Validity study of a multi-scaled fusion network using single-lead electrocardiogram signals for obstructive sleep apnea diagnosis
- S-Med
- May 30
- 1 min read
Here, we developed and evaluated the ability of the SE-MSCNN deep learning–based algorithm for detecting AH events in the FAH-ECG dataset and generated several novel findings. We found that the performance of our proposed single-lead ECG signal model was comparable to the high performance of the public AED dataset in a real clinical setting, with an accuracy, sensitivity, specificity, and an F1 score of 86.6%, 83.3%, 89.1%, and 0.843, respectively. Compared with other studies, the SE-MSCNN was superior by a large margin, particularly with regard to sensitivity. For the diagnosis and classification of OSA, the SE-MSCNN demonstrated good agreement with manual scoring. Finally, the results showed the high efficiency, reliability, validity, and potential wide-ranging applications of the automated scoring algorithm.
Given the extensive application of automated scoring software for sleep staging and sleep apnea detection, the AASM requires physicians to review and evaluate the accuracy of sleep study reports. As conventional automated scoring software is based on statistical reasoning, its accuracy is limited.Therefore, excessive editing is required, significantly increasing the time burden for sleep technicians. In particular, for inexperienced physicians or medical institutions without professional technicians, this significant reliance on automated scoring software may lead to misdiagnosis and missed diagnosis. Previous studies have shown that AH detection methods based on deep neural networks are effective, but there is still scope for improvement.This study aimed to develop and validate a deep neural network algorithm with high performance for respiratory event detection to further increase laboratory efficiency and standardize scoring results within and across sleep centers.
Comments