Structure Aware Diffusion Models for Enhanced Cardiac Motion Estimation in Echocardiography

Echocardiography is an important tool for evaluating cardiac function, where accurate characterization of myocardial motion is crucial for the diagnosis and treatment of heart disease. However, the inherent speckle noise and low spatial resolution property of ultrasound images restrict the calculation accuracy of myocardial motion. To this end, we propose an innovative multi-task learning framework that combines a myocardial segmentation network to provide reliable tissue semantic information for optical flow estimation. Specifically, we extract motion and context features based on the RAFT network, and we innovatively integrate the generation capability of the diffusion model to improve the quality of myocardial motion estimation. In addition, we introduce the myocardial tissue probability map generated by the segmentation network as a spatial structure prior to further improve the estimation accuracy of the motion boundary. The experimental results show that the proposed method significantly improves the estimation accuracy of the myocardial motion, providing a new solution for echocardiography analysis.Clinical relevance - This study presents a novel structure-aware diffusion model for accurate myocardial motion generation in echocardiography, thereby enhancing clinical cardiac assessment and supporting improved diagnosis and management of heart disease.