An enhanced hierarchical Mamba-based cardiac segmentation network for cardiovascular analysis and disease diagnosis using multiple MRI datasets

Automatic cardiac segmentation grew in popularity as it became feasible in clinical settings. Deep learning-based techniques are used throughout the world to gain access to cardiac functioning. Although the current deep learning architecture represents an important step toward clinical usability, it falls short in terms of accurately and effectively segmenting critical cardiac segments such as the left ventricle (LV), right ventricle (RV), and myocardium. Furthermore, the generalization of approaches is a significant difficulty when doing cardiac segmentation. Therefore, in this article we proposed the Mamba-based algorithm, which we named as H-MayoMamba Net, that uses hierarchical structure to undergo local processing, regional processing, global processing, and context processing in each level, one by one. The experiment reveals that our method has been evaluated by following performance evaluation metrics like dice similarity coefficient, Hausdorff distance (HD) with the record of as 95.7 ± 0.1, 95.7 ± 1.3, 92.1 ± 1.5, 93.61 ± 0.4, and 6.01 ± 0.8 for LV dice(%), RV dice(%), Myo dice (%), Avg dice (%), and HD (mm), respectively. The proposed H-MayoMamba Net confirms the cross-data generalization by testing on different datasets, like training on ACDC and testing on M&M dataset, and vice versa. Thus, the result shows that the proposed method has potential of applying the automatic cardiac segmentation using AI and proved to be a useful starting point for realizing the concept of digital cardiology.