Explainable Transformer-based Approach for ECG Anomaly Detection

Anomaly detection is critical in various fields, especially in healthcare. The ability to identify anomalies from normal patterns can significantly contribute to early interventions and enhanced patient outcomes. In Electrocardiogram (ECG) analysis, timely detection of abnormal signals is essential for diagnosing and treating potentially life-threatening conditions. Despite the high performance of many AI-based anomaly detection methods, they often function as ``black boxes", making it difficult to interpret the results. In this paper, we propose integrating explainable artificial intelligence (XAI) into a transformer-based network combined with a support vector data description control chart and multivariate exponential weighted moving average technique (MEWMA-SVDD chart) for robust ECG monitoring. By incorporating XAI, we aim to enhance the transparency and reliability of our model, providing clear and interpretable results. We will demonstrate our proposed approach's effectiveness using a well-known ECG dataset and provide important insights into the detection mechanism. This approach illustrates the importance of combining advanced deep learning techniques with explainability to improve the reliability and efficiency of anomaly detection systems in monitoring healthcare.