Classification of Parkinson's disease motor phenotype: a machine learning approach

Area of Science:

• Neuroscience
• Clinical Neurology
• Biomedical Engineering

Background:

• Parkinson's disease (PD) is a neurodegenerative disorder characterized by motor symptoms.
• Distinct motor phenotypes, such as tremor-dominant (TD) and postural instability and gait difficulty (PIGD), exist within PD.
• Cortical activity alterations may underlie these phenotypic differences.

Purpose of the Study:

• To investigate and compare electroencephalography (EEG) spectral activity between PD phenotypes (TD and PIGD) and healthy controls.
• To assess the impact of medication status (OFF and ON) on cortical activity in different PD phenotypes.
• To evaluate the efficacy of machine learning in classifying PD phenotypes based on EEG data.

Main Methods:

• Twenty-four individuals with PD (OFF and ON medication) and twelve healthy controls underwent resting-state and hand-movement EEG.
• Spectral ratio analysis was performed on EEG data.
• A machine learning approach using 35 EEG-derived attributes, including random forest and random tree algorithms, was employed for classification.

Main Results:

• Cortical activity slowing was observed in PD patients during the OFF medication state.
• Individuals with the TD phenotype showed cortical slowing at rest, while those with the PIGD phenotype exhibited it during hand movement.
• No significant differences in cortical activity between PD phenotypes were found during the ON medication state.
• Random forest machine learning demonstrated high accuracy in distinguishing between PD phenotypes.

Conclusions:

• Parkinson's disease phenotypes are associated with distinct patterns of cortical activity, particularly when medication is withdrawn.
• EEG spectral analysis, combined with machine learning, can effectively differentiate between PD motor phenotypes.
• Phenotypic characteristics may represent a significant factor influencing cortical activity in Parkinson's disease.