Objective To identify the heart sounds of aortic stenosis by deep learning model based on DenseNet121 architecture, and to explore its application potential in clinical screening aortic stenosis. Methods We prospectively collected heart sounds and clinical data of patients with aortic stenosis in Tianjin Chest Hospital, from June 2021 to February 2022. The collected heart sound data were used to train, verify and test a deep learning model. We evaluated the performance of the model by drawing receiver operating characteristic curve and precision-recall curve. Results A total of 100 patients including 11 asymptomatic patients were included. There were 50 aortic stenosis patients with 30 males and 20 females at an average age of 68.18±10.63 years in an aortic stenosis group (stenosis group). And 50 patients without aortic valve disease were in a negative group, including 26 males and 24 females at an average age of 45.98±12.51 years. The model had an excellent ability to distinguish heart sound data collected from patients with aortic stenosis in clinical settings: accuracy at 91.67%, sensitivity at 90.00%, specificity at 92.50%, and area under receiver operating characteristic curve was 0.917. Conclusion The model of heart sound diagnosis of aortic stenosis based on deep learning has excellent application prospects in clinical screening, which can provide a new idea for the early identification of patients with aortic stenosis.
Feature extraction methods and classifier selection are two critical steps in heart sound classification. To capture the pathological features of heart sound signals, this paper introduces a feature extraction method that combines mel-frequency cepstral coefficients (MFCC) and power spectral density (PSD). Unlike conventional classifiers, the adaptive neuro-fuzzy inference system (ANFIS) was chosen as the classifier for this study. In terms of experimental design, we compared different PSDs across various time intervals and frequency ranges, selecting the characteristics with the most effective classification outcomes. We compared four statistical properties, including mean PSD, standard deviation PSD, variance PSD, and median PSD. Through experimental comparisons, we found that combining the features of median PSD and MFCC with heart sound systolic period of 100–300 Hz yielded the best results. The accuracy, precision, sensitivity, specificity, and F1 score were determined to be 96.50%, 99.27%, 93.35%, 99.60%, and 96.35%, respectively. These results demonstrate the algorithm’s significant potential for aiding in the diagnosis of congenital heart disease.