Independent component model of the default-mode brain function: combining individual-level and population-level analyses in resting-state fMRI

Resting-state functional magnetic resonance imaging (RS-fMRI) is a technique used to investigate the spontaneous correlations of blood-oxygen-level-dependent signals across different regions of the brain. Using functional connectivity tools, it is possible to investigate a specific RS-fMRI network, referred to as "default-mode" (DM) network, that involves cortical regions deactivated in fMRI experiments with cognitive tasks. Previous works have reported a significant effect of aging on DM regions activity. Independent component analysis (ICA) is often used for generating spatially distributed DM functional connectivity patterns from RS-fMRI data without the need for a reference region. This aspect and the relatively easy setup of an RS-fMRI experiment even in clinical trials have boosted the combined use of RS-fMRI and ICA-based DM analysis for noninvasive research of brain disorders. In this work, we considered different strategies for combining ICA results from individual-level and population-level analyses and used them to evaluate and predict the effect of aging on the DM component. Using RS-fMRI data from 20 normal subjects and a previously developed group-level ICA methodology, we generated group DM maps and showed that the overall ICA-DM connectivity is negatively correlated with age. A negative correlation of the ICA voxel weights with age existed in all DM regions at a variable degree. As an alternative approach, we generated a distributed DM spatial template and evaluated the correlation of each individual DM component fit to this template with age. Using a "leave-one-out" procedure, we discuss the importance of removing the bias from the DM template-generation process.     

Multivariate analysis of fMRI time series: classification and regression of brain responses using machine learning

Machine learning and pattern recognition techniques are being increasingly employed in functional magnetic resonance imaging (fMRI) data analysis. By taking into account the full spatial pattern of brain activity measured simultaneously at many locations, these methods allow detecting subtle, non-strictly localized effects that may remain invisible to the conventional analysis with univariate statistical methods. In typical fMRI applications, pattern recognition algorithms "learn" a functional relationship between brain response patterns and a perceptual, cognitive or behavioral state of a subject expressed in terms of a label, which may assume discrete (classification) or continuous (regression) values. This learned functional relationship is then used to predict the unseen labels from a new data set ("brain reading"). In this article, we describe the mathematical foundations of machine learning applications in fMRI. We focus on two methods, support vector machines and relevance vector machines, which are respectively suited for the classification and regression of fMRI patterns. Furthermore, by means of several examples and applications, we illustrate and discuss the methodological challenges of using machine learning algorithms in the context of fMRI data analysis.     

CORSICA: correction of structured noise in fMRI by automatic identification of ICA components

When applied to functional magnetic resonance imaging (fMRI) data, spatial independent component analysis (sICA), a data-driven technique that addresses the blind source separation problem, seems able to extract components specifically related to physiological noise and brain movements. These components should be removed from the data to achieve structured noise reduction and improve any subsequent detection and analysis of signal fluctuations related to neural activity. We propose a new automatic method called CORSICA (CORrection of Structured noise using spatial Independent Component Analysis) to identify the components related to physiological noise, using prior information on the spatial localization of the main physiological fluctuations in fMRI data. As opposed to existing spectral priors, which may be subject to aliasing effects for long-TR data sets (typically acquired with TR >1 s), such spatial priors can be applied to fMRI data, regardless of the TR of the acquisitions. By comparing the proposed automatic selection to a manual selection performed visually by a human operator, we first show that CORSICA is able to identify the noise-related components for long-TR data with a high sensitivity and a specificity of 1. On short-TR data sets, we validate that the proposed method of noise reduction allows a substantial improvement of the signal-to-noise ratio evaluated at the cardiac and respiratory frequencies, even in the gray matter, while preserving the main fluctuations related to neural activity.     

Performance of blind source separation algorithms for fMRI analysis using a group ICA method

Independent component analysis (ICA) is a popular blind source separation technique that has proven to be promising for the analysis of functional magnetic resonance imaging (fMRI) data. A number of ICA approaches have been used for fMRI data analysis, and even more ICA algorithms exist; however, the impact of using different algorithms on the results is largely unexplored. In this paper, we study the performance of four major classes of algorithms for spatial ICA, namely, information maximization, maximization of non-Gaussianity, joint diagonalization of cross-cumulant matrices and second-order correlation-based methods, when they are applied to fMRI data from subjects performing a visuo-motor task. We use a group ICA method to study variability among different ICA algorithms, and we propose several analysis techniques to evaluate their performance. We compare how different ICA algorithms estimate activations in expected neuronal areas. The results demonstrate that the ICA algorithms using higher-order statistical information prove to be quite consistent for fMRI data analysis. Infomax, FastICA and joint approximate diagonalization of eigenmatrices (JADE) all yield reliable results, with each having its strengths in specific areas. Eigenvalue decomposition (EVD), an algorithm using second-order statistics, does not perform reliably for fMRI data. Additionally, for iterative ICA algorithms, it is important to investigate the variability of estimates from different runs. We test the consistency of the iterative algorithms Infomax and FastICA by running the algorithm a number of times with different initializations, and we note that they yield consistent results over these multiple runs. Our results greatly improve our confidence in the consistency of ICA for fMRI data analysis.     

Fractional amplitude of low-frequency fluctuation changes in functional dyspepsia: A resting-state fMRI study

Recently, there is an increasing interest in the study of the role of brain dysfunction in the pathogenesis of symptoms of functional dyspepsia (FD). More specifically, abnormal brain activities in patients with FD during the resting state have been proven by several positron emission tomography (PET) studies. Resting-state functional magnetic resonance imaging (fMRI) is also a valuable tool in investigating spontaneous brain activity abnormalities in pathological conditions. In the present study, we examined the amplitude of low-frequency fluctuations (ALFF) and fractional (f)ALFF changes in patients with FD by using fMRI. Twenty-nine patients with FD and sixteen healthy controls participated in this study. Between-group differences in ALFF/fALFF were examined using a permutation-based nonparametric test after accounting for the gender and age effects. The results revealed a significant between-group difference in fALFF but not in ALFF in multiple brain regions including the right insula, brainstem and cerebellum. Seed-based resting-state functional connectivity analysis revealed that FD patients have increased correlations between the right cerebellum and multiple brain regions including the bilateral brainstem, bilateral cerebellum, bilateral thalamus, left para-/hippocampus, left pallidum and left putamen. Furthermore, fLAFF values in the right insula were positively correlated with the severity of the disease. These findings have provided further evidence of spontaneous brain activity abnormalities in FD patients which might contribute to our understanding of the pathophysiology of the disease.     

Computer-generated fMRI phantoms with motion-distortion interaction

As an extension of previous work on computer-generated phantoms, more accurate, realistic phantoms are generated by integrating image distortion and signal loss caused by susceptibility variations. With the addition of real motions and activations determined from actual functional MRI studies, these phantoms can be used by the fMRI community to assess with higher fidelity pre-processing algorithms such as motion correction, distortion correction and signal-loss compensation. These phantoms were validated by comparison to real echo-planar images. Specifically, studies have shown the effects of motion–distortion interactions on fMRI. We performed motion correction and activation analysis on these phantoms based on a block paradigm design using SPM2, and the results demonstrate that interactions between motion and distortion affect both motion correction and activation detection and thus represent a critical component of phantom generation.     

Hadamard-encoded sub-slice fMRI for reduced signal dropout

T2?-weighted blood oxygen level-dependent functional magnetic resonance imaging is adversely affected by susceptibility-induced field gradients in brain regions adjoining air interfaces that cause image distortion and signal dropout. Reducing slice thickness diminishes signal dropout but is accompanied by reduced signal-to-noise ratio (SNR). This study proposes simultaneous excitation of subslices with total width equal to the desired slice thickness, employing alternating Hadamard-encoded radiofrequency pulses coupled with incoherent addition of the subslices to achieve reduction of through-plane dephasing with minimal SNR loss but at the expense of a reduction in temporal resolution. Using a sensory task and hypercapnic challenge with breathholding (BH), results with two subslices per slice and a twofold reduction in temporal resolution show improved activation relative to a conventional acquisition. Average (eight subjects) T-scores in the BH task increased by 16% (P<.0003), and activation extent increased by 12% (not significant). In frontal brain regions, significant improvements in BH activation extent (11.4%, P<.05) and T-scores (18%, P<.0002) were demonstrated. Higher temporal resolution can be achieved by tradeoff of SNR.     

Independent component analysis of photoacoustic depth profiles

Photoacoustic spectroscopy is a powerful tool to non-destructively retrieve depth related information with minimal sample preparation from heterogeneous samples. Using a phase modulated step-scan arrangement and digital signal processing a series of spectra with effectively different modulation frequencies, probing different thermal diffusion lengths within a sample, can be collected simultaneously. In this report independent component analysis is used to analyze photoacoustic data acquired from a multilayer sample in an attempt to isolate the spectra of the constituent layers. The ICA findings are compared to the approach of eliciting this information through difference spectroscopy. The results suggest that ICA may offer substantial benefits for segmentation of PAS depth profiles without user intervention. This approach can be useful in analysis of complex heterogeneous samples when one has no prior knowledge of their constituents or their distribution in the sample.     

Surface-based functional magnetic resonance imaging analysis of partial brain echo planar imaging data at 1.5 T

Surface-based functional magnetic resonance imaging (fMRI) analysis is more sensitive and accurate than volume-based analysis for detecting neural activation. However, these advantages are less important in practical fMRI experiments with commonly used 1.5-T magnetic resonance devices because of the resolution gap between the echo planar imaging data and the cortical surface models. We expected high-resolution segmented partial brain echo planar imaging (EPI) data to overcome this problem, and the activation patterns of the high-resolution data could be different from the low-resolution data. For the practical applications of surface-based fMRI analysis using segmented EPI techniques, the effects of some important factors (e.g., activation patterns, registration and local distortions) should be intensively evaluated because the results of surface-based fMRI analyses could be influenced by them. In this study, we demonstrated the difference between activations detected from low-resolution EPI data, which were covering whole brain, and high-resolution segmented EPI data covering partial brain by volume- and surface-based analysis methods. First, we compared the activation maps of low- and high-resolution EPI datasets detected by volume- and surface-based analyses, with the spatial patterns of activation clusters, and analyzed the distributions of activations in occipital lobes. We also analyzed the high-resolution EPI data covering motor areas and fusiform gyri of human brain, and presented the differences of activations detected by volume- and surface-based methods.     

Exploring vision-related acupuncture point specificity with multivoxel pattern analysis

Acupoint specificity is one of the central issues of functional magnetic resonance imaging (fMRI) studies of acupuncture and has been under discussed. However, strong and consistent proof has not been provided for the existence of acupoint specificity, and unsuitable analysis approach applied could be the reason. We observed that previous researches of acupoint specificity were mostly based on model-based methods which were limited to make exploration of acupoint specificity because of the inaccurate specified prior. Here we applied multi-voxel pattern analysis (MVPA) to investigate the specificity of brain activation patterns induced by acupuncture stimulations at a vision-related acupoint (GB37) and a nearby nonacupoint (NAP). Results showed that multiple brain areas could differentiate the central neural response patterns induced by acupuncture stimulation at these two sites with higher accuracy above the chance level. These regions included occipital cortex, limbic-cerebellar areas and somatosensory cortex. Our results support that the characteristic neural response patterns of brain cortex to the acupuncture stimulation at GB37 and a nearby NAP could differ from each other effectively with the application of MVPA approach.     

基于独立成分分析法ICA的多光谱重建

构造了一个6通道,8-bit多光谱相机(MSC)模型作为图像采集装置(包括一个三色数码相机和一个滤光片),在多光谱成像技术基础上,通过模拟相机的数码响应,对物体表面的反射光谱进行重建。由数码相机的响应值得到物体的光谱反射比。使用独立成分分析法来进行数据压缩,光谱重建结果分别用CIE1976色差,光谱反射比RMS误差和反射比相对误差进行评价。20例训练样本的平均CIE1976色差为0.1579,平均反射比均方差为0.0099,平均反射比相对误差为0.0107;12例待测样品的平均CIE1976色差为0.7150,平均反射比均方差为0.0309,平均反射比相对误差为0.0312。结果说明,基于独立成分分析法的光谱重建是一种有效的光谱重建方法。     

Realignment strategies for awake-monkey fMRI data

Functional magnetic resonance imaging (fMRI) experiments with awake nonhuman primates (NHPs) have recently seen a surge of applications. However, the standard fMRI analysis tools designed for human experiments are not optimal for NHP data collected at high fields. One major difference is the experimental setup. Although real head movement is impossible for NHPs, MRI image series often contain visible motion artifacts. Animal body movement results in image position changes and geometric distortions. Since conventional realignment methods are not appropriate to address such differences, algorithms tailored specifically for animal scanning become essential. We have implemented a series of high-field NHP specific methods in a software toolbox, fMRI Sandbox (http://kyb.tuebingen.mpg.de/~stoewer/), which allows us to use different realignment strategies. Here we demonstrate the effect of different realignment strategies on the analysis of awake-monkey fMRI data acquired at high field (7 T). We show that the advantage of using a nonstandard realignment algorithm depends on the amount of distortion in the dataset. While the benefits for less distorted datasets are minor, the improvement of statistical maps for heavily distorted datasets is significant.     

Bayesian wavelet-based analysis of functional magnetic resonance time series

Wavelet methods for image regularization offer a data-driven alternative to Gaussian smoothing in functional magnetic resonance (fMRI) analysis. Their impact has been limited by the difficulties in integrating regularization in the wavelet domain and inference in the image domain, precluding the probabilistic decision on which areas are activated by a task. Here we present an integrated framework for Bayesian estimation and regularization in wavelet space that allows the usual voxelwise hypothesis testing. This framework is flexible, being an adaptation to fMRI time series of a more general wavelet-based functional mixed-effect model. Through testing on a combination of simulated and real fMRI data, we show evidence of improved signal recovery, without compromising test accuracy in image space.     

On the nature of the BOLD fMRI contrast mechanism

Since its development about 15 years ago, functional magnetic resonance imaging (fMRI) has become the leading research tool for mapping brain activity. The technique works by detecting the levels of oxygen in the blood, point by point, throughout the brain. In other words, it relies on a surrogate signal, resulting from changes in oxygenation, blood volume and flow, and does not directly measure neural activity. Although a relationship between changes in brain activity and blood flow has long been speculated, indirectly examined and suggested and surely anticipated and expected, the neural basis of the fMRI signal was only recently demonstrated directly in experiments using combined imaging and intracortical recordings. In the present paper, we discuss the results obtained from such combined experiments. We also discuss our current knowledge of the extracellularly measured signals of the neural processes that they represent and of the structural and functional neurovascular coupling, which links such processes with the hemodynamic changes that offer the surrogate signal that we use to map brain activity. We conclude by considering applications of invasive MRI, including injections of paramagnetic tracers for the study of connectivity in the living animal and simultaneous imaging and electrical microstimulation.     

Depression recognition using resting-state and event-related fMRI signals

Purpose

This paper aimed to develop a method for depression detection using blood-oxygen-level-dependent (BOLD) response estimated from event-related signals and resting-state functional magnetic resonance imaging (fMRI) signals together.

Materials and Methods

Thirteen patients with unipolar depression and matched healthy subjects were recruited. Resting state data of each subject were collected. Thereafter, event-related paradigm was undertaken using sad facial stimuli. The resting-state fMRI signal was deemed as the baseline of each subject's activity. Coefficient marks were designed to sort and select temporal independent components of event-related signals. Thereafter, stimulus-evoked BOLD response components inside event-related signal were extracted and taken as features to discriminate depressive patients from healthy controls.

Results

Accuracy rate for depression recognition was 77.27% with P value of .017 for whole-brain analysis and 81.82% with P value of .009 for region-of-interest analysis. The effectiveness and the superiority of the proposed method for disease recognition were demonstrated via the performance comparison with three other typical methods.

Conclusions

The proposed model was effective in depression recognition.     

Fixed-point algorithms for constrained ICA and their applications in fMRI data analysis

Constrained independent component analysis (CICA) eliminates the order ambiguity of standard ICA by incorporating prior information into the learning process to sort the components intrinsically. However, the original CICA (OCICA) and its variants depend on a learning rate, which is not easy to be tuned for various applications. To solve this problem, two learning-rate-free CICA algorithms were derived in this paper using the fixed-point learning concept. A complete stability analysis was provided for the proposed methods, which also made a correction to the stability analysis given to OCICA. Variations for adding constraints either to the components or to the associated time courses were derived too. Using synthetic data, the proposed methods yielded a better stability and a better source separation quality in terms of higher signal-to-noise-ratio and smaller performance index than OCICA. For the artificially generated brain activations, the new CICAs demonstrated a better sensitivity/specificity performance than standard univariate general linear model (GLM) and standard ICA. Original CICA showed a similar sensitivity/specificity gain but failed to converge for several times. Using functional magnetic resonance imaging (fMRI) data acquired with a well-characterized sensorimotor task, the proposed CICAs yielded better sensitivity than OCICA, standard ICA and GLM in all the target functional regions in terms of either higher t values or larger suprathreshold cluster extensions using the same significance threshold. In addition, they were more stable than OCICA and standard ICA for analyzing the sensorimotor fMRI data.     

Fractional amplitude of low-frequency fluctuation changes in monkeys with spinal cord injury: A resting-state fMRI study

Purpose

Although functional magnetic resonance imaging (fMRI) has revealed that spinal cord injury (SCI) causes anomalous changes in task-induced brain activation, its effect during the resting state remains unclear. The aim of this study is to explore the changes of the brain resting-state function in non-human primates with unilateral SCI.

Materials and methods

Eleven adult female rhesus monkeys were subjected to resting-state fMRI: five with unilateral thoracic SCI and six healthy monkeys, to obtain the fractional amplitude of low-frequency fluctuations (fALFF) of the blood oxygenation level-dependent (BOLD) contrast signal to determine the influence of SCI on the cerebral resting-state function.

Results

The SCI-induced fALFF vary significantly in several encephalic regions, including the left cerebellum, the left thalamus, the right lateral geniculate nucleus, the right superior parietal lobule, and the posterior cingulate gyrus.

Conclusion

Analysis of the resting-state fMRI provides evidence of abnormal spontaneous brain activations in primates with SCI, which may help us understand the pathophysiologic mechanisms underlying the changes in neural plasticity in the central nervous system after SCI.     

Separating 4D multi-task fMRI data of multiple subjects by independent component analysis with projection

Independent component analysis (ICA) is a widely accepted method to extract brain networks underlying cognitive processes from functional magnetic resonance imaging (fMRI) data. However, the application of ICA to multi-task fMRI data is limited due to the potential non-independency between task-related components. The ICA with projection (ICAp) method proposed by our group (Hum Brain Mapp 2009;30:417–31) is demonstrated to be able to solve the interactions among task-related components for single subject fMRI data. However, it still must be determined if ICAp is capable of processing multi-task fMRI data over a group of subjects. Moreover, it is unclear whether ICAp can be reliably applied to event-related (ER) fMRI data. In this study, we combined the projection method with the temporal concatenation method reported by Calhoun (Hum Brain Mapp 2008;29:828–38), referred to as group ICAp, to perform the group analysis of multi-task fMRI data. Both a human fMRI rest data-based simulation and real fMRI experiments, of block design and ER design, verified the feasibility and reliability of group ICAp, as well as demonstrated that ICAp had the strength to separate 4D multi-task fMRI data into multiple brain networks engaged in each cognitive task and to adequately find the commonalities and differences among multiple tasks.     

A quality control method for detecting and suppressing uncorrected residual motion in fMRI studies

Motion correction is an important step in the functional magnetic resonance imaging (fMRI) analysis pipeline. While many studies simply exclude subjects who are estimated to have moved beyond an arbitrary threshold, there exists no objective method for determining an appropriate threshold. Furthermore, any criterion based only upon motion estimation ignores the potential for proper realignment. The method proposed here uses unsupervised learning (specifically k-means clustering) on features derived from the mean square derivative (MSD) of the signal before and after realignment to identify problem data. These classifications are refined through analysis of correlation between subject activation maps and the mean activation map, as well as the relationship between tasking and motion as measured through regression of the canonical hemodynamic response functions to fit both estimated motion parameters and MSD. The MSD is further used to identify specific scans containing residual motion, data which is suppressed by adding nuisance regressors to the general linear model; this statistical suppression is performed for identified problem subjects, but has potential for use over all subjects. For problem subjects, our results show increased hemodynamic activity more consistent with group results; that is, the addition of nuisance regressors resulted in a doubling of the correlation between the activation map for the problem subjects and the activation map for all subjects. The proposed method should be useful in helping fMRI researchers make more efficient use of their data by reducing the need to exclude entire subjects from studies and thus collect new data to replace excluded subjects.     

Neural specificity of acupuncture stimulation from support vector machine classification analysis

Acupoint specificity, as a crucial issue in acupuncture neuroimaging studies, is still a controversial topic. Previous studies have generally adopted a block-based general linear model (GLM) approach, which predicts the temporal changes in the blood oxygenation level-dependent signal conforming to the “on-off” specifications. However, this method might become impractical since the precise timing and duration of acupuncture actions cannot be specified a priori. In the current study, we applied a data-driven multivariate classification approach, namely, support vector machine (SVM), to explore the neural specificity of acupuncture at gall bladder 40 (GB40) using kidney 3 (KI3) as a control condition (belonging to different meridians but the same nerve segment). In addition, to verify whether the typical GLM approach is sensitive enough in exploring the neural response patterns evoked by acupuncture, we also employed the GLM method to the same data sets. The SVM analysis detected distinct neural response patterns between GB40 and KI3 — positive predominantly for the GB40, while negative following the KI3. By contrast, group analysis from the GLM showed that acupuncture at these different acupoints can both evoke similar widespread signal decreases in multiple brain regions, and most of these regions were spatially overlapped, mainly distributing in the limbic and subcortical structures. Our findings may provide additional evidence to support the specificity of acupuncture, relevant to its clinical efficacy. Moreover, we also proved that GLM analysis is prone to be susceptible to errors and is not appropriate for detecting neural response patterns evoked by acupuncture stimulation.