运动想象脑机接口技术作为一项创新治疗手段，通过解码大脑在想象肢体运动时产生的脑电信号，有望克服治疗手段和药物研发的限制，为治愈脑疾病和恢复受损脑功能提供新途径。研究关注不同人群在不同运动想象任务下脑电信号的差异以及对运动想象模式的准确识别，选取36名18～25岁的健康成年人，具备不同运动项目专长，设计3类运动想象任务，并采集了相应的脑电信号，对其特征进行深入分析。在受试者群体的差异性基础上，提出一种全新的深度学习框架RbMI-Net(Rhythm-based Motor Imagery Net)。该模型采用小波变换提取脑电信号特征信息，并将其输入到本研究设计的多层感知机模型中，以实现对不同运动想象脑电模式的准确识别。研究结果表明:1)在任务开始前和任务执行中，具有脚部运动优势的受试者表现出相对较高的大脑激活水平，较手部运动优势和手脚运动优势的受试者更为显著。手部运动优势的受试者在任务前后的大脑活动状态相对平稳，激活程度较低。2)RbMI-Net模型在十折交叉验证中展现出卓越的稳定性和准确性，三分类准确率达到82。59％，Kappa值为0。76。该模型在运动想象任务的脑电模式识别方面表现出色，优于当前领域内常见的脑电模式识别模型，成功突破了脑机接口技术的多分类难题。因此，通过深入研究不同运动专长人群的神经机制，本研究对于了解健康成年人在多项目运动想象模式方面具有重要意义，未来在推动脑机接口技术在大众运动健康中的应用也有深远的影响。
EEG Pattern Recognition on Multi-exercise Events of Motor Imagery Brain Computer Interface
The motor imagery brain computer interface technology can break through the limitations of treatment methods and drug development by decoding the EEG signals generated by brain-imagined limb movements,which becomes an innovative therapy for curing brain diseases and restoring damaged brain function.This study focuses on the differences in EEG signals among different groups of people under different imagery tasks,as well as the precise recognition of motor imagery patterns.We selected 36 healthy adults aged 18-25 years with expertise in three different sports events,designed three exercise imagery tasks,and collected EEG signals to analyze the dif-ferences in EEG signal characteristics.In addition,based on the group differences of the participants,this study proposed a novel deep learning framework,RbMI-Net(Rhythm based Motor Imagery Net),which used wavelet transform to extract EEG signal feature information from brain electrical signals and imput it into a multi-layer per-ceptron model designed in this study to identify different motor imagination EEG patterns.The results of this study are as follows:(1)Healthy adults with foot-motor dominance showed higher levels of brain activation before task initiation and during task performance compared to subjects with hand-motor dominance and hand-foot-motor domi-nance,and healthy adults with hand-motor dominance had the smoothest state of brain activity before and after the task,with the lowest levels of activation compared to each other;(2)The RbMI-Net model proposed in this study based on wavelet transform showed good performance with tenfold cross validation,with a triple classification accu-racy of 82.59％and a Kappa value of 0.76.The model excelled in the recognition of motor imagery brain electrical patterns,outperforming commonly used brain electrical pattern recognition models in the field,and successfully breaking through the multi-classification problem in brain-computer interface technology.Therefore,through in-depth research on the neural mechanisms of people with different motor expertise,this study is of great significance for exploring the multi-event exercise imagination patterns of healthy adults.It also has far-reaching implications for the promotion of brain computer interfaces in public exercise health in the future.
motor imagerybrain computer interfaceEEG signal patternprediction model
万方数据
维普
