A hybrid ECG-based arrhythmia detection system combining LSTM and physiological scoring achieved a peak F1-score of 74.5% on the MIT-BIH database, outperforming standalone baselines.
Does a hybrid ECG-based arrhythmia detection system improve detection performance compared to standalone models?
A hybrid AI and physiological modeling approach for ECG arrhythmia detection shows promising performance but highlights challenges in generalization across different clinical datasets.
This repository contains the framework and evaluation data for a hybrid ECG-based arrhythmia detection system developed at RV College of Engineering. The proposed approach addresses the limitations of standalone deep learning and physiological models by combining an LSTM-based beat classifier with a rule-based physiological scoring system. By extracting temporal morphology features alongside heart rate variability (HRV) and rhythm stability metrics , the framework provides an interpretable and robust pipeline for cardiac anomaly detection. Evaluated on the MIT-BIH Arrhythmia Database using strict patient-wise validation, the hybrid system's weighted adaptive fusion strategy achieved a peak F1-score of 74.5%, outperforming standalone baselines. Cross-database validation on the PTB (87.1% F1) and European ST-T (36.9% F1) datasets highlights the system's generalization capabilities and the ongoing challenges of clinical domain shifts. A Streamlit-based dashboard is included to support real-time clinical decision visualization.
R.V. College of Engineering (Mon,) conducted a other in Arrhythmia. Hybrid ECG-based arrhythmia detection system vs. Standalone baselines was evaluated on Peak F1-score on MIT-BIH Arrhythmia Database. A hybrid ECG-based arrhythmia detection system combining LSTM and physiological scoring achieved a peak F1-score of 74.5% on the MIT-BIH database, outperforming standalone baselines.