MRI-Based Machine Learning Classifier and its Association with Motor Progression in Parkinson’s Disease Based on Clinical Measures
Daniel Teixeira-Dos-Santos1, Anupa Ambili Vijayakumari1, Hubert Fernandez1, Benjamin Walter1
1Center for Neurological Restoration, Cleveland Clinic
Objective:
To determine whether a structural MRI–based machine learning classifier, originally trained on OFF-medication motor outcomes, can predict clinical outcomes commonly used in care.
Background:
Parkinson’s disease (PD) progresses at variable rates, and reliable tools to identify patients at risk for rapid decline are limited. We previously developed an MRI-based classifier that distinguished faster from slower motor progressors using OFF-medication MDS-UPDRS Part III scores. Because OFF assessments are rarely performed in routine practice, this study tested whether the classifier also relates to outcomes commonly collected in clinical care.
Design/Methods:
A support vector machine classifier integrating multivariate gray matter volumetric distance with baseline demographic and clinical features was applied to de novo PD patients in the Parkinson’s Progression Markers Initiative. Outcomes over 48 months included change in MDS-UPDRS Part II (motor experiences of daily living), Schwab & England Activities of Daily Living (S&E ADL), and levodopa equivalent daily dose (LEDD). Secondary analyses assessed thresholds for worsening: ≥2.5-point increase in MDS-UPDRS-II, ≥10% S&E ADL decline, ≥100 mg/day/year LEDD slope, and ≥1-stage Hoehn & Yahr (HY) progression.
Results:
The cohort included 88 patients with PD, classified as faster (n=42) or slower (n=46) progressors. Groups were similar at baseline except that slower progressors had higher MDS-UPDRS-III scores (22.4 vs 16.7, p=0.002). At 48 months, faster progressors showed greater decline in daily function (ΔMDS-UPDRS-II: 5.3 vs 2.8, p=0.01), independence (ΔS&E ADL: −9.3% vs −4.9%, p=0.009), and medication use (LEDD: 423 vs 278 mg/day, p=0.006). Clinically meaningful worsening was more common in faster progressors for MDS-UPDRS-II (81% vs 52%, OR=3.9, p=0.009) and HY progression (62% vs 28%, OR=4.1, p=0.003). Differences in S&E ADL and LEDD slope were not significant.
Conclusions:
An MRI-based classifier identified PD subgroups with distinct trajectories in motor function and treatment needs, highlighting its potential value for future studies of progression.
Disclaimer: Abstracts were not reviewed by Neurology® and do not reflect the views of Neurology® editors or staff.