PPG-Based Accurate Insomnia Detection System Using Convolutional Neural Networks With Self-Attention Mechanism and Gated Recurrent Units
Adequate sleep is essential for physical and mental well-being. The quality of sleep is significantly impacted by sleep disorders like narcolepsy, sleep apnea, nocturnal frontal lobe epilepsy, and insomnia. Insomnia is a prevalent sleep disorder characterized by difficulty initiating or maintaining...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11124893/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Adequate sleep is essential for physical and mental well-being. The quality of sleep is significantly impacted by sleep disorders like narcolepsy, sleep apnea, nocturnal frontal lobe epilepsy, and insomnia. Insomnia is a prevalent sleep disorder characterized by difficulty initiating or maintaining sleep, leading to daytime fatigue and restlessness. An accurate and automated detection system is essential for effective diagnosis and management. Usually, polysomnogram (PSG) signals are utilized to identify sleep disorders, but these signals are challenging to handle, computationally intensive to process, and affect the convenience of patients. The proposed method provides a non-invasive approach for insomnia detection. To the best of our knowledge, this is the first study that utilizes single-channel photoplethysmography (PPG) to automatically detect insomnia. In the proposed study, we performed experiments on two publicly accessible datasets: the Cyclic Alternating Pattern (CAP) sleep database and the MESA Sleep Dataset, considering a diverse set of subjects. This study introduces a novel approach for PPG-based insomnia detection, utilizing Convolutional Neural Network (CNN) with self-attention, CNN with Gated Recurrent Unit (GRU), and transformer-based models. The CNN with GRU-based approach demonstrated notable performance, achieving impressive classification accuracy of 96.00%, alongside robust precision, recall, and F1-score. Importantly, we further explored model explainability using Grad-CAM, providing visual insights into the learned model’s decision-making from PPG signals. The results emphasize the effectiveness and interpretability of these advanced models in capturing complex patterns in PPG signals, establishing PPG-based sleep monitoring as a practical and efficient solution for home use with wearable devices. |
|---|---|
| ISSN: | 2169-3536 |