Volume Alterations in Thalamic Subnuclei in Parkinson's Disease Dementia and Machine Learning-Based Prediction of Diagnosis and Severity

Abstract

Background: Cognitive decline in Parkinson's disease (PD) is associated with pathological alterations within the thalamus. Nevertheless, volumetric changes in the specific subnuclei of the thalamus in PD patients with dementia (PD-D) remain inadequately characterized. Furthermore, the clinical challenges of diagnosing PD-D at an individual level and forecasting the trajectory of cognitive decline persist.

Methods: This study acquired structural magnetic resonance imaging (MRI) data from 60 healthy normal controls (NC), 63 PD patients without dementia (PD-nD), and 57 PD-D patients. The volumes of 25 thalamic subnuclei were quantified using FreeSurfer and a novel thalamic segmentation algorithm. Subsequently, individual PD-D diagnosis and severity prediction of cognitive impairment were performed using support vector machines (SVMs).

Results: Our findings demonstrated atrophy in seven out of 25 left and two out of 25 right thalamic subnuclei in PD-D patients relative to PD-nD patients. When compared to NC subjects, the PD-D group exhibited volume reductions in two left and one right subnuclei, alongside enlargement in several others. Within the PD cohort, the volumes of four left thalamic subnuclei showed a negative correlation with cognitive impairment severity. Machine learning models achieved high accuracy in differentiating PD-nD from NC (89.19%), PD-D from NC (94.29%), and PD-D from PD-nD (83.33%). Moreover, the prediction of Mini-Mental State Examination (MMSE) scores yielded a Pearson correlation coefficient of 0.7568.

Conclusion: Specific thalamic subnuclei undergo atrophy in PD-D, and these morphological changes are linked to cognitive deficit severity. Leveraging these features with machine learning enables accurate individual diagnosis and severity prediction.

Keywords: Parkinson's disease; dementia; machine learning; magnetic resonance imaging.