Machine learning to diagnose, classify and predict phenoconversion in isolated REM sleep behavior disorder.

Area of Science:

• Neurology
• Artificial Intelligence
• Sleep Medicine

Background:

• Rapid eye movement sleep behaviour disorder (RBD) is an early indicator for neurodegenerative diseases like Parkinson's disease.
• Machine learning (ML) presents potential for improving diagnosis and risk assessment in individuals with RBD.

Purpose of the Study:

• To systematically review the application and validation of ML models for diagnosing and predicting phenoconversion in RBD.
• To identify methodological limitations and suggest future research priorities for ML in RBD.

Main Methods:

• Systematic review of PubMed, Embase, and Medline databases (2014-2025) following PRISMA guidelines.
• Inclusion of 75 studies on adult humans with clinically diagnosed RBD/iRBD using validated ML models.
• Analysis of studies focusing on diagnosis, phenotyping, and prediction of phenoconversion to α-synucleinopathies.

Main Results:

• Diagnostic ML models reported accuracies from ~63% to ~99.7% (median ~90%) using various biomarkers.
• Phenoconversion models achieved AUCs up to ~0.94 but often lacked external validation and used heterogeneous endpoints.
• A wide array of ML algorithms were employed, with Support Vector Machines, Random Forests, and Logistic Regression being most common.

Conclusions:

• ML approaches hold promise for scalable diagnosis and risk stratification in idiopathic RBD (iRBD).
• Current progress is hindered by methodological biases, inconsistent endpoints, data imbalance, and a lack of explainable, validated models.
• Methodological improvements are crucial for developing clinically interpretable and translatable ML tools for RBD management.