Automatic pediatric congenital heart disease classification based on heart sound signal

Congenital heart diseases (CHD) are the most common birth defects, and the early diagnosis of CHD is crucial for CHD therapy. However, there are relatively few studies on intelligent auscultation for pediatric CHD, due to the fact that effective cooperation of the patient is required for the acquisition of useable heart sounds by electronic stethoscopes, yet the quality of heart sounds in pediatric is poor compared to adults due to the factors such as crying and breath sounds. This paper presents a novel pediatric CHD intelligent auscultation method based on electronic stethoscope. Firstly, a pediatric CHD heart sound database with a total of 941 PCG signal is established. Then a segment-based heart sound segmentation algorithm is proposed, which is based on PCG segment to achieve the segmentation of cardiac cycles, and therefore can reduce the influence of local noise to the global. Finally, the accurate classification of CHD is achieved using a majority voting classifier with Random Forest and Adaboost classifier based on 84 features containing time domain and frequency domain. Experimental results show that the performance of the proposed method is competitive, and the accuracy, sensitivity, specificity and f1-score of classification for CHD are 0.953, 0.946, 0.961 and 0.953 respectively.