Data-Driven Based Approach to Aid Parkinson's Disease Diagnosis.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Neurology

Background:

• Parkinson's disease (PD) diagnosis relies on clinical symptoms, often leading to delayed detection.
• Objective diagnostic tools are needed to improve early and accurate PD identification.

Purpose of the Study:

• To develop and validate a machine learning methodology for diagnosing Parkinson's disease using vertical Ground Reaction Forces (vGRFs) during gait.
• To assess the methodology's ability to differentiate PD patients from healthy controls and individuals with other neurodegenerative diseases.

Main Methods:

• Utilized a four-step process: data pre-processing, feature extraction/selection (Random Forest wrapper), classification (supervised and unsupervised methods), and performance evaluation.
• Employed supervised classifiers (K-NN, DT, RF, NB, SVM) and unsupervised methods (K-means, GMM).
• Validated using an online dataset (93 PD patients, 72 healthy controls) and an additional dataset with Amyotrophic Lateral Sclerosis (ALS) and Huntington's disease (HD) patients.

Main Results:

• The methodology demonstrated high accuracy in differentiating PD subjects from healthy individuals.
• Achieved effective discrimination between PD patients and those with other neurodegenerative diseases (ALS, HD).
• Performance metrics included accuracy, precision, recall, and F-measure, evaluated via leave-one-out cross-validation.

Conclusions:

• The proposed machine learning approach using vGRFs is a promising tool for accurate Parkinson's disease diagnosis.
• This methodology offers potential for early detection and differential diagnosis of PD.
• vGRF analysis provides objective, quantifiable data for neurological disorder assessment.