Assessment of the long RR intervals using convolutional neural networks in single-lead long-term Holter electrocardiogram recordings.

Advancements in medical technology have extended long-term electrocardiogram (ECG) monitoring from the traditional 24 h to 7-14 days, significantly enriching ECG data. However, this poses unprecedented challenges for physicians in analyzing these extensive datasets. While artificial neural networks have shown great potential in the automatic recognition of arrhythmias and are gradually being adopted widely, their effectiveness still requires rigorous evaluation in clinical applications. Although there have been many research on the evaluation of atrial fibrillation recognition performance, systematic evaluation of automatic detection algorithms for prolonged RR interval(RRI) remains lacking. This study aims to comprehensively evaluate these algorithms based on convolutional neural networks (CNN). We collected 480 single-lead long-term dynamic ECG records from our hospital, all from patients with prolonged RRI. Both CNN algorithms and manual analysis methods were used to analyze the ECGs, with manual analysis serving as the gold standard for performance evaluation. Results indicated that the CNN algorithm achieved an average recall rate of 98.41%, an average precision of 98.68%, and an average F1 score of 98.54%. These metrics demonstrate that CNN algorithms can meet the clinical demands for recognizing prolonged RRI, thereby enhancing physicians' confidence, especially when dealing with large volumes of RRI data. AI provides invaluable support for quantitative analysis and significantly improves diagnostic efficiency. This evaluation offers a reliable basis for deploying single-lead intelligent monitoring devices in households, communities, nursing homes, and other settings.