Machine Learning for early detection of motor dysfunction in Parkinson’s disease using 6-axis IMU data

Parkinson's disease (PD) is a progressive neurodegenerative disorder characterized by distinct motor symptoms such as resting tremor, bradykinesia, muscle rigidity, postural instability and gait abnormalities including shuffling steps, shortened stride lengths, and freezing episodes. The gait impairments severely affect mobility and independence and the general quality of life. Early and accurate identification of PD- associated gait impairment is indispensable for optimal timing of intervention and efficacious management. New developments in wearable sensor technologies have allowed for the non-invasive, real-time measurement of biomechanical parameters of daily activities and the objective quantification of PD-associated gait patterns. This study investigates the effectiveness of traditional machine learning (ML) classifiers in differentiating gait characteristics of individuals with PD from healthy controls using 6-axis inertial measurement unit (IMU) data acquired via Moticon OpenGo smart insoles. This research specifically recognizes the significance of IMU-based features like the angles of roll and pitch, which identify key postural and kinematic patterns of importance in the differentiation of PD-associated gait impairment from normal gait patterns. Three- dimensional accelerations, angular velocities, roll, and pitch angles were assessed based on data collected from Moticon OpenGo smart insoles and subjected to supervising ML techniques such as k-nearest neighbours (KNN), Random Forest (RF) and Gradient Boosting (GB) classification. The results demonstrate that these ML classifiers effectively discriminate between PD and healthy gait patterns, with the RF classifier achieving the highest classification accuracy of 91%. This study highlights the significant role of IMU derived biomechanical features, particularly roll and pitch angles, and underscores the potential of integrating ML techniques with wearable sensor technologies for the early detection and continuous monitoring of PD, thus supporting clinical diagnosis, intervention planning and improved patient care.