A Generic Approach of ECG Signals Classification Using CNN

Abstract

Based on current medical applications, classification of The ECG signs performs a crucial role in the diagnosis of heart disease. The challenging part is to obtain an accurate classification of ECG. There are Machine learning algorithms available that provide solutions for classification as well as feature extraction of ECG data available. But these solutions only confine to basic feature learning architectures but not in-depth extraction. ANN
classifiers are one such mainstream classifier widely utilized in the relegation of ECG signals. In this paper, a versatile 1-D Convolutional Neural Network (CNN) is used for feature extraction and classification performed on two datasets (MIT-BIH and PTBDB) which are presented in the result section. Visualized 5000 sampled data points using the embeddings of our baseline CNNs. Here we show the visualization obtained from the application of PCA to extract the 2 most informative dimensions from these embeddings. We also use K-means
to perform clustering, showing the results in the same representation. The proposed method is efficient and generic due to its parameter invariant nature which makes it applicable for any ECG dataset.