Patient-adaptive and clinically interpretable arrhythmia detection from ECG using lightweight hybrid deep learning

A precise and reliable identification of cardiac arrhythmias from electrocardiogram (ECG) signals is critical for making informed clinical decisions. In this paper, a patient-adaptive and clinically interpretable a hybrid CNN-LSTM model with an attention mechanism is offered as a framework for ECG arrhythmia identification. The proposed framework combines signal preprocessing techniques with domain knowledge-driven morphological and time-frequency features to jointly analyze the waveform patterns and temporal dependencies in noisy ECG signals. In addition to beat-level classification, a composite arrhythmia burden score is also proposed to offer a patient-level summary of arrhythmic activity, and a lightweight patient-specific adaptation method is used to enhance reliability with limited calibration data. Over 99% accuracy and strong sensitivity to ventricular and supraventricular arrhythmias are demonstrated by experimental results employing a rigorous inter-patient methodology on the MIT-BIH Arrhythmia Database. The suggested framework's capacity to generalize on actual clinical datasets is further validated by external validation on the PTB-XL dataset without retraining. The proposed framework is also scalable for cloud-based deployment to enable continuous ECG monitoring, thus emphasizing its potential for practical and patient-centric cardiac care.