AUTHOR=Ma Lingyan , Lin Shinuan , Jin Jianing , Wang Zhan , Wang Xuemei , Chen Zhonglue , Ling Yun , Zhang Fei , Ren Kang , Feng Tao 

TITLE=Objective assessment of gait and posture symptoms in Parkinson’s disease using wearable sensors and machine learning

JOURNAL=Frontiers in Aging Neuroscience

VOLUME=Volume 17 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/aging-neuroscience/articles/10.3389/fnagi.2025.1618764

DOI=10.3389/fnagi.2025.1618764

ISSN=1663-4365

ABSTRACT=ObjectiveGait and posture symptoms—such as gait impairments, postural instability, and posture deformations—are common in Parkinson’s disease (PD) and closely linked to falls. Traditional assessments using clinical scales are time-consuming and prone to subjective bias. This study aims to predict the severity of gait and posture symptoms using data collected from wearable sensors during a single laboratory-based walking assessment, providing an objective, efficient, and automated evaluation approach.MethodsSensor-based gait parameters were collected from 225 PD participants (mean age 63.15 ± 10.46 years) through a standardized walking assessment. The dataset was randomly split into a training set (80%) and an independent test set (20%) with balanced age, sex, and PD duration. Two machine learning models—extreme gradient boosting (XGBoost) and support vector machine (SVM)—were trained to predict scores for five gait and posture items (#3.9–3.13) from the MDS-UPDRS III.ResultsXGBoost was chosen as the final model due to its better performance than SVM. Across all five gait and posture items, the models achieved over 80% acceptable accuracy. For items #3.9–#3.11, absolute accuracy surpassed 70%, and macro-F1 scores were above 0.60 in leave-one-out cross-validation (LOOCV). The model’s performance on the independent test set matched LOOCV results, confirming robustness. A total of 35, 35, 30, 30, and 40 gait features were selected for the predictive models of items #3.9–#3.13, respectively. Among these, key features with significant clinical relevance were identified. For example, Effective Trial Duration (R = 0.522, p < 0.001) had a positive correlation, while Shank—Swing RoM—mean (max) (R = −0.629, p < 0.001) had a negative correlation with scores on item #3.10. In addition, 180° Turn—Steps – mean (R = 0.482, p < 0.001) had a positive correlation with scores on item #3.11. These findings align with known clinical manifestations, reinforcing the clinical relevance of the identified gait features.ConclusionThis study demonstrates the feasibility of using wearable sensor data to objectively assess gait and posture symptoms in PD. Though conducted in a clinical setting, the approach may support clinicians through consistent assessments and more frequent monitoring, with potential for future home-based use to enable longitudinal symptom tracking.