实时手部精细运动意图识别方法的研究

关键词：脑卒中  手部运动意图  模板匹配  数据手套  实时识别

Keywords:stroke  hand motor intent  template matching  data glove  real-time identification

摘要：

      摘要 目的：利用脑卒中患者的自主运动意识进行主动神经康复是促进患者脑功能重塑、提高康复训练效果的重要工程学手段。针对脑卒中患者手部精细运动功能恢复速度慢、恢复程度有限等问题，本文提出了一种基于数据手套的模板匹配方法，用于识别患者的手部精细运动意图。 方法：利用自主研发的虚拟现实康复训练平台，将基于数据手套模板匹配的运动意图识别方法嵌入其中，并与基于表面肌电模式识别方法进行对比研究。招募了6例健康受试者参与实验，对16个手部精细动作的离线识别性能和实时识别性能分别进行分析，并对离线性能指标与实时性能指标之间的关系进行相关性研究。 结果：采用基于数据手套模板匹配方法取得的平均离线动作识别准确率为95.00%±3.66%，平均实时动作完成率为91.31%±1.17%，显著高于基于表面肌电模式识别方法的离线动作识别准确率（84.66%±4.66%，P<0.01）与实时动作完成率（71.86%±10.04%，P<0.01）。另外，基于数据手套模板匹配方法取得的离线动作识别准确率与实时动作完成率是线性相关的（P<0.05），而基于表面肌电模式识别方法取得的离线与实时性能指标不存在线性关系（P=0.4005）。 结论：与传统的肌电模式识别方法相比，基于数据手套模板匹配方法具有显著的比较性优势。因此，有望成为手部精细运动功能主动神经康复中的一种有效的运动意图识别方法。

Abstract:

      Abstract Objective: Active rehabilitation putting great emphasize on the concious effort of stroke survivors is an important engineering means to promote their brain plasticity and improve the rehabilitation effect. Towards the neurorehabilitation of hand motor function which was often slow and limited, a data-glove based template matching (Glove-TM) method was proposed for intent identification of fine hand movements in this study. Method: The proposed Glove-TM method was embedded into a custom-made virtual-reality rehabilitation training platform, and the electromyogram based pattern recognition (EMG-PR) approach was also included for comparison. The offline and real-time identification performance of 16 classes of hand movements were examined with six healthy subjects, respectively. In addition, the relationship between offline and real-time performance metric was investigated by linear regression method. Result: The average offline motion classification accuracy was 95.00%±3.66%, and the average real-time motion completion rate was 91.31%±1.17% when using Glove-TM method, which were significantly higher than those when using EMG-PR method (with offline classification accuracy of 84.66%±4.66% and motion completion rate of 71.86%±10.04%, P<0.01). And it is found that the offline classification accuracy and real-time motion completion rate was linear correlated when using Glove-TM method (P<0.05), but it was nonlinear correlated when using EMG-PR method (P=0.4005). Conclusion: Compared with the commonly used EMG-PR, the Glove-TM has obvious comparative advantage. It might be a promising method for the motor intent identification in active neurorehabilitation.