虚拟诱导患者下肢主动运动意图及其脑电精准感知方法

Virtual Induction of Patient’s Lower Limb Active Movement Intention and Electroencephalogram Precise Sensing Method

针对下肢运动功能障碍患者无法产生强烈主动运动意图造成外骨骼机器人在康复运动辅助时人机交互性差的问题,提出了一种虚拟诱导患者下肢主动运动意图及其脑电精准感知方法。首先,分析影响患者运动意图产生的因素,建立基于大脑注意机制的虚拟诱导患者意图产生模型,形成基于脑电信号的人机交互策略;然后,设计并搭建数据驱动沉浸式三维虚拟诱导场景,激发患者大脑产生主动运动意图;进而,采集患者脑电信号,通过基于自适应噪声完备经验模态分解(CEEMDAN)和独立成分分析(ICA)相结合的伪迹去除方法进行信号预处理;最后,利用深度卷积神经网络实现对患者运动意图的精准识别。实验结果表明:虚拟诱导方法能够有效增强受试者脑电信号特征,运动意图识别率明显提高,相比常规方法,采用虚拟诱导方法后,静息状态识别准确率达到80.5%,提高了10.33%,产生意图识别准确率达到92.17%,提高了20.5%,稳定维持在较高水平,为外骨骼机器人实现按需辅助控制奠定了基础。

Aiming at the problem that the patients with lower limb motor dysfunction cannot generate strong movement active intention,resulting in poor human-machine interaction of exoskeleton robot in rehabilitation motion assistance,a method of virtual induction of patient’s lower limb active movement intention and its electroencephalogram(EEG)precise sensing is proposed.Firstly,the factors that affect the generation of patients’motion intention are analyzed,a virtually induced patient intention generation model based on brain attention mechanism is established to form a human-computer interaction strategy based on EEG signals.Secondly,a data-driven immersive three-dimensional virtual induction scene is designed and constructed to stimulate patient’s brain to generate active motor intention.Furthermore,EEG signals of patients are collected and pretreated by the artifact removal method based on CEEMDAN-ICA.Finally,a deep convolutional neural network(CNN)is used to accurately identify the patient’s movement intention.Experimental results show that the virtual induction method can effectively enhance the characteristics of EEG signals,and the recognition of motion intention is significantly improved.Compared with the conventional method,when the virtual induction method is adopted,the accuracy rate of resting state recognition reaches 80.5%,10.33%higher than that of the conventional method,and the accuracy rate of generation intention recognition reaches 92.17%,20.5%higher than that of conventional method,which is stable at a high level.These results lay a foundation for the auxiliary control of exoskeleton robot.