An interactive web application to identify early Parkinsonian non-tremor-dominant subtypes

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computational Biology

Background:

• Parkinson's disease (PD) presents heterogeneous motor subtypes, specifically tremor-dominant (TD) and non-tremor-dominant (NTD).
• Accurate and rapid differentiation of these subtypes is crucial for tailoring personalized treatment strategies.
• Current diagnostic methods may not fully capture the nuances required for precise subtype classification.

Purpose of the Study:

• To develop and validate a machine learning model for the accurate classification of PD motor subtypes.
• To identify key predictors, including clinical assessments and biomarkers, that differentiate between TD and NTD PD patients.
• To create an accessible tool for real-time motor subtype identification in clinical practice.

Main Methods:

• Utilized data from the Parkinson's Disease Progression Marker Initiative (PPMI) cohort.
• Employed recursive feature elimination (RFE) to identify significant predictive features.
• Trained and evaluated seven classical machine learning models, including support vector machines, for motor subtype prediction, using AUC and follow-up data for validation.

Main Results:

• A feature subset identified by RFE, including clinical data and CSF alpha-synuclein (CSF α-syn), improved model performance.
• The polynomial support vector machine (P-SVM) achieved the highest AUC of 0.898.
• The P-SVM model incorporating CSF α-syn demonstrated superior performance compared to models without it (P=0.034).

Conclusions:

• An interactive web application was developed based on the validated P-SVM model.
• This tool enables rapid identification of PD motor subtypes, facilitating better patient status understanding.
• The application supports the development of personalized treatment plans for PD patients.