Cerebral hemodynamics and functional connectivity in the stroop test in Parkinson's disease patients: A machine learning approach to fNIRS features.

BACKGROUND

Cognitive impairment is a core non-motor feature of Parkinson's disease (PD). This study aimed to: (1) assess PD patients' performance on the color-word Stroop task and characterize task-related neural activity; (2) develop PD diagnostic models using task-based functional near-infrared spectroscopy (fNIRS) features to link neuroimaging mechanisms with clinical translation.

METHODS

Sixty-one participants (29 PD patients, 32 healthy controls [HC]) completed the Stroop task during fNIRS recording. Cerebral hemodynamic and brain network analyses were performed, and a two-way ANCOVA examined group (PD vs. HC) and task (congruent vs. incongruent stimuli) effects, adjusted for Mini-Mental State Examination (MMSE) and Montreal Cognitive Assessment (MoCA). Machine learning frameworks were applied to construct diagnostic models using task-based fNIRS features.

RESULTS

PD patients showed significantly higher omission rates than HC across both Stroop conditions, alongside enhanced dorsolateral prefrontal cortex and frontal eye field activation, and increased prefrontal interhemispheric functional connectivity. Logistic regression outperformed other models, achieving comparable accuracy with fewer features.

CONCLUSION

These findings advance our understanding of PD-related cognitive impairment and provide a framework for developing non-invasive, objective diagnostic tools.