Testing the Validity of a Spatiotemporal Gait Model Using Inertial Measurement Units in Early Parkinson’s Patients

The subtle gait characteristics of early Parkinson’s disease (EPD) patients are currently difficult to detect or require expensive, experimentally demanding testing equipment. The use of machine learning (ML) models in conjunction with inertial measurement unit (IMU) algorithms opens up new possibilities for the assessment of EPD patients. The aim of this study is to measure EPD gait using the IMU algorithm, select gait features using Recursive Feature Elimination (RFE), and classify EPD patients with healthy (HT) older adults using ML on the selected features. Firstly, 10 healthy subjects were recruited and the system parameters were validated using the double gold standard to ensure the reliability of the system. Second, 60 subjects (30 EPD patients and 30 HT elderly) were recruited to wear the system for linear walking activities and to obtain gait parameters. The results show that this system has good reliability, i.e. the best intraclass correlation coefficient (ICC) is between 0.521 and 0.941. The six best features of stride length, stance phase, stance time, swing phase variability, step speed and cadence were selected by REF and classified by decision tree (DT) with a model accuracy of 91.6%, sensitivity and specificity of 91% and 83% respectively, and an ROC value of 0.92. Our results show that the use of the IMU algorithm with precise accuracy can detect subtle gait features and that the use of optimal gait features can well assess patients with EPD, providing a new way to detect EPD.