Refined Myocardium Segmentation from CT Using a Hybrid-Fusion Transformer

Accurate segmentation of the left ventricle (LV) in cardiac CT images is crucial for assessing ventricular function and diagnosing cardiovascular diseases. Common semi-automatic segmentation often includes unwanted structures, such as papillary muscles, due to low contrast between the LV wall and surrounding tissues. In this study, we address this issue by proposing a two-input-channel method within a Hybrid-Fusion Transformer deep-learning framework. Our method refines coarse LV masks by incorporating both the CT images and the semi-automatic rough masks as input channels, effectively removing papillary muscles. Using a small number of manually refined labels, we evaluated the method through leave-one-out cross-validation. The results demonstrate an average Dice similarity coefficient of 95.2%, outperforming models that use only CT images or rough masks. This approach reduces the need for extensive manual labeling while maintaining high segmentation accuracy and stability, making it suitable for clinical applications.