Research On Dyskinesia In Parkinson’s Disease Assessment Method Based On Motion Video Feature Analysis

As the core symptom of Parkinson’s disease(PD)movement disorder,bradykinesia is a major basis for clinical screening of patients with early PD.At present,PD cannot be completely cured and affects the daily actions of patients.Therefore,early diagnosis is of great value for the recovery of motor function and follow-up treatment of patients.The clinical screening of PD is through manual examination by specialist doctors,which is time-consuming and depends on the personal experience of doctors.The assessment results are characterized by large individual differences and high subjective errors.Therefore,it is of great significance to seek an objective,efficient and quantifiable automatic assessment technology for PD bradykinesia.This thesis has conducted in-depth research on relevant research and found that the existing research has problems such as relying on professional equipment,single assessment task,few samples and low assessment accuracy.To address the above issues,this thesis proposes a new solution for the assessment of bradykinesia in PD,the main contents include:(1)A bradykinesia assessment scheme based on motion video feature analysis of PD patients is proposed.The new scheme can collect samples through ordinary smart phones,which solves the problem of relying on professional equipment for data collection and reduces the difficulty of collection.And there is no need for PD patients to go to the hospital outpatient clinic or receive professional guidance,and it has a wide range of applications.(2)A method for extracting the bradykinesia feature is proposed,which effectively alleviates the problem of small medical research samples.The method extracts reasonable bradykinesia feature data for training the classification model,and then realizes the assessment of bradykinesia,which reduces the dependence of video classification task training on a large number of datasets.At the same time,the feature extraction method effectively overcomes the shortcomings that it is difficult to unify the models of different assessment tasks and provides a new idea for the unification of the model of multitask assessment of bradykinesia.(3)An improved time series two-stream model is proposed.Through the construction of the twostream structure,the model integrates and analyzes the information of different dimensions,improves the feature analysis and feature fusion capabilities of the model,and can better correlate the feature data with the patient’s bradykinesia,which improves the overall performance of the model.In addition,this thesis improves the low resolution of ordinary video through the video frame interpolation algorithm,thereby improving the data quality and providing reliable data support for research.(4)A classification model based on improved attention is proposed.This model introduces an improved attention based on the time series two-stream model.Since the time feature has important information in the assessment of bradykinesia,the introduction of attention has effectively improved the processing ability of the network’s time feature,thus improving the accuracy of the model.In addition,this thesis improves the sample imbalance problem through the reweighting method,thereby improving the robustness of the model.In this thesis,a bradykinesia assessment scheme based on motion videos is proposed,and a solution based on feature extraction method and deep learning model is established,which effectively solves the problems in existing research,which has reference significance for related research.