Clinical responses to dopamine replacement therapy for individuals with Parkinson’s disease (PD) are often difficult to predict. We characterized changes in MDS-UPDRS motor factor scores resulting from a short-duration L-Dopa response (SDR), and investigated how the inter-subject clinical differences could be predicted from motor cortical magnetoencephalography (MEG). MDS-UPDRS motor factor scores and resting-state MEG recordings were collected during SDR from twenty individuals with a PD diagnosis. We used a novel subject-specific strategy based on linear support vector machines to quantify motor cortical oscillatory frequency profiles that best predicted medication state. Motor cortical profiles differed substantially across individuals and showed consistency across multiple data folds. There was a linear relationship between classification accuracy and SDR of lower limb bradykinesia, although this relationship did not persist after multiple comparison correction, suggesting that combinations of spectral power features alone are insufficient to predict clinical state. Factor score analysis of therapeutic response and novel subject-specific machine learning approaches based on subject-specific neuroimaging provide tools to predict outcomes of therapies for PD.
Bibliographical noteFunding Information:
The research was funded by the King Fahad Medical City, the National Science Foundation (00039202 to EP), and the National Institutes of Health (R01-NS094206 and P50-NS098573).
© Copyright © 2021 Peña, Mohammad, Almohammed, AlOtaibi, Nahrir, Khan, Poghosyan, Johnson and Bajwa.
- Parkinson’s disease
- machine learning
- motor cortex
- short duration L-Dopa response