The purpose of this study is to use whole-body kinematics data during walking to accurately identify motor subtypes of Parkinson’s disease (PD) and parkinsonism.

Two major motor subtypes of PD have been described, Postural Instability and Gait Disorder (PIGD) and Tremor Dominant (TD), which differ in the disease prognosis. Prior studies have explored quantifying motor subtypes via gait analysis, which lacks data for whole-body movements that are required when analyzing an entire gait cycle. In this study, we measure whole-body kinematics while walking, which is then analyzed to classify PIGD vs. TD subtypes using machine learning (ML) models.

The whole-body kinematic marker time series were collected from 57 patients (including 5 patients with atypical parkinsonism) while walking in off medication state using a motion capture system from 2015 to 2017. From all marker kinematics data, we extracted various features beyond gait-related features, including time series and spectral features. Then, ML models were used to classify the extracted features into the corresponding motor subtypes assessed with Movement Disorder Society-Unified Parkinson’s Disease Rating Scale (MDS-UPDRS). 

Compared to the clinically assessed motor subtypes, our model classified whole-body kinematics while walking with a 97.0% F1 score, which significantly outperformed the gait-based analysis which showed a 55.8% F1 score. Both upper and lower body movements were found to be useful indicators to distinguish between PIGD and TD.

Our findings strongly supported our hypothesis that accurately quantifying motor subtypes requires more than gait-related or lower-body movement features that are typically used in previous work. This feature analysis may lead to novel hypotheses for fine-grained upper-body motion phenotypes related to motor subtypes. More importantly, our techniques can help monitor changes in motor subtypes for guiding appropriate intervention through objective quantification of walking movements.