Enhanced Multi-Class Cardiac Sound Diagnosis using Advanced PCG Signal Analysis and Deep Learning Integration

The early detection of cardiac abnormalities is crucial for effective patient management and improved health out- comes. This research presents an innovative approach to cardiac sound diagnosis by integrating advanced phonocardiographic (PCG) signal analysis with deep learning (DL) techniques. To assess the system’s dependability over a range of patient demographics and recording circumstances, extensive testing is conducted on several benchmark datasets, such as PhysioNet and private hospital datasets. The recommended method outperforms the state-of-the-art benchmarks in multi-class classification tasks, attaining remarkable accuracy rates. Furthermore, thorough ablation investigations demonstrate the contributions of every element, emphasizing the significance of domain-specific feature engineering and transfer learning. Feature extraction was conducted through time-domain and frequency-domain analysis, facilitating the capture of critical sound characteristics. We implemented a DL architecture as per CNNs (convolutional neural networks) to enhance classification accuracy. Enhancing patient outcomes as well as lowering global burden of cardiovascular illnesses depend heavily on the precise diagnosis of heart problems. This work introduces a new method for diagnosing multi-class cardiac sounds by combining DL methods with sophisticated phonocardiogram (PCG) signal analysis. This study provides a reliable, effective, and scalable diagnostic tool by bridging the gap between conventional signal processing techniques and contemporary artificial intelligence (AI) by utilizing cutting-edge transfer learning frameworks. For a long time, cardiac auscultation has been a crucial clinical skill. However, because it is subjective, practitioners frequently perceive it differently. Automated PCG analysis offers a trustworthy and impartial method of detecting abnormal cardiac sounds in order to overcome this difficulty. Acquiring high-quality PCG signals from a variety of datasets covering a broad range of normal and pathological circumstances is the first step in the suggested approach. To guarantee the best input for the DL model, these signals go through pre-processing procedures like noise reduction, segmentation, and feature extraction.