Background Rheumatic heart disease (RHD) is the leading cause of cardiovascular morbidity and mortality in children and young adults worldwide. Early detection via echocardiography (echo) enables secondary antibiotic prophylaxis initiation, preventing further Strep A infections—the primary driver of RHD. Artificial intelligence (AI) with handheld devices and rapid training of non-physician healthcare workers could help scale early RHD diagnosis in low-resource settings. This AI model also has potential in scaling strep vaccine safety monitoring in a cost-effective manner. We previously developed a deep learning algorithm using mitral regurgitation color Doppler on standard portable echo machines, achieving 92% sensitivity and 79% specificity. Here, we present an updated single-view handheld echo AI algorithm for RHD diagnosis based on current guidelines.
Methods Our model was trained on 849 handheld echos and validated on 212 studies. Using the mitral valve-parasternal long-axis color Doppler view after quality control, we applied knowledge transfer from our prior deep learning algorithm developed from standard echo machines. Results were selected based on maximum accuracy for a balanced outcome.
Results The updated model achieved 82% accuracy (range 78–84%), 82% sensitivity (range 77–87%), and 82% specificity (range 78–86%) in detecting RHD on handheld images. Ventricular systole was identified with 94% accuracy, which is consistent across all models and similar to the results obtained with standard data.
Conclusions Our AI models for RHD diagnosis using single-view handheld echos demonstrate strong performance. This technology holds promise for scaling early RHD detection and strep vaccine safety monitoring globally.