Multi-task deep learning for sub-clinical screening of mood and sleep disturbances using physical activity biomarkers

Sub-clinical screening for mood disorders and sleep disturbances remains challenging due to limited accessibility of mental health screening. Physical activity patterns serve as accessible biomarkers reflecting psychological wellbeing, offering potential for community-based screening. This study aims to develop and validate a multi-task deep learning framework (MUSCLE) for concurrent risk identification of mood disturbances and sleep disorders using physical activity data.

We analyzed a comprehensive Taiwanese national dataset (n = 69,559) containing physical fitness assessments and self-reported wellbeing indicators. After preprocessing, 44,477 complete records underwent analysis using our Multi-task Synchronized Convolutional LSTM Evaluator (MUSCLE) model, which integrates convolutional neural networks with long short-term memory networks to simultaneously predict mood disturbance risk, sleep disorder indicators, and psychological wellbeing. Performance was evaluated through 10-fold cross-validation against nine alternative approaches.

MUSCLE demonstrated superior performance in detecting mood disturbances with precision of 0.929, recall of 0.899, and F₁-score of 0.914, while sleep disorder detection achieved precision of 0.881, recall of 0.983, and F₁-score of 0.930. Psychological wellbeing assessment achieved precision of 0.746, recall of 0.834, and F₁-score of 0.787. The model significantly outperformed traditional machine learning and single-task neural networks across all screening tasks.

This study establishes multi-task deep learning efficacy for sub-clinical screening of mood and sleep disturbances using accessible physical activity biomarkers. MUSCLE's ability to capture interdependencies between physical activity patterns and mental health enables comprehensive concurrent risk identification, providing a cost-effective and scalable complement to traditional mental health screening that supports timely clinical referral for at-risk individuals.