2020/02/12 by Sajad Mousavi, Mousavi, Sajad, Fatemeh Afghah +3
Medicine · #ECG Monitoring and Analysis #FOS: Biological sciences #FOS: Computer and information sciences #FOS: Electrical engineering #Machine Learning (cs.LG) #Quantitative Methods (q-bio.QM) #Signal Processing (eess.SP) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2002.05262
openalex publication_date 2020/02/12 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Atrial fibrillation (AF) is one of the most prevalent cardiac arrhythmias\nthat affects the lives of more than 3 million people in the U.S. and over 33\nmillion people around the world and is associated with a five-fold increased\nrisk of stroke and mortality. like other problems in healthcare domain,\nartificial intelligence (AI)-based algorithms have been used to reliably detect\nAF from patients' physiological signals. The cardiologist level performance in\ndetecting this arrhythmia is often achieved by deep learning-based methods,\nhowever, they suffer from the lack of interpretability. In other words, these\napproaches are unable to explain the reasons behind their decisions. The lack\nof interpretability is a common challenge toward a wide application of machine\nlearning-based approaches in the healthcare which limits the trust of\nclinicians in such methods. To address this challenge, we propose HAN-ECG, an\ninterpretable bidirectional-recurrent-neural-network-based approach for the AF\ndetection task. The HAN-ECG employs three attention mechanism levels to provide\na multi-resolution analysis of the patterns in ECG leading to AF. The first\nlevel, wave level, computes the wave weights, the second level, heartbeat\nlevel, calculates the heartbeat weights, and third level, window (i.e.,\nmultiple heartbeats) level, produces the window weights in triggering a class\nof interest. The detected patterns by this hierarchical attention model\nfacilitate the interpretation of the neural network decision process in\nidentifying the patterns in the signal which contributed the most to the final\nprediction. Experimental results on two AF databases demonstrate that our\nproposed model performs significantly better than the existing algorithms.\nVisualization of these attention layers illustrates that our model decides upon\nthe important waves and heartbeats which are clinically meaningful in the\ndetection task.\n