Objective: To externally validate a previously published machine learning (ML) classifier for predicting motor symptom progression in Parkinson’s disease (PD) using an independent clinical dataset.

Background: In a prior study using the Parkinson’s Progression Markers Initiative (PPMI) dataset [1], we developed an ML model that classified PD patients as slower or faster progressors based on changes in OFF-medication MDS-UPDRS-III scores from baseline to 48 months. Using baseline MRI-derived multivariate gray matter volume (M_GMV), age, sex, and MDS-UPDRS-III scores as input features, the model achieved an accuracy of 89% [2]. Evaluating the model’s generalizability in independent clinical cohorts is essential for assessing its potential clinical utility.

Method: MRI and MDS‑UPDRS‑III scores were retrospectively collected from 11 PD patients at the Cleveland Clinic Foundation (CCF), with data derived from standard clinical visits closest to baseline and approximately 48 months of follow‑up. T1‑weighted MRI data were processed using the Computational Anatomy Toolbox to extract gray matter volumes from motor‑related ROIs defined in the prior study [2]. Scanner effects were harmonized with NeuroHarmonize [3], and M_GMV distance was computed using Mahalanobis distance [4]. The PPMI-derived ML classifier [2] was applied directly to the CCF dataset without retraining, using age, sex, baseline MDS-UPDRS-III scores, and M_GMV as input features. Model performance was evaluated using sensitivity, specificity, and balanced accuracy.

Results:

The CCF cohort had a mean age of 69.9 ± 8.9 years, with 64% male participants and a mean baseline MDS-UPDRS-III score of 22.27 ± 8.07. Among the 11 PD patients, 3 were classified as faster progressors and 8 as slower progressors. Sensitivity for detecting fast progressors was 66.7% (2/3), while specificity for identifying slow progressors was 87.5% (7/8), yielding a balanced accuracy of 77.1%. The confusion matrix summarizing prediction performance on the CCF dataset is shown in Figure 1.

Conclusion: Our results provide preliminary external validation of the PPMI-derived ML classifier, demonstrating its potential to identify PD patients at risk for faster motor progression in a clinical setting. Future studies with larger cohorts are needed to confirm its clinical utility.

ML classifier confusion matrix in CCF cohort

References: 1. Marek K, et al. (2018) The Parkinson’s progression markers initiative (PPMI) – establishing a PD biomarker cohort. Ann Clin Transl Neurol 5:1460-1477. https://doi.org/10.1002/acn3.644
2. Vijayakumari AA, Fernandez HH, Walter BL (2023) MRI-based multivariate gray matter volumetric distance for predicting motor symptom progression in Parkinson’s disease. Scientific Reports 13:17704. https://doi.org/10.1038/s41598-023-44322-0
3. Pomponio R, et al. (2020) Harmonization of large MRI datasets for the analysis of brain imaging patterns throughout the lifespan. Neuroimage 208:116450. https://doi.org/10.1016/j.neuroimage.2019.116450
4. Vijayakumari AA, Mandava N, Hogue O, Fernandez HH, Walter BL (2023) A novel MRI-based volumetric index for monitoring the motor symptoms in Parkinson’s disease. J Neurol Sci 453:120813. https://doi.org/10.1016/j.jns.2023.120813

« Back to 2026 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/external-validation-of-a-machine-learning-classifier-for-motor-progression-in-parkinsons-disease/