Detection and Classification of Cardiac Arrhythmia using DeepCNNFeatures via Transfer Learning and Optimized -KNN

Deep Learning(DL) has emerged as a significant area of research across various fields, particularly in healthcare. The identification of irregularities in Electrocardiogram (ECG) readings is crucial for effective patient monitoring. Research indicates that a DL model when trained on available datasets, can identify arrhythmias with high accuracies. The ECG procedure is straightforward and non-invasive, serving as a vital tool for predicting and diagnosing Cardiac Arrhythmia. This paper introduces an innovative DL model that employs transfer learning to extract features using deep networks and enables the automatic classification of ECG beats into sixteen distinct categories using the Optimized K-Nearest Neighbour Classifier (KNN). The proposed research approach initiates with collecting ECG Dataset from MIT-BIH Database. Then noise is removed using DWT Filtering. Segmentation of beats is done to extract ECG beats and labelled as per the annotation file. Further deep features are extracted using Deep Learning Models-AlexNet and ResNet18 and Classification is done using optimized KNN Classifier. Feature maps learned by a model when trained on input ECG beats can be used as general descriptors of ECG signal. Proposed system records an accuracy of 97.98% using AlexNet Feature Set and 98.18% using ResNet18 Feature Set. Our proposed transfer learning based inference model is effective, promising when compared with existing Techniques