A dual-branch model for diagnosis of Parkinson’s disease based on the independent and joint features of the left and right gait

摘要

Clinical diagnosis of Parkingson’s disease (PD) requires the physician to assess the patient’s gait and other symptoms. A dual-branch model is proposed in this paper as an objective diagnostic tool to diagnose PD automatically. In this research, the joint features and independent features of left and right gait are fused innovatively. Convolutional neural network (CNN) and long short-term memory network (LSTM) are used to extract the spatial and temporal characteristics of sensors respectively. After the independent features extracted from the branches are collapsed, LSTM is used to incorporate the joint features between the left and right gait. Compared with other methods, the proposed model can learn the correlation between the two feet and extract higher discriminative features to effectively improve the accuracy of Parkinson detection. The model shows the state-of-the-art performance for the public dataset, with the accuracy, sensitivity, and specificity being 99.22%, 100%, and 98.04%, respectively. A simple, fast, and objective method proposed in this paper was believed to improve diagnostic performance.