News - AISSLab

AISSLab members and collaborators recently concluded an impactful week at the 29th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI 2026), held at the Palais de la Musique et des Congrès in Strasbourg, France. As one of the premier global venues for medical imaging and artificial intelligence, this year’s conference featured 1,167 accepted papers out of 4,601 worldwide submissions, with South Korea contributing strongly through over 300 registered delegates.

Representing our active international research initiatives between Sejong University, ITMO University, and the Almazov National Medical Research Centre, the team presented two groundbreaking research contributions advancing modern MRI analysis.

Featured Research Presentations

Towards Automated Cardiac MRI Assessment: A Multi-Agent CAD Framework for Functional Analysis and Tissue Characterization

Moving beyond single-task pipelines, this study introduces MRI-CARDAIX, an autonomous multi-agent clinical decision-support framework. The architecture bridges Cine and LGE MRI pipelines, unifying cardiac functional parameter extraction, AHA 17-segment fibrosis localization, and RAG-driven clinical diagnostic reports into an integrated, trustworthy workflow. Co-authors: Walid Al-Haidri, Mukhlis Raza, Anatoly Levchuk, Kseniya Belousova, Maksim Lukin, Yeong Hyeon Gu, Mugahed A. Al-antari, and Ekaterina Brui. (Link)

Prof. Mugahed A. Al-antari (left) and Dr. Walid Al-Haidri (right) presenting the multi-agent cardiac MRI framework (MRI-CARDAIX) during the interactive poster session at MICCAI 2026.

Sequence-Conditioned Flow-based Models for Digital Phantom Generation in MRI

Addressing clinical data scarcity and privacy bottlenecks, this research proposes a self-supervised, physics-informed generative framework. By conditioning continuous normalizing flows directly on pulse sequence parameters (TR, TE, ESP), the framework synthesizes high-fidelity digital phantoms (T1, T2, and PD maps) from conventional MR acquisitions without requiring paired ground truth.