Application of Artificial Intelligence-Based Auxiliary Diagnosis in Congenital Heart Disease Screening

Abstract

Background: To evaluate the application value of artificial intelligence-based auxiliary diagnosis for congenital heart disease. Methods: From May 2017 to December 2019, 1892 cases of congenital heart disease heart sounds were collected for learning- and memory-assisted diagnosis. The diagnosis rate and classification recognition were verified in 326 congenital heart disease cases. Auscultation and artificial intelligence-assisted diagnosis were used in 518 258 congenital heart disease screenings, and the detection accuracies of congenital heart disease and pulmonary hypertension were compared. Results: Female sex and age > 14 years were predominant in atrial septal defect (P < .001). Compared with no pulmonary arterial hypertension, a male predominance was seen in cases of congenital heart disease–pulmonary arterial hypertension (P < .001), and age was significantly associated with pulmonary arterial hypertension (P= .008). A high prevalence of extracardiac anomalies was found in the pulmonary arterial hypertension group. A total of 326 patients were examined by artificial intelligence. The detection rate of atrial septal defect was 73.8%, which was different from that of auscultation (P= .008). The detection rate of ventricular septal defect was 78.8, and the detection rate of patent ductus arteriosus was 88.9%. A total of 518 258 people from 82 towns and 1220 schools were screened including 15 453 suspected and 3930 (7.58%) confirmed cases. The detection accuracy of artificial intelligence in ventricular septal defect (P= .007) and patent ductus arteriosus (P= .021) classification was higher than that of auscultation. For normal cases, the recurrent neural network had a high accuracy of 97.77% in congenital heart disease–pulmonary arterial hypertension diagnosis (P= .032). Conclusion: Artificial intelligence-based diagnosis is an effective assistance method for congenital heart disease screening.