Sleep Posture Classification using a Convolutional Neural Network.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Sleep Medicine

Background:

• Sleep disorders like REM sleep behaviour disorder (RBD), sleep apnea, and restless leg syndrome (RLS) affect many individuals.
• Diagnosis typically requires polysomnography (PSG), an invasive method involving electrodes.
• Sleep disorder behaviors are often linked to specific body positions and movements.

Purpose of the Study:

• To develop a non-contact method for measuring sleep disorders.
• To utilize Infrared (IR) camera technology for monitoring body postures and movements during sleep.
• To classify different sleep postures using artificial intelligence.

Main Methods:

• An Infrared (IR) camera was employed for non-contact monitoring of body positions.
• Twelve participants were instructed to adopt and transition through 12 predefined sleep positions.
• Convolutional Neural Networks (CNNs) were utilized for automatic feature extraction and sleep posture classification from IR data.

Main Results:

• The proposed non-contact method achieved classification accuracies ranging from 0.76 to 0.91.
• Accuracy varied with and without blanket cover, indicating robustness.
• The system demonstrated effectiveness in classifying 12 distinct sleep postures across participants.

Conclusions:

• The non-contact IR camera and CNN approach is a promising tool for detecting common sleep postures.
• This method has the potential to characterize sleep disorder-related behaviors non-invasively.
• Further development could lead to improved, patient-friendly sleep disorder assessment.