Trustworthy diagnosis of Electrocardiography signals based on out-of-distribution detection.

Cardiovascular disease is one of the most dangerous conditions, posing a significant threat to daily health. Electrocardiography (ECG) is crucial for heart health monitoring. It plays a pivotal role in early heart disease detection, heart function assessment, and guiding treatments. Thus, refining ECG diagnostic methods is vital for timely and accurate heart disease diagnosis. Recently, deep learning has significantly advanced in ECG signal classification and recognition. However, these methods struggle with new or Out-of-Distribution (OOD) heart diseases. The deep learning model performs well on existing heart diseases but falters on unknown types, which leads to less reliable diagnoses. To address this challenge, we propose a novel trustworthy diagnosis method for ECG signals based on OOD detection. The proposed model integrates Convolutional Neural Networks (CNN) and Attention mechanisms to enhance feature extraction. Meanwhile, Energy and ReAct techniques are used to recognize OOD heart diseases and its generalization capacity for trustworthy diagnosis. Empirical validation using both the MIT-BIH Arrhythmia Database and the INCART 12-lead Arrhythmia Database demonstrated our method's high sensitivity and specificity in diagnosing both known and out-of-distribution (OOD) heart diseases, thus verifying the model's diagnostic trustworthiness. The results not only validate the effectiveness of our approach but also highlight its potential application value in cardiac health diagnostics.