Mitral regurgitation (MR) is the most prevalent valvular heart disease in China. Transcatheter edge-to-edge repair (TEER) is currently the preferred treatment for patients with severe MR who face high surgical risks. However, existing preoperative assessment methods for TEER suffer from numerous limitations, including complex measurement parameters, high technical demands, and significant subjectivity. Vision Mamba, a cutting-edge technology in the visual domain, overcomes the limitations of common computational units in convolutional neural networks and Transformers through bidirectional state space models and positional encoding, demonstrating exceptional performance in visual tasks. To date, no studies have applied Vision Mamba to ultrasound videos for constructing TEER preoperative assessment models. Our team previously established a Transformer-based evaluation model using a small, single-center cohort. This study innovatively introduces a Vision Mamba-based visual foundation model. By integrating multi-faceted, multi-modal ultrasound videos from multiple centers, we develop a one-stop preoperative TEER assessment model for MR patients \[slice identification → video analysis → multi-modal information fusion → preoperative assessment recommendation (suitable/challenging/unsuitable)\]. This model will optimize surgical patient identification and accurately screen patients with contraindications. Furthermore, the one-stop model is fast and objective, significantly improving clinical efficiency. It holds promise for deployment at primary care levels to optimize healthcare resource allocation.
Age: 18-90 years
Patients with moderate to severe or severe mitral regurgitation as indicated by echocardiography
Moderate or severe aortic stenosis or aortic regurgitation, or following aortic valve replacement
Patients with congenital heart disease
Poor image quality: Insufficient cross-sectional coverage or inability to perform effective mitral valve marking