Recognizing Gait Pattern of Parkinson's Disease Patients Based on Fine-Grained Movement Function Features

Abstract

Parkinson's Disease (PD) is one of the typical movement disorder diseases, which has a serious impact on the daily lives of elderly people. In this paper, we propose a novel framework for PD gait pattern recognition. The key idea of our approach is to distinguish PD gait patterns from healthy individuals by accurately extracting gait features that indicate three aspects of movement function, i.e., Stability, symmetry and harmony. Concretely, our framework contains three steps: gait phase discrimination, feature extraction and selection and pattern classification. In the first step, we put forward a key event based method to discriminate four gait phases from plantar pressure data. In the second step, based on the gait phases, we extract and select gait features that can indicate stability, symmetry and harmony of movement function. In the third step, we recognize PD gait pattern by employing BP neural network. We evaluate the framework using a real plantar pressure dataset that contains 93 PD patients and 72 healthy individuals. Experimental results demonstrate that our framework outperforms the baseline approach by 32.7% on average in terms of Precision, 42.2% on average in terms of Recall, and 24.0% on average in terms of AUC.