Abstract:
Parkinson’s disease mild cognitive impairment (PD-MCI), which is one of the major risk factors for dementia, is present in 26.7% of PD patients. In this study, we classified PD-MCI, cognitively normal Parkinson’s disease (PD-CN) and healthy control (HC) groups based on multimodal magnetic resonance imaging (MRI) using machine learning methods. We also investigated time dependent changes in PD-MCI patients through a longitudinal study. 33 PD-MCI, 27 PD-CN and 17 HC participated in this study. The participants were diagnosed by neurologists according to the neuropsychological test scores and physical examination results. MRI data was obtained at a 3T Philips clinical MR scanner using a 32-channel head coil. Mean cerebral blood flow (CBF), arterial blood volume (aBV) and bolus arrival time (BAT) maps obtained fromarterialspinlabelingMRI(ASL-MRI),fractionalanisotropy(FA)andmeandiffusivity (MD) maps obtained from diffusion tensor imaging (DTI), and metabolite peak ratios obtained from proton MR spectroscopic imaging (1H-MRSI) at various brain regions were used as features. Various machine learning methods were employed with appropriate hyperparameters. Random forest recursive feature elimination (RF-RFE) technique was used for feature selection. For longitudinal analysis, linear mixed effects model was utilized with age, education, gender, visuospatial disorder status, and genotype as covariants. The best classification accuracies were 77% for PD-MCI versus HC,71%forPD-MCIversusPD-CN,and86%forPD-CNversusHC.Machinelearning based on multimodal MRI might be helpful in early diagnosis of PD-MCI. Reduced aBV and FA, and higher MD values were observed in time in PD-MCI. Future studies will aim to improve the classification of PD-MCI in a larger patient cohort.|Keywords : Parkinson’s disease, mild cognitive impairment, multimodal MRI, machine learning, lineer mixed effects model.
