Deep Self-Integrated 3D Left Atrium Segmentation Network Based on Semi-Supervised Learning

Pixel-accurate tissue segmentation of the left atrium (LA) can assist clinicians in diagnosing atrial fibrillation and assessing cardiac function. However, existing LA segmentation methods suffer from poor performance in complex anatomical structures due to insufficient utilization of structural information. To address this issue, we propose a Deep self-integrated Uncertainty-aware Mean Teacher network (DU-MT), which comprises a Multilevel deep melting furnace module (MDFM) and a Bifocal V-Net (BV-Net) backbone. Specifically, MDFM fuses uncertainty maps from multiple decoder levels using an entropy-weighted strategy to dynamically adjust the consistency loss. BV-Net enhances boundary continuity through spatial-channel dual-branch attention. In addition, a Vision Transformer (ViT) is integrated into the bottleneck layer of the V-Net to model long-range dependencies. Extensive experiments on the MICCAI 2018 LA Dataset demonstrate that our approach achieves better or comparable performance to state-of-the-art methods, making it suitable for real-world clinical applications.