Опубликована: Май 7, 2025

Mental health issues have become a significant global public challenge. Traditional assessments rely on subjective methods with limited ecological validity. Passive sensing via wearable devices and smartphones, combined machine learning (ML), enables objective, continuous, noninvasive mental monitoring. This study aims to provide comprehensive review of the current state passive sensing-based (ML) technologies for We summarize technical approaches, reveal association patterns between behavioral features disorders, explore potential directions future advancements. Following PRISMA-ScR guidelines, we searched seven major databases (Web Science, PubMed, IEEE Xplore, etc.) studies published 2015 2025. A total 42 were included. Information was extracted from dimensions such as data collection, preprocessing, feature engineering, ML methods, validation, integrating (e.g., sleep, activity, social interaction) disorders depression, anxiety). The found that most commonly used digital biomarkers heart rate (n=28), movement index (n=25), step count (n=17), which significantly associated depression anxiety. Deep models CNN, LSTM) performed exceptionally well in processing time-series data. However, traditional random forest, XGBoost), due their higher interpretability, remain widely adopted. Current face challenges small sample sizes (median = 60.5 participants), short collection periods (45.24% had less than 7 days), device variety (76.19%). Additionally, only one conducted external limiting clinical generalizability models. On ethical front, few (14.29%) explicitly mentioned anonymization, highlighting need enhanced privacy protection algorithm fairness. combination offers innovative solutions key challenges, including quality, model generalization, standards, be addressed before translation. Future research should focus large-scale longitudinal multimodal integration, optimization, interdisciplinary collaboration drive widespread adoption application these technologies.

Язык: Английский