Research On Parkinson’s Disease Assessment Based On Multi-Activity And Multi-Feature Fusion Modeling

Parkinson’s disease is currently the second largest neurodegenerative disease,mainly manifested as tremor,bradykinesia,frozen gait and other motor symptoms,which seriously affect the quality of life of patients.A large number of studies believe that correct treatment can delay the development of the disease.Neurologists comprehensively evaluate the patient’s condition and give appropriate treatment plans based on the Parkinson’s Comprehensive Rating Scale(UPDRS)combined with the patient’s field performance and recall description.This method cannot monitor patients in an out-of-hospital environment,and it is difficult for doctors to understand the changes in the patient’s condition.In recent years,there have been a large number of studies on the detection or quantification of patients’ conditions through wearable devices,but there are few works to evaluate the overall condition of patients through a single wearable device.To address this issue,the work and contributions of this paper are as follows:(1)Aiming at the difficulty in assessing the overall condition of patients,a multiactivity fusion modeling strategy is proposed,and through the cross-fusion of different activity segments,the amount of data is greatly increased,and the continuity of the feature space is increased;(2)According to the advantages and disadvantages of traditional machine learning and deep learning methods and the characteristics of the data itself,a multi-feature fusion method is proposed,which is instant domain features + frequency domain features +patient basic information features + CNN feature map,which can complement each other.Inferior,more complete expression of data information;(3)A 2D-CNN network structure is designed according to the characteristics of the data itself,which can learn more robust features while reducing the amount of parameters,and introduces a label smoothing loss to reduce the model’s interference from label noise.;(4)Six experimental schemes are designed to verify the effectiveness of multiactivity fusion modeling,multi-feature fusion,custom 2D-CNN architecture and label smoothing.And compared with 8 previous works,the results prove the effectiveness of our proposed multi-activity multi-feature fusion CNN algorithm;(5)Combined with previous research work,a set of easy-to-implement PD patient activity data collection scheme was designed for data collection.And over a 16-month period,18 activity data were collected from 40 Parkinson’s patients and 40 healthy people.