Heart Disease Prediction Through Ecg Filtering And Classification Methodologies Based On Machine Learning And Deep Learning Techniques

Abstract

The preliminary identification and prevention of heart or Cardiovascular Disease newline(CVD) , responsible in a wide margin for morbidity and mortality in industrialized newlinecountries and developing countries, (affecting approx. 5 billion people around the newlineworld. Cardiac arrhythmias like atrial fibrillation are commonly diagnosed and newlineassessed using an electrocardiogram (ECG). Most of the ECG segmentation based newlinecomputer aided automated cardiac anomaly monitoring techniques need a robust newlinedetection of ECG segments like QRS complexes so as to yield an accurate result. But newlinereliable detection of QRS spikes can be problematic due to the presence of noise and newlineaberrations emerging from the ECGs, especially when the signals are collected using newlinewearables. The contaminated ECG data generated in some previous studies, however, newlinewere not obtained using real noisy data from ECG signals, as majority of the standard newlinedenoising methods were tested using fictitious noise addition to the clean ECG signal. newlineIn the first phase, we focused on developing an integrated scheme for ECG signal newlinefiltration and detecting diverse types of peaks in ECG. An idea built on the practice of newlinewavelet packet transform and Neural Networks (NN), the wavelet coefficients are newlinecalculated, and using the NN feed-forward module, the weights are updated. To newlinedetect the peaks of the ECG, we utilize the well-known convolutional layers and then newlinecombine them with the Bidirectional Long Short-Term Memory Networks newline(BiLSTM) architecture. One of the many advantages of the BiLSTM models is its newlineability to solve the unsteady gradient problems too while maintaining the long-term newlinedependency. Moreover, for classifying the ECG beats and recognizing five types of newlineheartbeat types (PVC, LBBB, RBBB, PACE, and APC), we adopt a combined feature newlineextraction-based approach that obtains both wavelet transform and morphological newlinefeatures. The feature extraction step utilizes high-order statistics, morphological newlinefeatures, and wavelet transform to create the robust feature. Now the PCA module is newlineused