A task-aligned multimodal machine learning framework for studying working memory dysfunction in Parkinson's disease

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Cognitive impairment is a significant non-motor symptom in Parkinson's disease (PD), impacting daily function.
• Current diagnostic methods for PD-related cognitive deficits rely on subjective assessments and lack temporal resolution.
• Objective, scalable methods are needed to detect subtle cognitive dysfunction in early PD.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) pipeline for detecting early cognitive impairment in Parkinson's disease (PD).
• To leverage neurophysiological signals (EEG and pupillometry) recorded during a cognitive task for objective assessment.
• To identify subtle cognitive markers of PD using ML that are not apparent through standard evaluations.

Main Methods:

• Recorded electroencephalography (EEG) and pupillometry data from 35 PD patients and 33 healthy controls (HC) during a visual change detection working memory task.
• Extracted 108 features using a custom toolbox and applied principal component analysis (PCA) for dimensionality reduction.
• Trained a support vector machine with radial basis function kernel (SVM-RBF) classifier, optimizing feature selection using an automated elbow method and recursive elimination (RE).

Main Results:

• The ML pipeline identified optimal feature sets, with a model using 14 principal components (PCs) and 7 PCA-weighted features achieving 71% accuracy and an F1 score of 0.701 on training data.
• The final model demonstrated a classification accuracy of 63% and an F1 score of 0.626 on a hold-out dataset.
• The study framework supports post-hoc feature contribution mapping for mechanistic investigation of cognitive dysfunction in PD.

Conclusions:

• A task-aligned ML pipeline shows potential for uncovering early, objective cognitive markers in Parkinson's disease (PD).
• This approach offers a scalable and objective alternative to current subjective assessments for cognitive dysfunction in PD.
• The developed framework facilitates mechanistic understanding of cognitive deficits by linking neurophysiological features to task performance.