Upcoming MultiMeDIA Presentations at IEEE ISBI 2026

This year's International Symposium on Biomedical Imaging (ISBI 2026) will be held in London, United Kingdom and is one of the world's premier conferences in biological and biomedical imaging. The conference brings thousands of researchers from across the world together to showcase the latest research in mathematical, algorithmic, and computational aspects of biological and biomedical imaging across all scales. We are excited to announce that three papers from members of the MultiMeDIA lab have been accepted for publication and oral presentation at this year's ISBI.

Zhengda's paper, titled 'Multi-class TransCompletion: Geometric Deep Learning for 3D Cardiac Reconstruction from Cine MRI' introduces Multi-class TransCompletion, a geometric deep learning framework for reconstructing four-chamber cardiac point clouds from cine-MRI. Evaluated on the UK Biobank dataset under cross-domain conditions, the method demonstrates substantial improvements in geometric accuracy and anatomical plausibility compared to state-of-the-art approaches, while maintaining strong robustness to slice misalignment and motion artifacts. Zhengda will be presenting their work in the Faster Cardiac MRI, Better Reconstruction session, happening on 9 April, from 16:30-17:30, in Room 4.

Thalia and Alex's paper, titled 'Geometric Deep Learning Model of Left Atrial Geometry and Conduction Velocity in Atrial Fibrillation', extended a geometric deep learning model to capture conduction velocity. In the paper, they use the features learned by the model to explore the non-linear relationship between the geometry and electrophysiology of the left atrium in atrial fibrillation. Thalia will be presenting their work in the New Modalities, New Measurements session, happening on 10 April, from 08:00-09:00, in Room 3. 

Mohammad Atwany and Mojtaba's paper is titled 'Feature Invariance via Interpretable Ablation for Single-Source Domain Generalisation in X-Ray Angiography Segmentation'. This paper explores how the behaviour of low-level feature channels in the first layer, where basic visual patterns are first captured, is linked to domain generalisation in X-ray coronary angiography vessel segmentation. More specifically, their analysis suggests that generalisation may depend not only on the coverage of domain-invariant channels, but also on their relative contribution. Atwany will be presenting their work in the X-Rays with Fewer Blind Spots session, happening on 10 April, from 16:30-17:30, in Room 15.

We congratulate Zhengda, Thalia, Alex, Atwany, and Mojtaba on these achievements, and wish them the best of luck with their upcoming oral presentations! The supporting papers to their presentations will be released in the upcoming IEEE conference proceedings.