Early Detection of Heart Diseases through Deep Learning–Based Electrocardiogram (ECG) Analysis

Abstract

Cardiovascular diseases are among the leading causes of death worldwide, and early diagnosis
plays a critical role in reducing mortality rates. Electrocardiogram (ECG) signals are a fundamental
diagnostic tool for identifying arrhythmias; however, manual interpretation requires expert knowledge and
is prone to diagnostic errors. In this study, a hybrid deep learning model combining Convolutional Neural
Networks (CNN) and Long Short-Term Memory (LSTM) networks was developed for automatic
arrhythmia detection from ECG signals. The model was trained using the MIT-BIH Arrhythmia Database,
and several preprocessing steps—including noise removal, windowing, standardization, and SMOTE-based
data balancing—were applied. The proposed CNN-LSTM model achieved high performance in five-class
(N, S, V, F, Q) arrhythmia classification, with an accuracy of 95.2%, an F1 score of 92%, and a loss value
of 0.14. The confusion matrix results indicate that the model demonstrates strong generalization capability,
particularly for the N, V, and Q classes. These findings suggest that the hybrid architecture can serve as an
effective decision-support tool in ECG signal analysis. Furthermore, integrating the proposed system into
portable devices may enable real-time and automated monitoring of cardiac rhythms; however, the model
has not been approved by regulatory agencies such as the FDA or CE, and further validation is required
before clinical use. Additionally, by reducing the diagnostic burden on clinicians, the system has the
potential to improve both accuracy and efficiency within clinical workflows. Future studies incorporating
cross-dataset validation and larger ECG collections may further strengthen the reliability and robustness of
the proposed approach.