Classification of Parkinson’s Disease-Associated Gait Patterns

Parkinson’s disease (PD) is a progressive neurological disorder that affects movement of millions of people worldwide. Many methods have been developed to identify and diagnose PD in patients. However, most of these approaches require extensive setup and involve costly equipment such as using depth cameras or devices worn on the body. In this study, we investigate the use of vertical ground reaction force (VGRF) sensor readings to classify PD subjects from non-PD subjects. This presents a low-cost and straightforward approach to identify PD through the gait characteristics associated with PD. By fusing together data points from individual sensors to create an ensemble and combining it with a deep short network capable of accurately identifying the gait characteristics of PD from the sensory input, we present a novel approach to classify gait characteristic of PD which is feasible in a clinical setting. We tested our model on a public dataset from PhysioNet which consists of VGRF sensor readings of PD and non-PD patients. Preprocessing was done by extracting out several meaningful features from the raw data when was then split and normalized. Classification was done using a multilayer feed-forward artificial neural network. Experimental results conclude that this model achieved 84.78% accuracy on the PhysioNet dataset, which is a significant improvement over various state-of-the-art models.