Sensor-Enhanced Deep Learning for ECG Arrhythmia Detection: Integrating Multiscale Convolutional and Self-Attentive GRU Networks

The real-time monitoring of electrical activities of the heart using a wearable electrocardiogram (ECG) sensor plays a vital role in providing real-time data and allows for the immediate detection of arrhythmia events for patients with high risk of cardiac diseases. This work presents an architecture that capitalizes on multiscale convolutional neural networks (CNNs) and gated recurrent units (GRUs), augmented with a self-attention mechanism, to thoroughly analyze both spatial and temporal aspects of ECG signals. This innovative integration enables the model to detect nuanced arrhythmic patterns effectively, thereby addressing the complex nature of ECG interpretation. The proposed model's performance is substantiated by its high diagnostic accuracy, reaching a peak accuracy of 99.63%, which is a marked improvement over the existing models. It is optimized for real-time analysis, featuring a significant reduction in computational complexity and memory usage, distinguishing it from other high-performing but computationally intensive frameworks. Moreover, this article delineates the signal lengths and datasets, ensuring a comprehensive validation against established benchmarks. The system demonstrates 98.39%, 99.63%, and 99.00% of precision, recall, and F1-scores, respectively. The work also elucidate the importance of sensor technology in enhancing diagnostic precision, detailing the role of sensor sensitivity and specificity in our system's overall efficacy.