Explainable AI Model Diagnoses Bicuspid Aortic Valve from Echocardiography
Researchers developed an explainable AI model to distinguish bicuspid aortic valve (BAV) from tricuspid aortic valve (TAV) using transthoracic echocardiography (TTE) parasternal long-axis (PLAX) cine loops. The model uses a multi-backbone video ensemble trained on 90 patient studies (48 BAV, 42 TAV) with a leakage-aware, stratified outer cross-validation protocol. The calibrated stacked ensemble achieved an outer-CV F1-score of 0.907 and recall of 0.877. Frame-level Grad-CAM localized evidence to the aortic root and leaflet plane, while SHAP values quantified each backbone's contribution. The study, published on arXiv (2605.13730), suggests PLAX-based AI can improve diagnostic consistency.
Key facts
- Model uses multi-backbone video ensemble on PLAX cine loops
- Trained on 90 patient studies (48 BAV, 42 TAV)
- Leakage-aware, stratified outer cross-validation protocol used
- Calibrated stacked ensemble achieved F1-score 0.907
- Recall of 0.877 achieved
- Grad-CAM localized evidence to aortic root and leaflet plane
- SHAP values quantified video backbone contributions
- Published on arXiv with ID 2605.13730
Entities
Institutions
- arXiv