We are pleased to announce that the following paper, co-authored by Weiwei Ma, Xiaobing Yu, Peijie Qiu, Jin Yang, Pan Xiao, Xiaoqi Zhao, Xiaofeng Liu, Tomo Miyazaki, Shinichiro Omachi, and Yongsong Huang (Assistant Professor at the AI So-Go-Chi center), has been published in the proceedings of the IEEE International Symposium on Biomedical Imaging (ISBI 2026). This premier conference is a key forum for the dissemination of cutting-edge research in biomedical imaging and image analysis.

Published Paper

Weiwei Ma, Xiaobing Yu, Peijie Qiu, Jin Yang, Pan Xiao, Xiaoqi Zhao, Xiaofeng Liu, Tomo Miyazaki, Shinichiro Omachi, Yongsong Huang. “U-Harmony: Enhancing Joint Training for Segmentation Models with Universal Harmonization.” In Proceedings of the IEEE International Symposium on Biomedical Imaging (ISBI), 2026.

This work addresses the challenge of training robust segmentation models from limited and heterogeneous medical datasets, which exhibit significant variations in imaging modalities, protocols, and annotation standards across institutions. To overcome the performance degradation caused by such domain shifts in conventional joint training, we propose the Universal Harmonization (U-Harmony) module. This framework introduces a two-stage feature harmonization and restoration mechanism that first normalizes instance-specific feature distributions and then selectively restores domain-specific information, enabling a single model to learn effectively from multiple heterogeneous sources while preserving dataset-specific knowledge. Integrated with a domain-gated head for dataset-agnostic inference, the proposed method demonstrates superior performance in extensive experiments on cross-institutional brain lesion segmentation datasets, establishing a new benchmark for generalizable and adaptable 3D medical image segmentation.

Congratulations to the author!