Machine Learning for Pediatric Echocardiographic Mitral Regurgitation Detection

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Echocardiography screening for pediatric valvular disease aids early diagnosis but is resource-intensive.
• Automated echocardiographic diagnosis could expand screening feasibility and improve outcomes, especially in resource-limited settings.
• This study explored building a machine learning model for identifying mitral regurgitation (MR) on echocardiography.

Purpose of the Study:

• To develop and assess a machine learning model for detecting mitral regurgitation (MR) in pediatric echocardiograms.
• To enable automated, widespread echocardiographic screening for valvular heart disease.

Main Methods:

• Two convolutional neural networks were developed: one for view classification and one for MR detection.
• Models were trained and evaluated using labeled echocardiogram data, including parasternal long-axis color Doppler views.
• Performance metrics included accuracy, precision, recall, F1 score, and receiver operating characteristic analysis.

Main Results:

• The view classification model achieved an F1 score of 0.97.
• The MR detection model demonstrated a testing accuracy of 0.86 and an area under the receiver operating characteristic curve of 0.91.
• These results indicate the model's capability in discerning MR.

Conclusions:

• A machine learning model can effectively identify mitral regurgitation on transthoracic echocardiography.
• This represents a significant advancement towards AI-driven diagnosis of pediatric valvular heart disease.
• Automated detection holds promise for more accessible and timely diagnosis.