Explainable Artificial Intelligence Models for Predicting Depression Based on Polysomnographic Phenotypes.

Area of Science:

• Artificial Intelligence in Medicine
• Mental Health Diagnostics
• Sleep Science

Background:

• Depression is a prevalent mental health disorder with significant mortality and morbidity.
• Current depression screening methods lack robustness and automation, delaying diagnosis and intervention.
• Explainable AI offers a path to transparent and reliable automated detection.

Purpose of the Study:

• To develop explainable AI models for predicting depression using polysomnographic phenotype data.
• To ensure high predictive performance and provide insights into depression risk factors.
• To enhance automated detection for timely mental health interventions.

Main Methods:

• Utilized advanced machine learning algorithms (Random Forest, XGBoost, CatBoost, LightGBM).
• Analyzed phenotype data including subjective questionnaires, clinical assessments, and demographics.
• Employed cross-validation for model performance evaluation.

Main Results:

• Achieved an F1-score of 85% for depression prediction, indicating high reliability.
• Identified key predictive features: anxiety disorders, sleep efficiency, and demographic factors.
• Demonstrated the transparency and actionable insights provided by explainable AI models.

Conclusions:

• Developed reliable, explainable AI models for automated depression detection.
• Highlighted the potential of AI in improving mental health diagnostics and enabling early intervention.
• Emphasized the clinical utility of identified risk factors for targeted patient care.