Deep learning for coronary arteries identification and measurement in pediatric echocardiography

Abstract Background Coronary artery lesions are a critical complication of Kawasaki disease. This study developed a deep learning model for the automated identification and measurement of coronary arteries in pediatric echocardiograms to facilitate accurate assessment. Methods We collected 2314 echocardiograms from children with suspected or diagnosed Kawasaki disease. A YOLOv8 model was trained on a curated set of 3000 images to identify the right coronary artery, left main coronary artery, left anterior descending artery, and left circumflex artery. 1500 standard images were selected to validate the model’s measurement performance, and compared against physicians. Results The AI model showed high identification performance on the test set, with the mean average precision for all coronary branches >80%. The identification rates were higher in toddlers and older children than in infants and in experts than in general physicians. The AI model demonstrated excellent agreement with expert physicians across all branches, with significantly shorter measurement time than manual methods. Conclusion The AI model developed in this study demonstrated high accuracy in pediatric coronary artery identification and measurement, showing promise as an efficient and reliable assistive technology. Impact A deep learning model was developed for automated identification and measurement of coronary arteries in pediatric echocardiography. It moves beyond existing identification-focused studies by solving the key challenge of automated, precise quantification. This AI model can serve as a valuable supplementary tool for standardizing echocardiographic evaluation and supporting quality control in clinical practice.