Automated valvular heart disease detection using heart sound with a deep learning algorithm.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Clinical auscultation by clinicians has limitations in diagnosing valvular heart disease (VHD).
• Artificial intelligence (AI) offers a potential solution to enhance VHD diagnosis by analyzing heart sounds.
• The efficacy of AI in automatically diagnosing VHD requires further investigation.

Purpose of the Study:

• To develop and evaluate a deep learning model for identifying patients with VHD requiring intervention using raw heart sound data.
• To compare the diagnostic performance of the AI model against established clinical standards.

Main Methods:

• Collected heart sound data from patients with VHD and healthy controls using an electronic stethoscope.
• Utilized echocardiography as the gold standard for VHD diagnosis.
• Trained a deep learning model on early-enrolled data and validated it on late-enrolled data.

Main Results:

• The study included 499 patients (354 for training, 145 for validation).
• The deep learning model demonstrated high sensitivity, specificity, and accuracy for identifying various VHDs (71.4-100.0%).
• Mitral stenosis showed the best diagnostic performance with 100% sensitivity, specificity, and accuracy.

Conclusions:

• Deep learning models can effectively identify patients with VHD from raw heart sound data.
• AI-assisted VHD diagnosis shows promise for improving screening, diagnosis, and follow-up processes.
• This technology can potentially compensate for limitations in human auscultation skills.