Arrhythmia detection algorithm based on domain knowledge and spatio-temporal network model

Cardiovascular disease (CVD) remains the leading cause of mortality globally, and arrhythmia detection serves as a critical foundation for CVD diagnosis. The electrocardiogram (ECG) is a pivotal tool for diagnosing arrhythmias. In recent years, end-to-end deep learning approaches have achieved remarkable advancements. Deep learning models can effectively extract spatiotemporal features from ECG signals, yielding favorable classification results on specific datasets. However, the inter-patient variability in ECG signals significantly reduces model accuracy when applied to cross-population scenarios (inter-patient paradigm). To address this challenge, we propose a hybrid model that integrates medical domain knowledge features with deep learning features. By employing an attention mechanism, the model dynamically assigns weights to different features rather than relying on simplistic concatenation. Our proposed model demonstrates its efficacy on the MIT-BIH dataset, overcoming the generalization limitations of traditional methods in cross-patient data analysis, and providing robust decision support for intelligent diagnostics.