Auxiliary Diagnosis System For Parkinson’s Disease Based On Gait Analysis

Parkinson’s disease is a common neurodegenerative disease with a high incidence and large number of patients in the elderly.The diagnosis and treatment of Parkinson’s disease in China is still in a state of low accuracy rate and high delayed diagnosis rate.On the one hand,it is difficult to distinguish Parkinson’s disease due to the lack of specific markers,and on the other hand,the existing diagnosis and treatment methods lack objective and quantitative evaluation standards.Therefore,it is necessary to improve the diagnostic accuracy through auxiliary diagnosis.Gait analysis,as a quantitative evaluation method,has gradually attracted attention in the auxiliary diagnosis of Parkinson’s disease.The reason is that most patients with Parkinson’s disease have gait disorders,which are manifested in different degrees at different stages of disease.Therefore,gait analysis has high clinical value in the auxiliary diagnosis.Compared with existing quantitative gait analysis techniques and related studies,it is found that the equipment cost of high-precision gait analysis method is too high,while the low cost gait analysis studies mostly use a single sensor to collect gait data for analysis,and the assessment accuracy is still limited.In view of the aforementioned difficulty in balancing accuracy and cost,a low-cost and accurate Parkinson’s auxiliary diagnosis system is designed and implemented to help clinicians improve the efficiency and accuracy of disease diagnosis and benefit more potential patients with timely diagnosis and treatment.Starting with gait analysis,the system adopts a three-layer architecture of "cloud-edge-device" and realizes data collection,comprehensive analysis and management display functions.In the data collection phase,multimodal gait data of patients were collected in a non-invasive manner.In the comprehensive analysis stage,the complementarity between different modal data is fully mined to improve the accuracy of gait quantitative evaluation.In the management presentation stage,the evaluation results will be visualized and managed online,which is convenient for doctors to refer to diagnosis and evaluate the disease progression of patients.Specifically,this paper contains three main tasks:(1)Design a multimodal gait information acquisition program combining inertial sensor and monocular camera.Multiple algorithms were designed to extract gait disorder features such as step length from the collected data for quantification to assist in diagnosis.The selection of gait features was based on the clinical diagnosis experience of the partner doctors and the existing research results.(2)In order to further improve the intelligent degree of the system,an auxiliary diagnosis model based on multimodal deep learning is designed and implemented.Firstly,a multimodal dataset containing two modes of limb kinematics data and human side gait image was sorted out,and then the model was trained by the dataset.Finally,the original data of the two modes of kinematics and image could be input into the model to automatically extract features and give auxiliary diagnosis results.(3)Design and develop a cloud platform for doctors to visually view the results of gait analysis and manage patient treatment records,and display the quantized values of gait characteristics and model prediction results obtained in Work 1 and 2 through Web pages.The test results show that the system meets the expectation and can play a good role in auxiliary diagnosis.The gait quantization function was accurate in feature extraction of cooperative hospitals during trial operation.Multimodal model has high performance in gait data recognition of Parkinson’s disease patients.The cloud platform passed the function test,and the performance test met the requirements of the hospital scenario.In conclusion,the developed Parkinson’s auxiliary diagnosis system has practical value and good application prospect.