Multimodal AI for predicting comorbid REM sleep behavior disorder in major depressive disorder

Area of Science:

• Neuroscience
• Artificial Intelligence
• Psychiatry

Background:

• REM sleep behavior disorder (RBD) is a significant prodromal marker for α-synucleinopathies, with high conversion rates to Parkinson's disease (PD).
• Major depressive disorder (MDD) patients with comorbid RBD represent a subgroup at increased risk for developing PD, but this comorbidity is often overlooked in clinical settings.

Purpose of the Study:

• To develop and validate a multimodal artificial intelligence (AI) framework for detecting comorbid RBD in patients diagnosed with MDD.
• To identify potential digital biomarkers indicative of prodromal synucleinopathy within this patient group.

Main Methods:

• A dual-stream multimodal AI model was developed, integrating facial dynamics from video and vocal features from speech recordings.
• The model was trained and evaluated on a cohort of 31 patients (19 with MDD-RBD, 12 with MDD-only controls), utilizing reading and spontaneous-speech tasks.
• Explainability analysis was performed to identify candidate digital biomarkers associated with RBD detection.

Main Results:

• The best-performing AI model achieved 80.5% accuracy and an F1-score of 0.848 in predicting comorbid RBD in MDD patients.
• Explainability analysis pinpointed reduced lower-face tension and variability, along with brow lowering, as potential biomarkers.
• Predicted RBD risk showed significant correlation with the RBD Questionnaire (RBDQ) score and a trend towards correlation with UPDRS motor scores.

Conclusions:

• Multimodal AI demonstrates significant promise for predicting RBD in MDD patients, potentially aiding in the early identification of individuals at risk for PD.
• The study identified interpretable digital markers that could serve as novel tools for detecting prodromal synucleinopathy in psychiatric populations.