Objective: To develop a machine learning model using MRI data that predicts the likelihood of cognitive impairment in newly diagnosed patients with Parkinson’s disease (PD).

Background: Mild cognitive impairment (MCI) often exists in newly diagnosed PD patients and more than half of patients can develop dementia within 10 years of diagnosis [1]. Therefore, early identification of MCI in PD is essential for proper management of the disease.

Method: We analyzed the baseline  T1-weighted MRI of  35 newly diagnosed PD patients and 72 age-matched healthy controls, with a MoCA score of 26 or higher at entry into the Parkinson’s Progression Marker Initiative (PPMI) [2]. Based on the follow-up MoCA scores at 4-years, PD patients were subsequently classified into two groups: cognitively-intact (MoCA=26 or greater) and cognitively-impaired (MoCA=25 or lower). To predict the likelihood of developing cognitive impairment after 4 years, for each patient’s baseline MRI, gray matter volumes from 48 regions of interest, including frontal, parietal, and subcortical regions, were calculated using Mahalanobis distance (M_GMV) [3]. M_GMV calculates the distance between brain regions in each patient, using a reference distribution from the healthy control brains. Random forest model was then used to create the prediction model for the development of cognitive impairment using the M_GMV. We then calculated the accuracy, sensitivity and specificity of our machine learning model in predicting cognitive-impairment.

Results: Our PD cohort had a mean age of 59.8 years; 71% (n=25) males; mean baseline MoCA score of 28.4. The mean age of the age-matched controls was 60.2 years; 66% (n=48) males; average baseline MoCA score of 28.3. At 4 years, 22 PD patients remained cognitively-intact (mean MoCA = 27.5) and 13 were cognitively-impaired (mean MoCA =22.2). Using baseline MRI, our machine learning model predicted cognitively-impaired patients with an accuracy of 90.9% having 80% specificity and 100% sensitivity.

Conclusion: Our study proposes a machine learning model that utilizes baseline MRI data to distinguish between PD patients with cognitive impairment and those without, as defined by their follow-up MoCA scores. Further investigations using larger samples along with more detailed cognitive assessment tools will be necessary to confirm the accuracy and improve the generalizability of this model.

References: 1- Cammisuli DM., et al. Parkinson’s Disease-Mild Cognitive Impairment (PD-MCI): A Useful Summary of Update Knowledge. Front Aging Neurosci. 2019. 8(11):303.
2- Marek, K., et al., The Parkinson’s progression markers initiative (PPMI) – establishing a PD biomarker cohort. Ann Clin Transl Neurol, 2018. 5(12): p. 1460-1477.
3- Mahalanobis, P.C. On the generalized distance in statistics. . in Proceedings of the National Institute of Sciences. 1936. Calcutta.

« Back to 2023 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/leveraging-mri-and-machine-learning-for-early-detection-of-cognitive-impairment-in-parkinsons-disease/