Research On Upper Limb Rehabilitation System Based On Motor Imagery Decoding With Virtual Reality

Upper limb dysfunction is a problem that exists in most patients with dysfunction.Since traditional rehabilitation therapy cannot produce sufficient stimulation of patients’ motor nerves,more and more researchers are focusing on the field of brain-computer interface neurorehabilitation.The virtual reality upper limb rehabilitation system based on motor imagery,which can achieve the effect of neurorehabilitation training by using sensory stimulation,has become a popular research direction at present.However,the development period of the current virtual reality system is quite long and the construction cost is high,and the most classical Common Spatial Pattern(CSP)algorithm for motor imagery signal decoding relies heavily on the selection of filter bands due to the variability among subjects,and the classification accuracy is generally low under multi-class motor imagery tasks.To address the above issues,the following research components are carried out in this paper:(1)To address the problem of unstable classification performance of broadband CSP algorithm,a Narrow Filter Bank Common Spatial Pattern(NFBCSP)algorithm is proposed in this paper.Through a novel band search tree method,the optimal narrow band for the two-class motor imagery task is automatically determined,and the optimal narrow band is used for band-pass filtering to remove the redundant signals in EEG.Finally,the optimal narrow band is combined with the CSP algorithm to extract the dynamic energy features in the EEG signal and use SVM for classification.The results show that the NFBCSP algorithm achieves an average classification accuracy of 86.43%on the BCI-VR dichotomous motor imagery dataset,which is a significant improvement over the traditional broadband CSP algorithm as well as the FBCSP algorithm.On the other hand,on the BCI competition public dataset(Dataset 2a),the NFBCSP algorithm outperforms the best performing Ro CSP-SRIT2 NFIS algorithm in terms of the average classification accuracy as well as the average deviation.(2)To address the problem that CSP algorithms generally have low classification accuracy under multi-category tasks,a novel NFBCSP-DCNN feature input network is proposed in this paper.Firstly,the multi-category motor imagery task is transformed into multiple one-to-many(One vs Rest,OVR)binary classification tasks,and the optimal narrow band for each binary classification task is determined separately.After extracting the time-frequency domain features of each optimal narrow band,a deep convolutional neural network is used to fuse the band features and classify the multicategory motor imagery task.In this paper,the algorithm is validated on the Dataset 2a four-class motor imagery dataset,and the experimental results show that the algorithm has a significant improvement in classification performance and has a lower average bias compared to the multi-category motor imagery algorithms proposed in other works.(3)To address the problems of long development cycles and high construction costs of virtual reality systems,this paper designs a virtual reality upper limb rehabilitation system based on Web VR technology.Compared with traditional virtual reality systems,this system uses Web VR technology instead of Unity3 D to rapidly build virtual reality scenes,which significantly reduces the cost required to build virtual environments without degrading the sensory experience of subjects and speeds up the process of applying the upper limb rehabilitation system to clinical care.To enhance subject immersion,the system uses a panoramic view as the virtual reality background,and allows subjects to customize their virtual environment during rehabilitation training by photographing realistic scenes and importing them using a visualization interface.Finally,a comparison experiment was conducted with several subjects under virtual reality and normal 3D desktop animation effects.All subjects reported that the immersion provided by the virtual reality scenes could improve motivation during training,and they were less likely to become fatigued after long-term training,making it more suitable for clinical neurological rehabilitation training.