CASIA OpenIR  > 毕业生  > 博士学位论文
面向偏瘫康复的人体下肢运动数据智能分析方法与实验
崔承坤1,2
学位类型工学博士
导师侯增广
2018-05-29
学位授予单位中国科学院大学
学位授予地点北京
关键词脑卒中偏瘫 多模态信息融合 运动行为识别 运动异常分析 康复疗效评定
摘要脑卒中偏瘫是由于脑血管阻塞或破裂而引起的一侧肢体的运动障碍,也是影响脑卒中患者日常生活能力和生活质量的主要原因。脑卒中发病后的运动康复治疗对于患者肢体运动功能的恢复有着不可替代的作用。为了缓解当前康复医疗资源的供需矛盾,一些智能康复辅助系统逐渐被引入到康复治疗领域,用于辅助患者完成相应的训练任务。临床研究已证实,患者主动参与的康复训练对于其运动功能的恢复具有更加积极的促进效果,由此产生了两个方面的科学问题:一方面,需要准确解码患者的运动行为,从而实现患者与辅助系统之间的柔顺交互;另一方面,需要精确评估患者运动功能的康复效果,以便能够及时调整训练方式。解决以上问题的关键技术在于如何从与患者运动相关的数据中挖掘出关于人体运动行为和肢体运动功能的重要信息,该技术也是实现智能康复的首要前提,目前已成为康复医学和康复工程领域的一个重点研究方向。本文以多模态人体运动数据作为切入点,对人体下肢运动行为的感知监测和肢体运动功能的量化评估等问题进行了深入的研究。
本文的主要研究内容和创新点如下:
(1) 提出了一种基于脑-肌生理信号融合的运动姿态分类方法,以解决人体下肢多关节运动姿态的定性分类问题。首先,设计了相关人体运动实验,包括踏车、步行、上下台阶等日常生活中最常见的下肢运动形式,从受试者身上同步采集了脑电、肌电、肌音信号等三种不同模态的生理数据。然后,提出了一种基于多模态生理信息的决策融合分析结构,以建立脑-肌生理信号与运动姿态模式间的映射关系。在该结构中,我们实现了多种先进的模式分类算法和决策融合算法。最后,开展了充分的数据分析实验,统计结果表明,所提出的分类方法可以有效利用大脑活动信息与肌肉活动信息之间的互补性,来提高对人体下肢多关节运动姿态的识别精度,其中,基于支持向量机的三模态决策融合模型获得了98.61%的平均准确率,显著优于目前常用的基于两模态融合或基于单一模态的分类方法。
(2) 提出了一种基于脑-肌生理信号融合的关节角度估计方法,以解决人体下肢各关节连续运动的定量估计问题。我们主要研究了如何利用生理信号来准确估计受试者在踏车运动过程中连续变化的关节角度,实验中所收集的数据包括了脑电、肌电、肌音等生理信号,以及下肢髋、膝、踝等关节角度信号。我们提出了一种基于多模态回归融合的估计方法来建立脑-肌生理信号与关节角度之间的映射关系,并进行了大量的对比实验研究。数据分析结果表明,所提出的估计方法能够有效提高对人体下肢各关节角度的估计精度,其中,基于三种生理信号的最小二乘支持向量回归融合模型,在不同的踏车负载条件下,分别获得了2.51度、2.81度和3.13度的平均估计误差,显著优于目前最常用的基于肌电信号的估计方法,且对于踏车负载的变化也表现出较好的鲁棒性。
(3) 提出了一种能够用于定性检测和定量评估步态异常的智能步态分析方法,以解决脑卒中偏瘫患者步态异常的自动检测与评估问题。首先,设计了包括脑卒中偏瘫患者和健康志愿者参与的临床步行对照试验,从受试者身上同步采集了标记球运动位置、地面反作用力和肌电等三种最具有代表性的步态数据。然后,利用这些不同模态的步态数据并基于信息融合理论建立了一个步态异常分析系统,该系统可用于定性检测脑卒中患者的步态异常模式,同时,为了进一步量化评估患者的下肢步行功能,我们提出了一种新颖的基于概率的自定义步态得分。最后,进行了大量的数据分析实验,实验结果表明所提出的方法对于脑卒中患者的步态异常具有高达98.21%的检测准确率,而且对于患者步行能力的概率估计得分与临床步态量表评分之间存在着显著的强相关性,其相关系数的绝对值达到了0.84,在一定程度上证实了所提出方法的临床有效性。
(4) 提出了一种身心复合康复疗法和基于多元数据的综合康复评定方法,以探索和研究针对脑卒中偏瘫患者下肢步行功能的新型治疗方法和疗效评价问题。我们首先分析了脑卒中偏瘫患者在步行康复治疗过程中经常遇到的运动障碍和心理问题,在此基础上将运动康复疗法与心理康复疗法有机结合起来,提出了身心复合康复疗法,该疗法可以从身、心两个层面上为患者提供双重的治疗体验。然后,设计了临床康复对照试验,将脑卒中患者随机分成两组,对研究组进行身心复合治疗,对照组只进行传统的运动治疗,试验中分别收集了两组患者在治疗前后的多种运动数据。最后,结合多种数据分析方法,从多个角度对患者康复治疗前后的步行能力进行了全方位的综合评定。统计分析结果表明,身心复合康复疗法对脑卒中患者步行功能的改善具有非常积极的促进作用,其改善程度显著优于传统的运动疗法。

其他摘要Post-stroke hemiparesis is a condition caused by cerebrovascular blockage or rupture that lead to movement disorders of one side of the body. Movement disorders seriously affect the patients' daily activities and quality of life. The motor rehabilitation treatment after hemiparesis is essential to recover the patients' motor function. In order to alleviate the contradiction between supply and demand of current rehabilitation medical resources, some intelligent rehabilitation aid systems have been introduced into the field of rehabilitation to assist patients in completing the rehabilitation training tasks. Clinical studies have confirmed that the rehabilitation training with patients' active participation has a more positive effect on their motor function recovery. This will derive two scientific issues: on the one hand, it is necessary to accurately decode the patient's motion behaviors so as to achieve a flexible interaction between the patient and the aid system; on the other hand, it is necessary to accurately assess the patient's rehabilitation effect so as to adjust the training mode in time. The key technique to solve the above problems is how to extract important information about human motion behavior and limb motor function from the data which are related to the patient's motions. This technique is also the primary prerequisite for the realization of intelligent rehabilitation, which has become a key research direction in the field of rehabilitation medicine and rehabilitation engineering. This thesis focuses on the perceptual monitoring of human lower limb motion behaviors and quantitative assessment of limb motor function by analyzing multimodal human motion data.
The main research contents and innovation points are as follows:
(1) A motion pattern classification method based on the fusion of brain-muscle physiological signals is proposed to qualitatively distinguish human lower limb multi-joint movements. Different lower limb exercises are considered. They are cycling, walking, and going up and down a step, which are the most common activities of daily living. Electroencephalogram (EEG), electromyogram (EMG) and mechanomyogram (MMG) signals were simultaneously recorded from the subjects while performing these lower limb motions. Then, a decision fusion analysis framework based on multimodal physiological information is proposed to establish the mapping relationship between the brain-muscle physiological signals and motion patterns. In this framework, a lot of advanced pattern classification algorithms and decision fusion algorithms are implemented. Sufficient data analysis experiments are carried out and the results show that the proposed method can effectively take the advantages of the complementarity between the brain and muscular activities to improve the recognition accuracy of motion patterns. A support vector machine based three-modal fusion model can achieve an average accuracy of 98.61%, which is significant higher than the commonly used two-modal fusion based methods or single-modal based methods.
(2) A joint angle estimation method based on the fusion of brain-muscle physiological signals is proposed to quantitatively estimate the continuous movements of human lower limb joints. We mainly study how to use physiological signals to accurately estimate the subjects' continuously changing joint angle during cycling exercises. The collected data includes EEG, EMG and MMG signals, as well as the joint angle signals of the hip, knee and ankle. Then, an estimation method based on multimodal regression fusion is proposed to establish the mapping relationship between the brain-muscle physiological signals and joint angles. A lot of comparative experiments are conducted. The results show that the proposed method can effectively reduce the estimation errors of the joint angles. The least squares support vector regression fusion model based three physiological signals obtain the mean estimated errors of 2.51, 2.81, and 3.13 degrees under different cycling load conditions, which is significant better than the most commonly used estimation methods based on the EMG signal, and also has good robustness against changes in cycling loads.
(3) An intelligent gait analysis method is proposed to qualitatively detect and quantitatively assess the gait abnormality of the post-stroke hemiparetic patients. A clinical walking trial involving the post-stroke hemiparetic patients and healthy volunteers is designed. Three of the most representative gait data, i.e., marker motion position, ground reaction force and EMG signals, were simultaneously acquired from these subjects during walking. Then, a gait abnormality analysis system is established based on these different modal gait data by using information fusion theory, which can be used to qualitatively detect the gait abnormal patterns after stroke. In order to further quantify the patients' lower limb walking ability, a novel probability based custom-defined gait score is designed for this system. Extensive data analysis experiments have been carried out. The results show that for the detection performance, the accuracy is as high as 98.21%. For the estimation performance, there is a significant strong correlation between the probability based gait scores and the clinical gait scale ratings. The absolute value of the correlation coefficient between them is 0.84, which validates the clinical effectiveness of the proposed method to some extent.
(4) A kind of physical and mental compound rehabilitation therapy, and a comprehensive rehabilitation assessment method based on multivariate data are proposed to study new treatment methods and rehabilitation effect assessment for the post-stroke hemiparetic patients. The movement disorders and psychological problems of the post-stroke hemiparetic patients during walking rehabilitation are analyzed. On this basis, the physical rehabilitation therapy and psychological rehabilitation therapy is appropriately combined to propose a physical and mental compound rehabilitation therapy, which can provide the patients with a dual treatment experience from body and mind. A clinical rehabilitation trial is designed. Post-stroke patients are randomly divided into two groups. For the study group, physical and mental compound therapy is performed for every patient. For the control group, only traditional physical therapy is carried out. Various data of the two groups of patients are collected before and after rehabilitation. Then, the walking ability of patients before and after rehabilitation is comprehensively assessed from many aspects by using a lot of data analysis methods. The results show that physical and mental compound rehabilitation therapy has a very positive effect on improving the walking ability of the post-stroke patients and is significantly better than traditional physical therapy.

文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/20989
专题毕业生_博士学位论文
作者单位1.中国科学院自动化研究所
2.中国科学院大学
推荐引用方式
GB/T 7714
崔承坤. 面向偏瘫康复的人体下肢运动数据智能分析方法与实验[D]. 北京. 中国科学院大学,2018.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
面向偏瘫康复的人体下肢运动数据智能分析方(13682KB)学位论文 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[崔承坤]的文章
百度学术
百度学术中相似的文章
[崔承坤]的文章
必应学术
必应学术中相似的文章
[崔承坤]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。