A wearable sensor-based biofeedback system to predict the freezing of gait (fog) in parkinson’s patients

Abstract

Freezing of Gait (FoG) is a disabling condition in Parkinson’s patients, which could cause their motion to be temporarily hindered. Wearable sensors have been commonly used to detect FoG, but this will be of limited use since the patient cannot be notified prior to the FoG event. In this research, a system that can be used to predict the FoG in Parkinson’s patients well ahead of the event was developed. This is a multimodal sensor system that consists of Plantar pressure sensors placed in the insole of the shoes and Inertial Measurement Units placed on the shank of each leg. 3 machine learning algorithms were developed: K means, K Nearest Neighbors, and Support Vector Machines. Out of the 3 algorithms used, the KNN algorithm was determined to have the highest FoG prediction sensitivity of 0.88. The Pressure sensors captured more swift variations in Gait during freezing compared to the IMUs. This system could identify FoG 4.42s ahead of the event and used an optimal time window of 1s. This system could be developed to yield more accurate predictions, and this could be integrated with an auditory biofeedback system to warn the patient of FoG, to minimize its occurrence.