LSTM-Based Arrhythmia Classification in Electrocardiogram Signals

Cardiac arrhythmia, a condition characterized by abnormal heart rhythms, necessitates precise detection methods for effective diagnosis and treatment. This study proposed and evaluated a deep learning-based method, utilizing Long Short Term Memory (LSTM) for arrhythmia detection. The study employed the MIT-BIH arrhythmia dataset, which was partitioned into training, testing, and validation sets. To ensure data quality, denoising processes were performed on the electrocardiogram signals. Additionally, normalization and Rpeak detection techniques were applied to extract relevant arrhythmia features. Addressing the challenge of imbalanced data, oversampling techniques were utilized to augment minority data representation. The proposed deep learning model underwent meticulous evaluation using multiple performance metrics, including accuracy, precision, sensitivity, specificity, F1 score, and root mean square error (RMSE). The results showcased LSTM's remarkable performance, achieving 96% overall accuracy. These finding provide compelling evidence for the effectiveness of deep learning in significantly enhancing arrhythmia detection accuracy, thus advancing the field of cardiac health monitoring and diagnosis.