OCADN: Improving Accuracy in Multi-Class Arrhythmia Detection From ECG Signals With a Hyperparameter-Optimized CNN

Abstract

Arrhythmia, a heart rhythm disorder, remains a serious global health problem due to its potential to cause complications such as stroke and heart failure. Early detection and accurate classification of arrhythmia are crucial for appropriate medical intervention. Although deep learning holds promise for automatic arrhythmia detection using ECG signals, several challenges need to be addressed. Previous studies often utilize limited and imbalanced datasets and lack exploration of optimal pre-processing and feature extraction methods. To overcome these limitations, this study proposes the Optimized Cardiac Arrhythmia Detection Network (OCADN), a CNN-based model with hyperparameter optimization and advanced pre-processing techniques such as Discrete Wavelet Transform (DWT) and Z-score normalization. As a comparison to OCADN, this research also develops an arrhythmia detection model using the LSTM algorithm. Experimental results demonstrate that OCADN outperforms LSTM, achieving high accuracy, precision, sensitivity, specificity, and F1-score on both training and test data. The consistent performance of OCADN on both datasets indicates its robustness and potential for clinical implementation. OCADN with hyperparameter tuning exhibits accuracy, precision, sensitivity, specificity, and F1-score of 99.97%, 99.97%, 99.97%, 99.99%, and 99.97%, respectively, on the training data. Meanwhile, the performance on the testing data for accuracy, precision, sensitivity, specificity, and F1-score is 98.87%, 95.23%, 98.09%, 99.65%, and 96.59%, respectively.