Quantitative Evaluation Method of Stroke Association Based on Multidimensional Gait Parameters by Using Machine Learning

CHENG WANG ¹ Zhou Long ¹ Xiangdong Wang ² You-Qi Kong ³ Li-Chun Zhou ³ Weihua Jia ⁴ Pei Li ³ Jing Wang ³ Tian Tian ^3*

• ¹ Jinan Zhougke Ubiquitous-Intelligent Institute of Computing Technology, , 250000, China, Jinan, China
• ² Institute of Computing Technology, Chinese Academy of Sciences (CAS), Beijing, Beijing Municipality, China
• ³ Beijing Chaoyang Hospital, Capital Medical University, Beijing, Beijing Municipality, China
• ⁴ Beijing Shijingshan Hospital, Capital Medical University, Beijing, Beijing Municipality, China

Objective: NIHSS for stroke is widely used in clinical, but it's complex and subjective. The purpose of the study is to present a quantitative evaluation method of stroke association based on multidimensional gait parameters by using machine learning.Methods: 39 ischemic stroke patients with hemiplegia were selected as the stroke group and 187 healthy adults from the community as the control group. Gaitboter system was used for gait analysis. Through the labeling of stroke patients by clinicians with NIHSS score, all gait parameters obtained were used to select appropriate gait parameters. By using machine learning algorithm, a discriminant model and a hierarchical model were trained.The discriminant model was used to distinguish between healthy people and stroke patients. The overall detection accuracy of the model based on KNN, SVM and Randomforest algorithms is 92.86%, 92.86% and 90.00% respectively. The hierarchical model was used to judge the severity of stroke in stroke patients. The model based on Randomforest, SVM and AdaBoost algorithm had an overall detection accuracy of 71.43%, 85.71% and 85.71% respectively.The proposed stroke association quantitative evaluation method based on multidimensional gait parameters has the characteristics of high accuracy, objectivity, and quantification.