Signal Processing Techniques for Classifying Heartbeat Patterns in ECG Signals for Cardiac Disease Detection Using Graph Convolutional Networks

In recent days, the information about electrical activity is obtained in the form of Electro-Cardo-Diograms (ECG) which is become a typical way of monitoring heart rhythm and function of patients. In existing, the Convolutional Neural Networks (CNN) is utilized for the effective detection of Cardiac Disease (CD). However, the CNN model may not extract the spatial features from the signals effectively. To overcome this issue, Graph Convolutional Network (GCN) is proposed to extract the spatial features from signal effectively. Initially, the data is taken from Massachusetts Institute of Technology and Beth Israel Hospital (MIT-BIH) dataset. Then, min-max normalization is utilized to preprocess the data and the high frequency noise is removed with Butterworth Frequency Filter (BFF). After that, the preprocessed data is further processed into classification with proposed GCN model. In the classification, the convolutional layers are utilized to extract the spatial features by using graph-based structure. Finally, the fully connected layer with softmax activation function is utilized to classify the CD into normal or abnormal labels. From the results, the proposed GCN model established better results than existing CNN including precision (99.75%), recall (98.60%), F1-score (98.20%) and Area Under Curve (99.70%) respectively.