Feature Extraction and Fusion of ECG Signals Using MFCC and DWT for Cardiovascular Disease Diagnosis

Abstract

Cardiovascular diseases (CVDs) remain a leading cause of mortality globally, making early and accurate detection crucial for effective treatment. Electrocardiogram (ECG) signals play a vital role in diagnosing these conditions, and advanced feature extraction and classification techniques can enhance detection accuracy. The objective of this research is to develop and implement feature extraction and classification methods for ECG signals using lightweight deep learning models, tailored for real-time CVD detection on IoT devices. The proposed method integrates Mel Frequency Cepstral Coefficients (MFCC) and Discrete Wavelet Transform (DWT) to extract significant features from ECG signals, capturing both frequency-domain and time-domain information. By combining these complementary features, the method enhances the ability to identify patterns indicative of cardiovascular diseases. For classification, Convolutional Neural Networks (CNNs) and two lightweight deep learning models, VGG16 and MobileNet, are employed to ensure computational efficiency while maintaining high accuracy. The lightweight nature of these models makes them suitable for deployment on resource-constrained IoT devices, enabling real-time monitoring. The models are validated using real-world datasets to ensure robust performance, with sensitivity and specificity as key metrics for evaluating their effectiveness in detecting CVDs. The results demonstrate that the proposed feature extraction and fusion method, coupled with optimized lightweight models, achieves high accuracy in classifying ECG signals, contributing to early detection and intervention of cardiovascular diseases. The research shows promising potential for real-time, on-device implementation in healthcare systems, offering an efficient, scalable solution for CVD detection.