基于多目标优化约束独立成分分析方法的fMRI数据分析研究

约束独立成分分析(CICA)通过加入先验信息,可极大地提高独立成分分析(ICA)的盲源信号分析性能,但还存在先验信息难以获取、先验信息约束条件阈值参数难以选择以及先验信息难以被有效利用等问题,需要进一步研究和解决。在多目标优化框架的基础上,建立一种同时融合时空先验信息的CICA模型,可有效规避CICA中阈值参数选择的问题。此外,提出一种从多被试fMRI数据中提取本真先验信息来指导fMRI组分析的自适应挖掘算法,从而为CICA获取先验信息提供一种新途径。最后,利用10例模拟数据、5例任务态和23例静息态fMRI数据,验证所提方法的有效性。结果表明:基于多目标优化的CICA(MOPCICA)获得的时空源信号总体上优于ICA、包含时间信息的CICA(CICA-tR)和包含空间信息的CICA(CICA-sR)(P<0.05)(如在模拟数据中,对应的空间AUC和时间相关系数分别0.75±0.05、0.62±0.02、0.72±0.03、0.71±0.06和0.81±0.13、0.67±0.04、0.74±0.09、0.77±0.13),而空间独立性则优于CICA-tR和CICA-sR(P<0.05)(如在任务态数据中,对应的峭度和负熵分别为69.20±23.36、17.60±13.22、36.71±13.43和0.031 2±0.007 7、0.003 7±0.002 1、0.018 4±0.004 5),从而说明它具有更好的源信号恢复性能。同时,在静息态数据中利用fMRI本真先验信息,MOPCICA获得的组成分与每个被试相应成分之间的相关系数平均高于ICA、基于牛顿迭代法的CICA(CICA-nR)和基于不动点迭代法的CICA(CICA-fR)(P<0.05)(分别为0.46±0.08、0.44±0.08、0.45±0.08和0.44±0.08),从而更能代表组中被试的共性。研究表明,所提出的方法对fMRI脑功能连通性检测具有重要意义。     

基于偶极子特征优化的情绪重评同步EEG-fMRI源定位研究

为研究情绪重评时的大脑皮层源活动,针对情绪重评实验范式下采集的15例健康人同步EEG-fMRI数据,首先提出一种新颖的基于偶极子特征优化的融合源定位方法:根据fMRI加权最小范数估计源定位结果,采用20 ms EEG滑动时间窗,提取每个时窗内的偶极子空间融合特征,将其作为动态融合先验进行加权最小范数估计溯源;随后将该结果与fMRI加权最小范数估计源定位结果进行情绪重评机制上的对比;最后采用样本熵进行脑电源复杂度分析。实验结果表明,该方法可以在高时间和空间分辨率下,有效地追踪情绪重评任务下大脑皮层上的脑电源动态并识别出相关脑区。情绪重评过程中,随着后枕顶叶晚期正电位的出现,显著活跃脑区从左顶叶下部、右侧额中回下部、左侧脑岛转移到右侧颞上回和左外侧枕叶,最后在晚期正电位慢波阶段激活了右侧梭状回、右侧额中回下部和右侧扣带回峡部(P<0.05)。通过脑电源样本熵的计算,提取出被试在接受不同情绪刺激后1500 ms内的显著脑区(P<0.05):情绪响应的活跃脑区为左外侧枕叶(负性:0.688±0.124,中性:0.590±0.126);情绪重评的活跃脑区为右侧额中回下部(负性重评:0.814±0.114,负性:0.736±0.123);情绪重评的抑制脑区为右侧颞上回(负性:0.642±0.152,负性重评:0.546±0.090)。这些结果为情绪重评相关的皮层脑电源定位研究提供了脑区参考。     

基于核磁共振扩散张量成像技术的脑组织3D分割算法研究

无论在临床诊断上还是在脑电/脑磁研究中,脑组织的准确分割都具有重要的作用.由于DT-MRI比传统MRI能够提供更多的关于脑组织生理结构特性的物理信息,所以基于DT-MRI的分割将更加准确有效.基于扩散张量核磁共振成像技术,提出把脑内部分割成灰质、白质、脑脊液3种组织的新方法,对脑脊液采用扩散张量特征值阈值法进行分割,对白质与灰质采用3D不变参数均值聚类算法进行分割.在Matlab 7.1环境下,采用该方法对病人DT-MRI数据进行实际计算验证,得到较好的结果.对DT-MRI张量空间的准确分割,需要高效率算法与不同脑组织生理结构特性先验知识的有机结合.     

脑成像与脑连接

美国NIH在2009年发布了人脑连接组计划(HCP),该计划的实施将主要依赖三大关键技术:脑电与功能磁共振(fMRI)信息的融合技术、静息态fMRI技术、脑纤维束成像等脑结构分析技术.文中简要介绍了相关技术的研究现状,以及未来的主要发展方向.     

利用小波神经网络求解脑电等效偶极子源参数

提出了多维单尺度径向基小波神经网络的构造性算法 ,并将之应用到脑电等效偶极子源定位问题 ,从而避开对模型系统描述的困难和现有的迭代类求解方法计算时间较长的问题。通过对解空间的随机抽样由正向计算形成学习样本 ,使小波神经网络在训练过程中建立起自己的逆映射联想记忆 ,以正确推断出头皮观测电位与脑内源发生器之间的内在联系 ,使之对于新的脑电数据能实时地、准确地估算出等效偶极子源的参数 ,为脑电的动态分析提供一条途径。计算机仿真计算结果证明了此方法的有效性     

一种脑磁定位中背景噪声的去除方法

脑磁源的定位问题是脑磁图(MEG)研究的一个基本问题。在脑磁源的定位研究中，噪声对定位的准确性有很大的影响。自发的脑神经活动所产生磁场与感兴趣的诱发磁场不相关，可将其称为背景噪声。传统的消噪方法无法有效去除这种背景噪声。利用不含信号信息的纯噪协方差矩阵结合MUSIC算法；可以有效地抑制背景噪声而快速获得较好的定位结果。     

空间ICA在fMRI数据上的应用与分析

独立成分分析(ICA)技术试图将多维数据分解成若干个相互统计独立的分量。时间ICA和空间ICA都可以用于分析功能核磁共振成像(fMRI)数据。但由于fMRI数据空间维数远远大于时间维数，为计算方便，在分析fMRI数据时。则更多的使用空间ICA方法。本文在单任务激励实验中，利用ICA方法从fMRI数据中分离出若干个与任务相关的独立分量，其中包括与任务相关的恒定分量(CTR)和与任务相关的暂态分量(TTR)；通过将这些独立分量进行空间映射，得到了与任务相关的脑部激活区域。将此结果与SPM的分析比较，得到了一致的结果。在对结果的分析中，我们进一步指出了ICA方法的特点和局限性。     

血氧水平依赖功能磁共振成像的基本原理及方法学应用

功能磁共振成像(functional magnetic resonance imaging,fMRI)是研究脑功能的一种强有力的方法,具有无创定位、高空间分辨率和时间分辨率的特点。最基本的fMRI的成像技术是血氧水平依赖功能磁共振成像(blood oxygen level dependent fMRI,BOLD-fMRI)。概述了BOLD-fMRI的基本原理以及在方法学上的应用。     

三目标最优化结合磁场信号SVD正交分解用于脑磁源的定位计算(MEG)

脑磁源的定位是脑磁图（ｍａｇｎｅｔｏｅｎｃｅｐｈａｌｏｇｒａｐｈｙ简称ＭＥＧ）研究的一个基本问题。该问题属不定问题，即根据探头测量的微小磁场所建立的方程组求得的未知脑磁源有无穷组解。因此需要通过建立合理的数学模型补充适当信息。脑磁场测量信号ＳＶＤ正交分解且进行单偶极子搜索方法（ｍｕｓｉｃ）已被广泛采用。该方法假定脑磁源是许多孤立的偶极子源，且偶极子源之间在统计上是互相独立的。若脑磁源不满足上述假定，一般讲该方法失效。三目标最优化法是最近推出的一方法，若偶极子源分布在脑空间内任意深度平行于探头表面的一层附近，该方法可给出正确的解。三目标最优化法与脑磁场测量信号的ＳＶＤ正交分解方法相结合可得一新方法，它将脑空间划分成几个子区域，每一区域在平行探头表面不同深度的一层附近。新方法仅要求每一区域内的脑磁源与其它区域内的源在统计上互相独立。而区域内的源是允许互相相关的。与前两种方法相比，新方法进一步放松了对脑磁源的要求，因此有更广泛的适用性。本文详细推导并论证该新方法并讨论它对磁场的源的独立性的要求。本文所述新方法也适用对脑电图ＥＥＧ源的定位计算。     

基于高频颅内脑电的致痫区定位和网络分析

目的通过癫痫发作期颅内高频脑电(60~90 Hz)的分析,实现一种基于非参数统计模型的致痫区定位方法,并研究致痫区内外的脑电信号相干性。方法首先构造高频脑电归一化指标——高频能量指数,再用该指标量化各导联的致痫性,并定位致痫区,最后分析致痫区内外电极的幅度相干和相位相干性。结果分析了8例有良好术后效果的癫痫患者共1252个电极触点的颅内脑电数据,以临床癫痫灶定位结果作为参考,高频能量指数平均敏感性29.57%、特异性96.67%。8位患者的致痫电极之间的平均相干性显著高于非致痫电极之间的相干性。结论构造了颅内脑电高频能量指标,并用非参数统计模型定位致痫区,特异性高于已有研究,能够帮助医生在癫痫手术规划中寻找关键相关致痫电极。     

Multimodal Integration of fMRI and EEG Data for High Spatial and Temporal Resolution Analysis of Brain Networks

Two major non-invasive brain mapping techniques, electroencephalography (EEG) and functional magnetic resonance imaging (fMRI), have complementary advantages with regard to their spatial and temporal resolution. We propose an approach based on the integration of EEG and fMRI, enabling the EEG temporal dynamics of information processing to be characterized within spatially well-defined fMRI large-scale networks. First, the fMRI data are decomposed into networks by means of spatial independent component analysis (sICA), and those associated with intrinsic activity and/or responding to task performance are selected using information from the related time-courses. Next, the EEG data over all sensors are averaged with respect to event timing, thus calculating event-related potentials (ERPs). The ERPs are subjected to temporal ICA (tICA), and the resulting components are localized with the weighted minimum norm (WMNLS) algorithm using the task-related fMRI networks as priors. Finally, the temporal contribution of each ERP component in the areas belonging to the fMRI large-scale networks is estimated. The proposed approach has been evaluated on visual target detection data. Our results confirm that two different components, commonly observed in EEG when presenting novel and salient stimuli, respectively, are related to the neuronal activation in large-scale networks, operating at different latencies and associated with different functional processes.     

Integrated MEG/fMRI Model Validated Using Real Auditory Data

The main objective of this paper is to present methods and results for the estimation of parameters of our proposed integrated magnetoencephalography (MEG) and functional magnetic resonance imaging (fMRI) model. We use real auditory MEG and fMRI datasets from 7 normal subjects to estimate the parameters of the model. The MEG and fMRI data were acquired at different times, but the stimulus profile was the same for both techniques. We use independent component analysis (ICA) to extract activation-related signal from the MEG data. The stimulus-correlated ICA component is used to estimate MEG parameters of the model. The temporal and spatial information of the fMRI datasets are used to estimate fMRI parameters of the model. The estimated parameters have reasonable means and standard deviations for all subjects. Goodness of fit of the real data to our model shows the possibility of using the proposed model to simulate realistic datasets for evaluation of integrated MEG/fMRI analysis methods.     

The impact of task load on the integration of explicit contextual priors and visual information during anticipation

There is limited knowledge about the impact of task load on experts’ integration of contextual priors and visual information during dynamic and rapidly evolving anticipation tasks. We examined how experts integrate contextual priors––specifically, prior information regarding an opponent's action tendencies––with visual information such as movement kinematics, during a soccer-specific anticipation task. Furthermore, we combined psychophysiological measures and retrospective self-reports to gain insight into the cognitive load associated with this integration. Players were required to predict the action of an oncoming opponent, with and without the explicit provision of contextual priors, under two different task loads. In addition to anticipation performance, we compared continuous electroencephalography (EEG) and self-reports of cognitive load across conditions. Our data provide tentative evidence that increased task load may impair performance by disrupting the integration of contextual priors and visual information. EEG data suggest that cognitive load may increase when contextual priors are explicitly provided, whereas self-report data suggested a decrease in cognitive load. The findings provide insight into the processing demands associated with integration of contextual priors and visual information during dynamic anticipation tasks, and have implications for the utility of priors under cognitively demanding conditions. Furthermore, our findings add to the existing literature, suggesting that continuous EEG may be a more valid measure than retrospective self-reports for in-task assessment of cognitive load.     

Noise reduction and convergence of Bayesian algorithms with blobs based on the Huber function and median root prior

Iterative image reconstruction algorithms have the potential to produce low noise images. Early stopping of the iteration process is problematic because some features of the image may converge slowly. On the other hand, there may be noise build-up with increased number of iterations. Therefore, we examined the stabilizing effect of using two different prior functions as well as image representation by blobs so that the number of iterations could be increased without noise build-up. Reconstruction was performed of simulated phantoms and of real data acquired by positron emission tomography. Image quality measures were calculated for images reconstructed with or without priors. Both priors stabilized the iteration process. The first prior based on the Huber function reduced the noise without significant loss of contrast recovery of small spots, but the drawback of the method was the difficulty in finding optimal values of two free parameters. The second method based on a median root prior has only one Bayesian parameter which was easy to set, but it should be taken into account that the image resolution while using that prior has to be chosen sufficiently high not to cause the complete removal of small spots. In conclusion, the Huber penalty function gives accurate and low noise images, but it may be difficult to determine the parameters. The median root prior method is not quite as accurate but may be used if image resolution is increased.     

Equivalent Charge Source Model Based Iterative Maximum Neighbor Weight for Sparse EEG Source Localization

How to localize the neural electric activities within brain effectively and precisely from the scalp electroencephalogram (EEG) recordings is a critical issue for current study in clinical neurology and cognitive neuroscience. In this paper, based on the charge source model and the iterative re-weighted strategy, proposed is a new maximum neighbor weight based iterative sparse source imaging method, termed as CMOSS (Charge source model based Maximum neighbOr weight Sparse Solution). Different from the weight used in focal underdetermined system solver (FOCUSS) where the weight for each point in the discrete solution space is independently updated in iterations, the new designed weight for each point in each iteration is determined by the source solution of the last iteration at both the point and its neighbors. Using such a new weight, the next iteration may have a bigger chance to rectify the local source location bias existed in the previous iteration solution. The simulation studies with comparison to FOCUSS and LORETA for various source configurations were conducted on a realistic 3-shell head model, and the results confirmed the validation of CMOSS for sparse EEG source localization. Finally, CMOSS was applied to localize sources elicited in a visual stimuli experiment, and the result was consistent with those source areas involved in visual processing reported in previous studies.     

基于加权压缩感知的MR图像重建方法

背景：压缩感知理论已广泛应用于MR图像的快速重建中。在对K空间数据进行随机欠采样后,通过非线性优化算法求解带约束的范数最小化问题,可恢复出在变换域具有稀疏性的MR图像。 目的：为了增强图像在变换域中的稀疏性,改善MR图像重建质量,提出了对待重建图像的稀疏表示进行加权的方法。 方法：采用非线性共轭梯度下降算法求解该加权范数最小化问题,在迭代过程中,根据所求取的图像稀疏表示来更新权值矩阵,增强MR图像的稀疏性。 结果与结论：通过比较带加权矩阵和不带加权矩阵的压缩感知图像重建方法,结果表明带加权矩阵改进的算法提高了图像重建能力。     

Spatial Correspondence Between Functional MRI (fMRI) Activations and Cortical Current Density Maps of Event-Related Potentials (ERP): A Study with Four Tasks

We investigated the spatial correspondence between functional MRI (fMRI) activations and cortical current density maps of event-related potentials (ERPs) reconstructed without fMRI priors. The presence of a significant spatial correspondence is a prerequisite for direct integration of the two modalities, enabling to combine the high spatial resolution of fMRI with the high temporal resolution of ERPs. Four separate tasks were employed: visual stimulation with a pattern-reversal chequerboard, recognition of images of nameable objects, recognition of written words, and auditory stimulation with a piano note. ERPs were acquired with 19 recording channels, and source localisation was performed using a realistic head model, a standard cortical mesh and the multiple sparse priors method. Spatial correspondence was evaluated at group level over 10 subjects, by means of a voxel-by-voxel test and a test on the distribution of local maxima. Although not complete, it was significant for the visual stimulation task, image and word recognition tasks (P < 0.001 for both types of test), but not for the auditory stimulation task. These findings indicate that partial but significant spatial correspondence between the two modalities can be found even with a small number of channels, for three of the four tasks employed. Absence of correspondence for the auditory stimulation task was caused by the unfavourable situation of the activated cortex being perpendicular to the overlying scalp, whose consequences were exacerbated by the small number of channels. The present study corroborates existing literature in this field, and may be of particular relevance to those interested in combining fMRI with ERPs acquired with the standard 10-20 system.     

新的独立成分分析算法结合主成分分析实现fMRI信号的盲分离

用ICA算法来实现fMRI信号的盲源分离,可以提取出产生fMRI信号的多种源信号。但是在处理过程中存在两个困难:(1)fMRI数据的规模比较大,计算耗时;(2)计算量太大难免产生误差,给结果的分析带来不便。所以我们考虑对数据进行降维,但是如何确定源信号的个数也是一个难题。我们利用信息论的方法来估计源信号的个数,再使用主成分分析对数据进行降维。通过这样的处理,有效地确定了源信号的个数,减少了计算量。然后将一种新的ICA算法(New fixed-point,NewFP)用于处理降维后的数据。最后通过对实际的fMRI信号进行处理,结果表明新算法可以快速有效的分离fMRI信号,且准确性优于FastICA算法。     

A Study on Asymmetry of Spatial Visual Field by Analysis of the fMRI BOLD Response

The asymmetry of the left-right and upper-lower visual field is analyzed in this paper by a model approach based on the functional magnetic resonance imaging (fMRI) blood oxygenation level dependent (BOLD) response. The model consists of the convolution between a Gaussian function and the perfusion function of neural response to stimulus. The model parameters are estimated by a nonlinear optimal algorithm, and te asymmetry of the left-right and upper-lower visual field is investigated by the differences of the model parameters. The results from eight subjects show that reaction time is significant shorter and the response is significant stronger when the lower field is stimulated than that when the upper field is stimulated. For the left and right fields, the response is different. These results provide the fMRI BOLD response evidence of the asymmetry of spatial visual fields.     

Incorporating fMRI Functional Networks in EEG Source Imaging: A Bayesian Model Comparison Approach

Brain functional networks extracted from fMRI can improve the accuracy of EEG source localization. However, the coupling between EEG and fMRI remains poorly understood, i.e., whether fMRI networks provide information about the magnitude of neural activity, and whether neural sources demonstrate temporal correlations within each network. In this paper, we present an improved version of the NEtwork-based SOurce Imaging method (iNESOI) through Bayesian model comparison. Different models correspond to various matching between EEG and fMRI, and the appropriate one is selected by data with the model evidence. Synthetic and real data tests show that iNESOI has potential to select the appropriate fMRI priors to reach a better source reconstruction than some other typical approaches.