Date: August 3, 2025

Source: Jinan University Integrated Media Center

Editor: Chang Kaili

Lead: A research team led by Associate Professor Bai Jieyun and Professor Zhang Xiaoshen from Jinan University, in collaboration with international institutions, has published groundbreaking research in the top medical imaging journal IEEE Transactions on Medical Imaging. The study successfully developed the first deep learning-based framework for automated right atrial segmentation, providing a crucial tool for precise cardiac disease diagnosis and treatment.

The right atrium plays a crucial role in cardiac function yet has long been underemphasized in clinical diagnosis. With advances in interventional cardiology, precise assessment of its structure and function has become increasingly important. Recently, Bai Jieyun and Zhang Xiaoshen's team at Jinan University, collaborating with researchers from the University of Auckland, University of Manchester, and other institutions, published their study titled A Benchmark Framework for the Right Atrium Cavity Segmentation From LGE-MRIs in IEEE Transactions on Medical Imaging.

To address challenges in right atrial segmentation such as class imbalance and significant anatomical variations, the research team developed a two-stage segmentation framework based on the 3D deep learning network RASnet. This architecture integrates convolutional neural networks with vision transformers, employing multi-path input, multi-scale feature fusion, global context capture, and edge optimization modules to achieve high-precision automated segmentation of the right atrium in delayed gadolinium-enhanced magnetic resonance imaging.

Experiments were conducted on a dataset containing 354 LGE-MRI cases. Results demonstrated that RASnet achieved a Dice coefficient of 92.19%, Jaccard coefficient of 0.8563, and Hausdorff distance of 13.19 on the primary dataset, outperforming mainstream models like nnUnet and MedSAM. The framework also showed excellent performance on independent validation datasets, confirming its strong generalization capability.

The study's breakthrough lies in establishing the first benchmark framework for right atrial segmentation in LGE-MRI, addressing the long-standing lack of standardized methods and public datasets in this field. RASnet accurately captures the complex anatomical structures of the right atrium, providing clinicians with quantitative tools for measuring cavity size, morphology, and volume. The team has made their code and dataset publicly available to promote standardization and clinical translation in the field.

Example image

In clinical practice, this framework can assist in evaluating right atrial remodeling in atrial fibrillation patients and predicting outcomes in pulmonary hypertension—for instance, by quantifying right atrial volume changes to enable earlier prediction of tricuspid regurgitation progression and optimize surgical timing. Automated segmentation also significantly reduces manual annotation time, supporting efficient large-scale clinical studies and follow-up monitoring.

The corresponding authors of this study are Associate Professor Bai Jieyun and Director Lu Hua from Jinan University. The research received support from the National Natural Science Foundation of China, Guangdong Basic and Applied Basic Research Foundation, European Research Council, and other funding agencies.

Paper link: https://doi.org/10.1109/TMI.2025.3590694

Open-source code: https://github.com/zjinw/RAS

Dataset: https://zenodo.org/records/15524472