Research On Classification Algorithms Of Surface EMG Signals For Interventional Surgery Robot Control

In recent years,the number of cardiovascular diseases has continued to increase,and the most effective method for it is interventional surgery.However,the current gap in interventional surgery doctors is relatively large,and the operating environment affects the health of doctors.The emergence of interventional surgery robots is urgently needed.Accurate,safe,and stable control of surgical robots is a huge challenge,so we proposed a new control method for interventional surgical robots based on surface EMG signals.The classification and of surface EMG signals is one of the most critical steps.This paper is a research on the classification algorithm of surface EMG signals for interventional surgery robot control.The main research contents are as follows:In the first research stage,significant differences were analyzed for the extraction of relevant features of surface EMG signals.Nine volunteers were recruited,four movements were set to collect data,and the signal was preprocessed and feature extracted.The two features of average electromyogram and root mean square were selected,and one-way analysis of variance was used to analyze the significant differences of different movements of different muscles.The difference between different actions is weak,and there is similarity between the two characteristic patterns.In this part,our innovation is proposed a novel control method for interventional surgery robot based on surface EMG signals.In the second research stage,the public standard dataset was used for classification experiments,and good results were obtained.A series of pre-processing operations were performed on the data to extract seven common features and two new features,ASS and MSR.The designed random forest algorithm performs classification experiments on a single feature and multi-features,and the best classification accuracy is 92.94%.And the experimental results are analyzed to verify the stability of the model from evaluation indicators of the confusion matrix,ROC curve and PR curve.In this part,our innovation is designed a random forest algorithm to obtain good classification performance on the standard dataset.Finally,we cooperated with the Shenzhen Hospital of the University of Chinese Academy of Sciences to collect data from experienced doctors for in-depth research.Seven cardiologists were recruited to collect three multimodal data of surface myoelectric signal,electromagnetic position signal,and tactile force signal during the operation of the guidewire to explore the operation skills of clinical hand movement.After signal processing and feature extraction,nineteen features were extracted and classified using the designed random forest algorithm.Experiments show that there is a positive correlation between muscle activity and the contact force of the blood vessel wall,and there are also significant differences between different muscles and different actions.The greater the number of features used,the better the classification performance,with a maximum accuracy of 94.11%.Our innovation in this part is the collection of a multimodal dataset of professional doctors.There is no precedent in the world.A classification and recognition framework based on random forest algorithm is proposed,and a good classification performance is obtained.This research has a high scientific significance and potential application value.It innovatively proposed a control method for interventional surgery robot based on surface EMG signals and proposed a classification framework for surface EMG signals based on random forest algorithm,which can be used for remote surgical control and achieve force feedback during surgery.