A Deep Neural Network for Heart Valve Defect Classification From Synchronously Recorded ECG and PCG

Heart valve defects (HVDs) are commonly analyzed by using heart sound or phonocardiogram (PCG) signals. In many cases, additional information along with PCG analysis helps in obtaining effective decisions. Since time-synchronously recorded electrocardiogram (ECG) and PCG signals can help each other to handle HVD analysis, in this letter, an automatic classification approach is introduced by utilizing both of these signals in a proposed convolutional neural network (CNN)-based architecture. Motivated by the temporal characteristics of these signals, multilayer 1-D CNN is applied on each signal separately to extract 1-D temporal variation pattern. Prior to applying the signals to the proposed network, a preprocessing step followed by three-stage data augmentation is performed. In the multilayer CNN architecture, operation in each layer convolution is followed by batch normalization and dropout operations. Features extracted from the output of the CNN networks are combinedly utilized in the classifier stage. It is found that the use of combined features achieves an overall accuracy of 95.06%, which is around 5% and 2.6% higher than that obtained by using only ECG and only PCG signal, respectively. Moreover, the performance of the proposed method outperforms some existing methods.