Quantitative Assessment Of Parkinsonian Motor Symptoms Based On Characteristic Analysis And Machine Learning Method

Parkinson's disease(PD)is one of the most common neurodegenerative diseases,affecting about one percent of people over the age of 60.With the advent of an aging society,the number of PD in China has increased dramatically.The main motor symptoms of PD include tremor,bradykinesia,gait disturbance and rigidity.These symptoms seriously affect the quality of life of patients,but also bring a certain burden on families and medical institutions,so it is very important to diagnose and treat PD timely and effectively.In a traditional clinical evaluation,the severity of PD symptoms is rated against the Unified Parkinson's Disease Rating Scale(UPDRS),which is a subjective assessment in nature and relies mainly on the visual judgment of the neurologists.Therefore,studying on quantitative assessment method for PD can eliminate inconsistencies and interrater assessment disagreements between different evaluators,so as to improve the evaluation and monitoring of PD.Trying to overcome the limitations of the current assessment methods,the quantitative assessment method based on characteristic analysis and machine learning were used to quantify tremor,bradykinesia,and rigidity in this study.The major innovative works includes:1?At present,most tremor quantitative analyses rely on raw signals from the inertial sensors,which ignores the influence of gravity component.In this study,the attitude estimation-based gradient descent algorithm is used to separate the linear acceleration from the accelerometer output.Tremor signal features extracted from the linear accelerations and angular velocities during were fitted to the clinicians'ratings with a multiple regression model..The prediction results show that the proposed quantitative method improves the assessment accuracy(rest tremor r~2=0.95,postural tremor r~2=0.93).2?To overcome the limitation that cannot directly distinguish patients with different grades of UPDRS bradykinesia score,the axis-angle representation approach was employed to obtain the 1-dimensional(1D)combined orientation angle to express the 3D motion.The significant bradykinesia features extracted from the 1-D combined orientation angle were used to train the SVM classification algorithm to acquire an objective score on the bradykinesia severity.Clinical experiments showed that the classification accuracy of the proposed method was95.349%.3?Aiming at the difficulty in measuring and evaluating on the Parkinsonian rigidity,a multi-sensor based mechanical measuring device was designed to capture the motion and biomechanical information of the elbow joint of PD patients in this study.Various quantitative characteristic parameters were extracted based on the mechanical impedance model,and the statistical analysis was carried out to find the optimal quantitative parameter.Clinical experiments showed that the mechanical impedance has the best correlation(Pearson correlation coefficient r=0.87)with the clinical UPDRS scores.In addition,the SVM classifier was trained using the extracted characteristic parameters to classify PD and healthy people.The classification results showed that the proposed SVM classifier has a sensitivity of96%and a specificity of 85.7%,which can contribute to the accurate diagnosis of PD and to the monitoring of therapies to control parkinsonian rigidity.