

📣New #ICLR paper 🥳
🤖 "You Point, I Learn: Online Adaptation of Interactive Segmentation Models in Medical Imaging"
#Interactive: Works *with the user*, not replace them
#Adapts: *Learns from user* after each interaction.
Handles #distribution_shifts, eg new MRI sequences

📜 arxiv.org/abs/2503.067...

Sincerely thank you @miccaisociety.bsky.social for this huge honor. I am sure this will be very motivating for Ziyun and the lab to keep pushing forward. Thank you! 🫡 🥰 🤗

IterMask3D: Unsupervised anomaly detection and segmentation with test-time iterative mask refinement in 3D brain MRI Unsupervised anomaly detection and segmentation methods train a model to learn the training distribution as ‘normal’. In the testing phase, they ident…

🤖 The paper introduces, IterMask3D, an unsupervised (no training labels) 3D model for detection & segmentation of unexpected artifacts (quality control/reliability) and brain lesions in brain MRI 🏥

😀So proud for the growth of our students. Thanks team for the hard work🫡

🥳🎊🥂 She won it! She won it! She won it! 🎉🎊🥂
🏆🏅Best Paper Award #MICCAI2025 for our #MedIA article!🏅🏆

One of most prestigious awards in #Medical #Imaging that I could only dream when starting a lab 3 years ago, but @ziyunliang.bsky.social made it!

📜 Paper: www.sciencedirect.com/science/arti...
🧵👇

Really proud of the lab's progress and the growth of our students: Decentralised Isolation Networks (DIsoN) for out-of-distribution detection accepted to #NeurIPS2025 !

📣 New article on MedIA, a fantastic job by Ziyun:

🤖🏥IterMask3D: Unsupervised Anomaly Detection and Segmentation in 3D Brain MRI

-AI that detects problematic image artifacts (eg Quality Control for reliability)
-Advances unsupervised lesion segmentation
-No labels for training

Open Access & code👇

Towards foundational models for Brain Lesion Segmentation with Multi-Modal MRI, vol.2 from our lab:

We show it s feasible to train 1 model with Federated Learning on clients with different brain lesions and MRI modalities, with performance similar to centralised.
Oral @wacvconference.bsky.social