独立成分分析及其应用的研究进展

独立成分分析(ICA)是一项把混合信号分解成具有统计独立性成分的新技术。ICA近年已在生物医学和雷达等领域的信号分离中展示了很好的应用前景。我们比较系统地介绍了ICA的基本原理、主要算法、应用和将来ICA研究的发展方向，旨在进一步推动有关的理论与应用研究工作。     

独立成分分析在视觉运动核磁共振数据处理中的应用

目的:验证独立成分分析(ICA)方法在处理视觉运动核磁共振数据中的有效性。方法:将ICA方法应用于视觉运动任务态的功能磁共振的数据处理。选用Fast ICA算法,根据有效的筛选标准选择最佳的独立成分,并将独立成分与功能模板数据进行比较。结果:选用Fast ICA算法进行数据的ICA处理,并选取成分8与功能数据进行对比。结果显示成分8显示的脑部活跃区域与功能数据较为相符。结论:采用Fast ICA方法所分离出来的独立成分,能够比较准确地显示脑部与运动视觉相关的活跃区域,同时也验证了ICA方法在分离视觉信息处理的背侧通路的有效性。     

基于独立分量分析的脑电噪声消除

作为一种新的多元统计处理方法，独立分量分析(ICA)是解决盲源分离(BSS)问题的一个有效手段。在简要分析ICA理论及其算法的基础上，提出将其应用到脑电中的眼电伪迹的去除任务。实际采集的生理信号大多由相互独立的成分线性迭加而成，符合ICA要求源信号统计独立的基本假设。与传统方法相比，ICA这种空间滤波器不受信号频谱混迭的限制，消噪的同时能对有用信号的细节成分做到很好的保留，很大程度上弥补了时频域方法的不足。此外解混矩阵的逆可以用来反映独立源的空间分布模式，具有重要的生理意义。     

功能磁共振图像的独立成分分析与提取

目的用独立成分分析方法(Independent Component Analysis,ICA)处理视觉任务态f MRI数据,并从f MRI信号中分离出任务相关和非相关的独立成分。方法运用快速不动点算法(Fast ICA)处理-功能磁共振数据,得到各独立成分的时间多元回归系数和时间进程图,结合对实验任务的分析,选取识别出各类独立成分。结果分别识别出视觉任务相关独立成分、类似周期信号独立成分、头动信号独立成分。结论把独立成分分析方法应用到对f MRI数据的处理当中,不仅能够找到真正与任务相关的独立成分,而且能够识别出其他相关因素引起的独立成分,从而为科研实验或图像的分析诊断提供科学依据。     

ICA在心音信号处理中的应用

独立成分分析(ICA)是近年来涌现的用于盲信号分离的新技术.本研究利用独立成分分析成功地把心音信号分离为三个独立的成分.借助ICA方法我们能够有效地区分正常人和房间隔缺损病人的心音信号.研究中所用的心音信号由自行研发的心音采集仪所采集.文章首先介绍了独立成分分析的基本原理,然后介绍了一种基于四阶统计量的算法-快速定点算法,并给出了利用此算法的ICA步骤.试验结果表明,ICA在心音信号的处理中是一种很有潜力的方法.     

一种基于独立成份分析的fMRI时-空模型数据处理方法

独立成份分析(ICA)是信号处理领域中斯近发展起来的一种很有应用前景的方法，而脑功能磁共振(fMRI)信号的有效分离与识别是一个正在研究和试验之中的技术领域。因此，发展基于ICA的fMRI数据处理方法具有明显的理论价值和应用前景。本文首先介绍了ICA原理，分析了现行ICA—fMRI方法采用的信号与噪声的空域分布相互独立的信号模型所存在的明显不足，然后提出了微域中的信号与噪声的时域过程相互独立的fMRI信号模型，从而建立了一种新的fMRI数据处理方法：邻域独立成份相关法。合理的fMRI实验数据处理结果验证了新方法的合理性。     

ROC分析及其在fMRI信号中的一个应用

详细阐述了受试者工作特征（Receiver operating characteristic，ROC）分析的基本原理，分析了它与其它的诊断评价性标准相比存在的优点，并给出了ROC曲线的绘制方法，最后采用ROC对独立成分分析（Independent component analysis，ICA）和统计参数图（Statistical parametric mapping，SPM）处理功能磁共振成像（Functional magnetic reasonance imaging，fMRI）数据的结果进行了比较。     

基于独立分量分析的生理信号盲源分离

用于盲源分离的独立分量分析（ICA）和扩展ICA算法，基于极大似然估计，给出一个衡量输出分量统计独立的目标函数，最优化目标函数，得到一种用于独立分量分析的迭代算法。扩展ICA算法的优点在于迭代过程中不需要计算信号的高阶统计量，收敛速度快，同时适用于超高斯和亚高斯信号的分离。应用该算法实现了脑电、心电信号以及语音信号的分离，并给了实验结果。     

基于独立成分分析的遗尿症儿童脑功能网络研究

目的:用独立成分分析（ICA）的方法对原发性单症状夜间遗尿症（PMNE）儿童的脑功能网络成分间连接进行研究。方法:采集35例PMNE儿童和25例健康儿童脑功能磁共振图像,通过ICA获得每个被试的脑功能网络成分,然后计算每个被试脑功能网络成分间的功能连接强度,比较PMNE儿童与健康对照组的强度差异。结果:与对照组对比,PMNE患者的右侧执行控制网络与默认模式网络、左侧执行控制网络均存在功能连接异常（FDR, P<0.05）。结论:PMNE儿童存在脑功能网络成分间的连接异常,这可能为理解PMNE儿童的病理机制提供一些新的影像学依据。     

独立分量分析在功能磁共振成像信号功能活动区检测中的应用

主要讨论独立分量分析(ICA)在功能磁共振成像(fMRI)信号功能区检测中的应用。fMRI利用血氧水平依赖(BOLD)效应成像，根据大脑神经元兴奋后局部血氧饱和度增高的原理间接显示神经元活动。假设fMRI信号中包含反映血氧饱和度事件相关的信号、生理噪声和仪器产生的随机噪声等独立分量，首先对fMRI信号进行去噪、配准等预处理，然后利用fastlCA算法对独立分量进行分离，有效抑制噪声对功能区检测的影响，利用相关原理检测出fMRI信号的功能活动区。     

Analysis of fMRI data by blind separation of data in a tiny spatial domain into independent temporal component

Independent Component Analysis (ICA) is a promising tool for the analysis of functional magnetic resonance imaging (fMRI) time series. In these studies, mostly assumed is a spatially independent component map of fMRI data (spatial ICA). In this paper, we assume that the temporal courses of the signal and noises are independent within a Tiny spatial domain (temporal ICA). Then with fast-ICA algorithm, spatially neighboring fMRI data were blindly separated into several temporal courses and were preassumed to be formed by a signal time course and several noise time courses where the signal has the largest correlation coefficient with the reference signal. The final functional imaging was completed for the signals obtained from each voxel. Simulations showed that compared with the spatial ICA method, the new temporal ICA method is more effective than the spatial ICA in detecting weak signal in a fMRI dataset. As background noise, the simulations include simulated Gaussian noise and fMRI data without stimulation. Finally, vivo fMRI tests showed that the excited areas evoked by a visual stimuli are mainly in the region of the primary visual cortex and that evoked by auditory stimuli are mainly in the region of the primary temporal cortex.     

独立成分分析在生物医学信号处理中的应用

独立成分分析(independentcomponentanalysis熏ICA)已经成功地应用到生物医学信号处理中,并被证明是一种分析生物医学信号的强有力的工具,近年来一直受到国内外学者的广泛关注。本文系统地介绍了独立成分分析在生物医学信号(EEG,MEG,fMRI)处理中的应用,分析了其应用方法,最后简要地探讨了独立成分分析应用到生物医学信号中的优势及存在的一些不足。     

基于独立成分分析的时间自相关方法在功能磁共振激活区自适应提取中的应用

传统基于ICA的激活区检测手段是将分离后的独立成分与参考信号做相关性分析。实际问题中,不同区域的脑血流动力学响应情况不同,因此往往得不到标准的参考信号。针对此类问题,提出时间自相关方法(TSC)与ICA方法结合,在不需要参考信号的情况下,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。应用5 邻域ICA方法对fMRI数据逐点处理,然后应用时间自相关算法检测各时间序列周期间的相关性,选择最大的自相关系数作为该体素点的信号值。再通过Z变换将相关系数分布转换为服从N(0,1)的Z分布,提取出具有显著性差异（a=0.05）的激活区。将自相关算法应用于仿真数据和12组双手握拳运动的真实fMRI数据的处理,结果表明该方法能够准确提取出仿真数据中的激活区。对真实数据的处理,该方法在空间准确性上与GLM方法无显著性差别(0.4653±0.1368 vs 0.4905±0.1341),在时间准确性上显著优于GLM方法 (0.6364±0.0111 vs 0.3692±0.0109),具有良好的脑功能激活区检测及空间定位能力。     

An Improved Fixed-point Algorithm for Independent Component Analysis of Functional MRI Data

The fixed-point algorithm and infomax algorithm are two of the most popular algorithms in independent component analysis （ICA）. However, it is hard to take both stability and speed into consideration in processing functional magnetic resonance imaging （fMRI） data. In this paper, an optimization model for ICA is presented and an improved fixed-point algorithm based on the model is proposed. In the new algorithms a small step size is added to increase the stability. In order to accelerate the convergence, an improvement on Newton method is made, which makes cubic convergence for the new algorithm. Applying the algorithm and two other algorithms to invivo fMRI data, the results show that the new algorithm separates independent components stably, which has faster convergence speed and less computation than the other two algorithms. The algorithm has obvious advantage in processing fMRI signal with huge data.     

Investigation of human brain hemodynamics by simultaneous near-infrared spectroscopy and functional magnetic resonance imaging

The aim of this study was to compare functional cerebral hemodynamic signals obtained simultaneously by near infrared spectroscopy (NIRS) and by functional magnetic resonance imaging (fMRI). The contribution of superficial layers (skin and skull) to the NIRS signal was also assessed. Both methods were used to generate functional maps of the motor cortex area during a periodic sequence of stimulation by finger motion and rest. In all subjects we found a good collocation of the brain activity centers revealed by both methods. We also found a high temporal correlation between the BOLD signal (fMRI) and the deoxy-hemoglobin concentration (NIRS) in the subjects who exhibited low fluctuations in superficial head tissues.     

一种基于稀疏分解去除EEG信号中MRI伪迹的新方法

在脑电图(Electroencephalography,EEG)和功能磁共振成像(Functional magnetic resonance imaging, FMRI)同时记录时,如何有效的去除混入EEG信号中的强磁共振(Magnetic resonance imaging,MRI)伪迹干扰信号是当前在EEG和FMRI的联合研究中面临的一个信号前期处理难点。主要从MRI干扰信号和EEG信号在时空上的差别出发,提出了一种基于混合过完备库的稀疏成分分析的分解方法,实现了强MRI干扰下的EEG信号的估计。在方法实现中,首先利用小波和离散余弦构造能体现MRI干扰和EEG时空特性差别的混合过完备库,然后通过匹配追踪(Matching pursuit,MP)方法在混合过完备库中的学习,实现MRI伪迹的消除。对模拟数据以及真实记录的混入了MRI干扰的EEG信号的估计实验结果,证实了该方法的有效性。     

Decomposition of working memory-related scalp ERPs: crossvalidation of fMRI-constrained source analysis and ICA.

Both functional magnetic resonance imaging (fMRI)-constrained source analysis and independent component analysis (ICA) claim to estimate the neuronal sources of electroencephalographic (EEG) scalp signals. In fMRI-constrained source analysis, event-related potential (ERP) generator locations are defined by fMRI activation patterns, and their contribution to the scalp ERP signal is probed. In contrast, ICA assumes that networks of cortical generators can be separated on the basis of their statistical independence. While good arguments can be put forward to justify both approaches, it is unclear how results from both methods compare. A clarification of these issues is of utmost importance to reconcile findings made using identical paradigms but these two complementary analysis methods. As both methods share the concept of spatially static sources a natural space to compare both methods and to crossvalidate the respective findings is at the level of source activity in the form of dipole source waves and independent component time courses and their corresponding maps. We used fMRI-constrained source analysis and ICA followed by clustering using the Kuhn-Munkres algorithm to analyze data from a working memory experiment. We demonstrate that crossvalidation is indeed possible using an appropriate statistical test. However, the sensitivity of this crossvalidation approach is ultimately limited by the low number of dimensions that contribute significant variance to the grand average scalp ERP. We conclude that testing at the single-subject level is preferable for crossvalidation purposes if the signal-to-noise ratio of the data allows for this approach.     

Data-based functional template for sorting independent components of fMRI activity

In human brain imaging with naturalistic stimuli, hemodynamic responses are difficult to predict and thus data-driven approaches, such as independent component analysis (ICA), may be beneficial. Here we propose inter-subject correlation (ISC) maps as stimulus-sensitive functional templates for sorting the independent components (ICs) to identify the most stimulus-related networks without stimulus-dependent temporal covariates. We collected 3-T functional magnetic resonance imaging (fMRI) data during perception of continuous audiovisual speech. Ten adults viewed a video, in which speech intelligibility was varied by altering the sound level. Five ICs with strongest overlap with the ISC map comprised auditory and visual cortices, and the sixth was a left-hemisphere-dominant network (left posterior superior temporal sulcus, inferior frontal gyrus, anterior superior temporal pole, supplementary motor cortex, and right angular gyrus) that was activated stronger during soft than loud speech. Corresponding temporal-model-based analysis revealed only temporal- and parietal-lobe activations without involvement of the anterior areas. The performance of the ISC-based IC selection was confirmed with fMRI data collected during free viewing of movie. Since ISC–ICA requires no predetermined temporal models on stimulus timing, it seems feasible for fMRI studies where hemodynamic variations are difficult to model because of the complex temporal structure of the naturalistic stimulation.     

Isometric force-related activity in sensorimotor cortex measured with functional MRI

Isometric force-related functional magnetic resonance imaging (fMRI) signals from primary sensorimotor cortex were investigated by imaging during a sustained finger flexion task at a number of force levels related to maximum voluntary contraction. With increasing levels of force, there was an increase in the extent along the central sulcus from which a fMRI signal could be detected and an increase in the summed signal across voxels, but these parameters were related in such a way that the signal from each voxel was similar for each level of force. The results suggest that increased neuronal firing and recruitment of corticomotor cells associated with increased voluntary isometric effort are reflected in an expansion of a relatively constant fMRI signal over a greater volume of cortex, rather than an increase in the magnitude of the response in a particular circumscribed region, possibly due to perfusion of an increase in oxygen-enriched blood over a wider region of the cortex. Received: 16 June 1997 / Accepted: 3 November 1998