Enhanced Identification of Valvular Heart Diseases through Selective Phonocardiogram Features Driven by Convolutional Neural Networks (SFD-CNN)

Abstract

Valvular Heart Disease (VHD) is a significant cause of mortality worldwide. Although extensive research has been conducted to address this issue, practical implementation of existing VHD detection results in medicine still falls short of optimal performance. Recent investigations into machine learning for VHD detection have achieved commendable accuracy, sensitivity, and robustness. To address this limitation, our research proposes utilizing Selective Phonocardiogram Features Driven by Convolutional Neural Networks (SFD-CNN) to enhance VHD detection. Notably, SFD-CNN operates on phonocardiogram (PCG) signals, distinguishing itself from existing methods based on electrocardiogram (ECG) signals. We present two experimental scenarios to assess the performance of SFD-CNN: one under default parameter conditions and another with hyperparameter tuning. The experimental results demonstrate that SFD-CNN surpasses other existing models, achieving outstanding accuracy (96.80%), precision (93.25%), sensitivity (91.99%), specificity (98.00%), and F1-score (92.09%). The outstanding performance of SFD-CNN in VHD detection suggests that it holds great promise for practical use in various medical applications. Its potential lies in its ability to accurately identify and classify VHD, enabling early detection and timely intervention. SFD-CNN could significantly improve patient outcomes and reduce the burden on healthcare systems. With further development and refinement, SFD-CNN has the potential to revolutionize the field of VHD detection and become an indispensable tool for healthcare professionals.