基于独立成分分析的大脑运动功能激活模式研究

发布时间：2018-01-30 05:11

本文关键词： 功能磁共振成像时空独立成分分析一般线性模型激活模式脑卒中康复　出处：《沈阳工业大学》2014年博士论文　论文类型：学位论文

【摘要】：“脑卒中”发病率的增加，严重的危害着人们的身体健康和生命安全，致残率居高不下。功能磁共振成像（fMRI）能够通过实时的检测大脑血液中脱氧血红蛋白含量的变化来间接反映参与活动任务的神经元的激活状况，它的出现为探索脑卒中后康复机制、评价和判断预后方面提供了新的研究思路，因此，在康复医学领域具有良好的应用前景。然而，fMRI数据具有成分复杂（时空性）、数据量大、多噪声且信号弱等特点，导致数据处理十分不易。独立成分分析（ICA）是现有的方法中唯一一种基于四阶统计量的方法，能够对fMRI信号与噪声的统计信息进行更深层次的挖掘，因此，在fMRI中被广泛应用并成为国际上的研究热点。本文深入分析了fMRI对比度机制、噪声的产生和原有数据处理方法的不足，针对fMRI数据特点，通过提出新算法、优化老算法或结合多种数据处理算法的方式，，来完善ICA在fMRI中的应用，并以此为基础，探索正常人下肢运动的大脑皮层激活模式和卒中后患者康复期内运动功能皮层的重组规律，为卒中后的下肢运动功能临床康复提供理论基础。（1）针对邻域相关ICA算法严重依赖参考函数的问题，提出一种邻域自相关ICA算法，在不需要参考函数的情况下，通过检测体素点各周期的时间序列相关性，对fMRI数据进行激活区提取。将该算法分别应用于对仿真数据和对12组真实fMRI数据的处理，并与前人方法进行了对比，分析了算法的准确性和稳定性。（2）针对传统ICA对fMRI数据时空特性假设不合理的问题，提出了基于infomax判据的stICA优化算法。该算法通过同时最大程度的优化时间源和空间源的独立性，来建立两个领域的平衡，服从物理上更真实的假设。通过对仿真数据的处理讨论了改进算法的准确性。将该算法应用于踝关节主动运动与被动运动激活模式的对比，判断被动运动是否可以作为无法进行主动运动时的替代刺激手段。（3）讨论了GLM和stICA两种模型的共性和差别，针对stICA模型稳定性差和GLM违反其基本假设的缺点，提出stICA-GLM联合算法。通过对仿真数据的处理讨论了联合算法的准确性和稳定性。通过同个体同条件下的不同被动fMRI实验指出受试者在进行被动运动时，大脑思维不受控制，会产生大量神经性噪声。将stICA-GLM联合算法应用于对神经性噪声的消除，并将其结果与GLM结果进行了对比。（4）针对fMRI数据量庞大，基于梯度算法的收敛方式很难满足fMRI数据处理的速度要求的问题，提出了基于固定点的stICA联合算法（Fast-stICA-GLM）。分析了Fast-stICA-GLM算法的收敛性能，提出在算法中添加步长因子来优化其收敛性能。通过真实fMRI数据对Fast-stICA-GLM和Infomax-stICA-GLM两种算法进行了对比，对比内容包括准确性、稳定性和运算速度。最后应用Fast-stICA-GLM算法分析了不同个体之间被动运动的激活状况。（5）应用Fast-stICA-GLM算法作为数据处理手段对卒中后患者进行为期6周（共4次）的跟踪fMRI研究，记录卒中患者在康复训练期间，大脑下肢运动功能皮层的重组情况，通过定量和定性指标给出功能皮层的重组规律。定量指标包括：偏侧化指数LI、峰值点坐标、激活体积和峰值点体素信号强度；定性指标包括：下肢运动功能所在的解剖区域和激活体素所在的brodmann分区。通过这些指标，可以得知某个单独关节的运动功能的恢复情况，从而对制定针对性的康复计划起到指导作用。
[Abstract]:"The increase in stroke incidence, serious harm to people's health and life safety, the disability rate is still high. Functional magnetic resonance imaging (fMRI) can change real-time detection of brain blood deoxyhemoglobin content to indirectly reflect the activation of neurons participate in the activities of the task, it appears in order to explore the mechanism of rehabilitation after stroke, provided research ideas, new evaluation and prognosis so it has good application prospect in the field of rehabilitation medicine. However, fMRI data has a complex composition (time and space), the large amount of data, much noise and weak signal characteristics, resulting in data processing is not easy. Independent component analysis (ICA) is the existing methods only a method of four order statistics based on the statistical information of the fMRI to the signal and the noise of deeper mining, therefore, is widely used in fMRI and It is a hot spot of research in the world.
This paper analyzes the fMRI contrast mechanism, lack of noise and the original data processing, according to the characteristics of fMRI data, the proposed new algorithm, combined with a variety of data processing algorithms or the old algorithm, to improve the application of ICA in fMRI, and on this basis, to explore the rehabilitation of patients with lower limb reorganization law the motion of normal human brain cortex activation patterns and motor function after stroke within the cortex, provide a theoretical basis for clinical rehabilitation of lower limb motor function after stroke.
(1) the neighborhood ICA algorithm relies heavily on the reference function, proposes a neighborhood self correlation ICA algorithm, without reference function, through the detection of voxel time series correlation of each cycle, activation of the fMRI data extraction. The algorithm is respectively applied to the simulation data and treatment of 12 groups of real fMRI data, and compared with the previous methods, the algorithm accuracy and stability are analyzed.
(2) according to the traditional ICA on the spatial and temporal characteristics of fMRI data that is not reasonable, put forward the optimization algorithm of stICA Infomax based on the criterion of the algorithm. At the same time through the optimization of time and space independent source source to the maximum extent, to build two areas of physical balance, to a more realistic assumption. Accuracy of the improved algorithm the discussion by analyzing the simulation data. The algorithm is applied to active ankle movement and passive movement active mode of comparison, judgment whether passive motion can be used as an alternative to active movement of the stimulus.
(3) discuss the similarities and differences between GLM and stICA two models, stICA model for stability and GLM violation of the basic assumptions of the shortcomings, proposed stICA-GLM algorithm. The algorithm's accuracy and stability is discussed by analyzing the simulation data. Through different passive fMRI experiments with the same individual conditions are pointed out the subjects in the passive movement, the mind is not controlled, will produce a lot of noise. The neural stICA-GLM algorithm is applied to eliminate the noise of nerve, and the results were compared with the results of GLM.
(4) fMRI for the huge amount of data, based on the convergence of gradient algorithm is difficult to meet the requirements of the fMRI data processing speed, put forward stICA combined algorithm based on fixed point (Fast-stICA-GLM). The convergence performance of Fast-stICA-GLM algorithm is analyzed, it is suggested to add the step factor in the algorithm to optimize the convergence performance by real fMRI. The data of Fast-stICA-GLM and Infomax-stICA-GLM two kinds of algorithms are compared, including the comparison of accuracy, stability and speed of operation. Finally the application of Fast-stICA-GLM algorithm to analyze the passive motion activation between different individuals.
(5) the application of Fast-stICA-GLM algorithm as a data processing method for patients after stroke for 6 weeks (4 times) the fMRI tracking research, recording stroke patients in rehabilitation training during the reorganization of the brain cortex of lower extremity motor function, through the reorganization law of quantitative and qualitative indicators are functional cortex. Quantitative indicators include: laterality the LI index, the peak point coordinates, the activation volume and peak voxel signal strength; qualitative indicators include: Brodmann partition anatomical regions of lower extremity motor function and the activated voxel location. These indicators can be that a single joint movement function recovery, so as to develop guidance for rehabilitation plan the.

【学位授予单位】：沈阳工业大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：R743.3;R445.2

【参考文献】