CHMMConvScaleNet: a hybrid convolutional neural network and continuous hidden Markov model with multi-scale features...

Area of Science:

• Sleep Medicine
• Biomedical Engineering
• Wearable Technology

Background:

• Sleep posture is crucial for sleep wellness and has implications for conditions like obstructive sleep apnea.
• Monitoring sleep posture is vital for bedridden patients to prevent pressure ulcers.
• Current methods for sleep posture recognition often require numerous sensors.

Purpose of the Study:

• To develop and validate a novel method, CHMMConvScaleNet, for accurate sleep posture recognition.
• To assess the efficacy of CHMMConvScaleNet using limited piezoelectric ceramic sensors.
• To demonstrate the potential of CHMMConvScaleNet for portable home sleep monitoring.

Main Methods:

• CHMMConvScaleNet utilizes pressure signals from a limited sensor array.
• A Movement Artifact and Rollover Identification (MARI) module detects rollover events.
• Multi-scale spatiotemporal features are extracted using sub-convolution networks and optimized with a Continuous Hidden Markov Model (CHMM).

Main Results:

• CHMMConvScaleNet achieved high performance metrics: 92.91% recall, 91.87% precision, and 93.41% accuracy.
• The method demonstrated comparable performance to state-of-the-art techniques using significantly fewer sensors.
• Data was collected from 22 participants using a 32-sensor array, yielding 8583 samples.

Conclusions:

• CHMMConvScaleNet offers an effective solution for sleep posture recognition with minimal sensors.
• The method shows significant potential for developing portable devices for home sleep monitoring.
• This approach can contribute to improved sleep wellness and patient care, particularly for those at risk of sleep apnea or pressure ulcers.