Abstract
Asymmetry of symptom onset in Parkinson's disease (PD) is strongly linked to differential diagnosis, progression of disease, and clinical manifestation, suggesting its importance in terms of specifying a therapeutic strategy for each individual patient. To scrutinize the predictive value of this consequential clinical phenomenon as a neuromarker supporting a personalized therapeutic approach, we modeled symptom-side predominance at disease onset based on brain morphology assessed with magnetic resonance (MR) images by utilizing machine learning classification. The integration of multimodal MR imaging data into a multivariate statistical model led to predict left- and right-sided symptom onset with an above-chance accuracy of 96%. By absolute numbers, all but one patient were correctly classified. Interestingly, mainly hippocampal morphology supports this prediction. Considering a different disease formation of this single outlier and the strikingly high classification, this approach proves a reliable predictive model for symptom-side diagnostics in PD. In brief, this work hints toward individualized disease-modifying therapies rather than symptom-alleviating treatments.