Anxiety Status Assessment In Upper Extremity Rehabilitation Training

In recent years,the number of patients with upper limb movement disorders is gradually increasing for diseases,accidents and aging population.The rehabilitation robot assisted training technology has become a research hotspot because of the low efficiency of traditional artificial assisted rehabilitation technology.However,the current rehabilitation robot system mainly focuses on the patients’ movement intention and physical fatigue,while not taking the patients’ emotional state into consideration.Failed to accurately understand the emotional state of the patients will result in poor rehabilitation and even secondary injury.The GA-PSO-SVM anxiety state assessment model is constructed to solve the problems of low recognition rate and lack of effective classification of patients’ anxiety in rehabilitation training,which is based on electroencephalogram(EEG)and electromyography(EMG)signals.The model is used in the upper limb rehabilitation training to evaluate the patient’s anxiety state in real time,which can realize the recognition of the three-level anxiety state.The specific research work is as follows:(1)The EMG signals and EEG signals are collected through experiments.With the consideration of the particularity of anxiety,three training tasks are designed for underweight,normal and overweight,while the task completion methods are selected to induce three anxiety states: mild,moderate and severe.A total of 10 subjects with 6 males and 4 females are selected to collect EEG and EMG signals.(2)The EMG and EEG signals are processed,as well as the features are extracted from EMG and EEG signals.A normal method is proposed to remove the high frequency noise of EMG signals,which is based on the wavelet threshold and permutation entropy algorithm.The combination of notch filter and band-pass filter is applied to remove power frequency noise and high-frequency noise in EEG signals.The ICA algorithm is adopted to eliminate electrooculogram interference in EEG signals.The features are extracted from EMG signals including 5 time domain features and 3 frequency domain features,while 3 time domain features and 2 frequency domain features are extracted from EEG signals.Considering the nonlinearity and non-stationarity of EEG and EMG signals,the features such as IMDF,IMNF,SC,LZC,and Samp En are acquired by variational modal decomposition algorithm.(3)The offline model of anxiety assessment is constructed based on GA-PSO-SVM.The PCA-Relief F-SVM method is used to obtain 16 feature fusion components,which removes redundant EEG and EMG features.The features of IEMG and LZC are selected through the higher recognition rate,whose inputs go into the support vector machine(SVM)respectively optimized by genetic algorithm(GA),particle swarm optimization(PSO)and improved GA-PSO fusion algorithm.Based on indicators such as the recognition rate and running time,the optimal classifier GA-PSO-SVM is obtained.In the end,the recognition rate of EEG and EMG fusion features is 83.325%,which is 9.4% and 9.1% higher than that of only relying on EMG or EEG features.(4)Anxiety status is recognized online based on upper limb rehabilitation platform.The Fourier M2 upper limb rehabilitation platform is adopted to design an active training experiment for anxiety state evaluation.The online recognition rate of anxiety states is 71%through the GA-PSO-SVM offline recognition model.In addition,the rehabilitation robot system adaptively adjusts the training level according to the anxiety state to improve the effect of rehabilitation training.