Polysomnographic identification of anxiety and depression using deep learning

Area of Science:

• Psychiatry
• Computational Neuroscience
• Sleep Medicine

Background:

• Anxiety and depression are prevalent psychiatric conditions with substantial morbidity and healthcare costs.
• A bidirectional relationship exists between anxiety, depression, and sleep quality.
• Polysomnography (PSG) is a standard tool for sleep analysis.

Purpose of the Study:

• To investigate the efficacy of deep learning methods in detecting anxiety and depression from PSG data.
• To develop and validate a machine learning model for psychiatric condition identification using sleep patterns.

Main Methods:

• Utilized PSG data from 940 patients, categorized into anxiety/depression, no anxiety/depression, and likely anxiety/depression groups.
• Transformed 12-channel PSG data into three-channel RGB images for analysis.
• Trained and validated the Xception deep learning model on sleep study data, including hypnograms and composite patient images.

Main Results:

• The Xception model achieved high accuracy (0.9782 on validation, 0.9688 on test set).
• The model demonstrated strong performance with precision (0.9533), recall (0.9630), and F1-score (0.9581) on the independent test set.
• Comparable classification performance was observed with other mainstream deep learning models.

Conclusions:

• Deep learning analysis of PSG data shows significant potential for accurately detecting anxiety and depression.
• These machine learning techniques offer a promising avenue for psychiatric diagnosis and further research in sleep medicine.
• Future studies should explore the broader clinical utility of these AI-driven approaches in psychiatry.