Abstract:
Exoskeletons that are activated by the muscles and brain have been suggested to train the
motor skills of stroke victims. Training can incorporate task variety since an exoskeleton allows
for the execution of various movement types.Differentiating between movement types at the
same time from brain activity is challenging, but it might be accessible from residual muscular
activity that many patients retain regain.This study examines whether forearm EMG from five
stroke patients can be used to decode seven distinct motion classes of the hand and forearm. This
study evaluates classifiers like Support vector machine (SVM), Lineardiscriminant analysis
(LDA) and K nearest neighbor (KNN). It investigated the relation of motor impairment with
classification accuracy by the classifiers. During the following motion classes: Supination,
Pronation, Hand Close, Hand Open, Wrist Extension, Wrist Flexion, and Pich, five surface EMG
channels were recorded.Every motion was performed by patients three times repetition over the
course of eight weeks.Support vector machines, k nearest neighbor, and linear discriminant
analysis were used to classify decoding of hand moments for stroke patients. On average,73.69 ±
6.39%SVM,71.6 ± 5.09% KNNand 50±4.56 LDA of the movements were correctly
classified.Seven motion classes were demonstrated to be decoded from residual EMG, and SVM
proved to be the most effective classification method when compared to the other three
classifiers for decoding of hand motion for stroke patients.The results of this study may have
implications for the development of exoskeletons, suits, or gadgets, that are powered by EMG
signals. These devices might be utilized in the comfort of the patient's home to assist stroke
sufferers with their training activities. Therefore, the findings of this study may assist in
improving the effectiveness and accessibility of these useful tools for stroke survivors.
