A Multi-Task Deep Learning Approach for Simultaneous Sleep Staging and Apnea Detection for Elderly People.

Accurately detecting sleep stages and sleep apnea is essential for assessing sleep quality and diagnosing related health issues, especially in the elderly. However, most existing deep learning models focus on single-task scenarios, overlooking the synergy between sleep staging and apnea detection. Moreover, they don’t consider the impact of age on sleep. To address these issues, this study selects 40 subjects aged 60–90 from the Human Sleep Project dataset as research participants. Supervised principal component analysis is applied to process their polysomnography signals, and S-transform is used to generate S-dynamic spectrograms as model inputs. Furthermore, a deep learning model (RGMNet) based on a multitasking has been designed specifically for older adults. By leveraging its unique feature extraction and interaction mechanisms, RGMNet effectively enables simultaneous sleep staging and apnea detection on datasets of older adults. Experimental results demonstrate that the proposed RGMNet model achieves sleep staging and apnea detection accuracies of 86.61% and 84.12%, respectively, outperforming existing multi-task and single-task methods. These findings highlight the effectiveness of RGMNet in improving the accuracy of simultaneous sleep staging and apnea detection for older adults, providing valuable insights for advancements in sleep medicine and related fields.