Research On Automatic Rating Method Of Motor Examination In UPDRS

Parkinson’s disease shows an increasing incidence in recent years,and how to diagnose Parkinson’s disease quickly and accurately becomes an important research direction.Currently,the common diagnosis method of Parkinson’s disease is scale evaluation,where the Unified Parkinson’s Assessment Scale(UPDRS)is one of the most widely used scales.However,scale evaluation method is instability in objectivity and costs much time and effort.As a result,based on motor data collected by wearable sensors,this paper focus on automatic scoring method of UPDRS sports-related indicators,aiming at improve the accuracy and convenience of Parkinson’s disease diagnosis.Firstly,according to the requirements of the motor-evaluation in the 3rd part of UPDRS,combined with clinical needs and algorithm requirements,we designs a set of paradigmatic movements correspond to the UPDRS motor-evaluation items.Paradigm movements include 7movements divided into 4 categories: tremor test,balance test,upper/lower limb flexibility test and gait test.These movements are group into smooth motion and repetitive motion by their own traits for further processing.A total of 110 valid data of Parkinson’s patients with Hoehn-Yahr grades 1-4and healthy controls are collected in cooperative hospital,and a data set management tool was developed to automate data export and data set construction.After that,motion data processing and automatic scoring model construction are performed.For smooth motion which includes tremor tests and balance tests,data pre-process contains denoising and valid channel selection.Then,the motion features are extracted from the time domain and frequency domain.Considering the long time range of smooth motion,we proposes a method to extract the parameter sequence from the segmented signal and combine the parameter sequence to increase information quality.When building an automatic scoring model later,we develop a SVM classification model based on PSO thyperparameters-optimize.Using the segmental parameter sequence combination as input,we obtain an accuracy of 0.90 in the tremor test and 0.87 in the balance test.For the repetitive motion including upper limb/lower limb flexibility test and gait test,denoising and coordinate calibration are applied in data pre-processing.Afterwards,paradigm motions are split into several action units,and the completion time,amplitude,speed,etc.of the action units are extracted as features.As the large amount of features in repetitive motions,feature engineering and PCA dimensionality reduction are used for feature screening.Using the SVM-PSO model,the accuracy of the upper limb flexibility test and the lower limb flexibility test can reach 0.83 and 0.81.The accuracy of the gait test using PCA dimensionality reduction method is improved,and the optimal accuracy is 0.78.It shows that the automatic scoring method in this paper has certain reference value in clinical application.