Developed ST-Net, an end-to-end framework for 4D cardiac MRI dynamic segmentation that introduces a dual-path spatiotemporal modeling strategy. The network integrates spatial morphology extraction, temporal attention recalibration, and spatio-temporal smoothness regularization to achieve both high spatial accuracy and temporal consistency across the cardiac cycle.

Built upon 3D TransUNet and CA-Net architectures, trained on the ACDC and CAMUS benchmarks with multi-round training pipelines.