Integration of fNIRS and Machine Learning for Identifying Parkinson's Disease

Parkinson's disease (PD) is a neurodegenerative disorder where early diagnosis is crucial for effective management. However, current diagnostic methods are often invasive or delayed, hindering early intervention. This study evaluates the effectiveness of combining functional near-infrared spectroscopy (fNIRS) with machine learning to distinguish individuals with PD from age-matched controls.
Data were collected using fNIRS from 28 people with PD and 32 age-matched controls while performing the Timed Up and Go (TUG) test under three conditions: simple TUG, cognitive dual-task TUG, and motor dual-task TUG. Changes in cerebral blood oxygenation in the prefrontal cortex (PFC) were analysed using four machine learning models: Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest (RF), and Extreme Gradient Boosting (XGB), along with statistical analyses. Two feature selection models identified key features and channels for differentiating PD from controls.
The SVM model achieved the highest accuracy (0.85
0.35) in distinguishing PD from CG. Feature selection and statistical analysis showed that dual-task activities were more effective than simple tasks for distinguishing PD from CG. Specific PFC subregions exhibited distinct activation patterns, which could serve as potential biomarkers for PD detection. Combining fNIRS with machine learning shows promise for PD diagnosis, with dual-task activities enhancing accuracy. Further investigation into PFC subregion behaviour could reveal stronger biomarkers.