Functional connectivity-based searchlight multivariate pattern analysis for discriminating Parkinson's disease patients and predicting clinical variables.

OBJECTIVES

Disruptions in brain functional connectivity (FC) have been verified to be one of the characteristics of Parkinson's disease (PD) and PD patients with mild cognitive impairment (MCI). This study aims to differentiate PD patients from healthy controls (HCs), distinguish PD-MCI from PD patients with normal cognition (PD-NC), and evaluate the extent to which connectivity-based searchlight multivariate pattern analysis (CBS-MVPA) could provide predictive information on clinical severity.

METHODS

A cohort of 261 participants from the Parkinson's Progression Markers Initiative (PPMI) was included (98 PD-MCI patients, 98 PD-NC patients, and 65 HCs). Resting-state functional magnetic resonance imaging was used to examine whole-brain FC subnetworks. CBS-MVPA, a novel unbiased whole-brain analysis capable of capturing distributed predictive signals, was applied to identify subnetworks that contributed to group discrimination, and to assess their predictive effects on clinical measures.

RESULT: CBS-MVPA identified 20 FC subnetworks that reliably differentiated PD patients from HCs. Classification accuracy was moderate for PD vs HC (71.25 % to 75.48 %) and modest for PD-MCI vs PD-NC (57.21 % to 62.74 %). Specifically, four subnetworks centered on the posterior cingulate cortex, hippocampus, pallidum, and central postcentral gyrus showed modest predictive value for motor severity (MDS-UPDRS I-III scores). Additional subnetworks involving the paracentral lobule, lingual gyrus, and medial orbital superior frontal gyrus showed modest associations with cognitive performance. It should be noted, however, that this accuracy is only modestly above chance level, reflecting the complexity and heterogeneity of PD.

CONCLUSION

This study introduces a novel approach for identifying distributed FC subnetworks patterns relevant to PD and PD-MCI. While predictive performance was modest, the detected subnetworks nonetheless contained measurable predictive signals related to both diagnostic status and clinical severity, highlighting the potential utility of whole-brain multivariate FC analyses for future mechanistic research and therapeutic interventions in PD.