Classification of fMRI independent components using IC-fingerprints and support vector machine classifiers

Functional connectivity MR imaging of the language network in patients with drug-resistant epilepsy

BACKGROUND AND PURPOSE

Subtle linguistic dysfunction and reorganization of the language network were described in patients with epilepsy, suggesting the occurrence of plasticity changes. We used resting state FC-MRI to investigate the effects induced by chronic epilepsy on the connectivity of the language-related brain regions and correlated it with language performance.

MATERIALS AND METHODS

FC-MRI was evaluated in 22 right-handed patients with drug-resistant epilepsy (11 with LE and 11 with RE) and in 12 healthy volunteers. Neuropsychological assessment of verbal IQ was performed. Patients and controls underwent BOLD fMRI with a verb-generation task, and language function was lateralized by an LI. Intrinsic activity fluctuations for FC analysis were extracted from data collected during the task. Six seeding cortical regions for speech in both hemispheres were selected to obtain a measure of the connectivity pattern among the language networks.

RESULTS

Patients with LE presented atypical language lateralization and an overall reduced connectivity of the language network with respect to controls. In patients with both LE and RE, the mean FC was significantly reduced within the left (dominant) hemisphere and between the 2 hemispheres. In patients with LE, there was a positive correlation between verbal IQ scores and the left intrahemispheric FC.

CONCLUSIONS

In patients with intractable epilepsy, FC-MRI revealed an overall reduction and reorganization of the connectivity pattern within the language network. FC was reduced in the left hemisphere regardless of the epileptogenic focus side and was positively correlated with linguistic performance only in patients with LE.

Cognitive and default-mode resting state networks: do male and female brains "rest" differently?

Variability in human behavior related to sex is supported by neuroimaging studies showing differences in brain activation patterns during cognitive task performance. An emerging field is examining the human connectome, including networks of brain regions that are not only temporally-correlated during different task conditions, but also networks that show highly correlated spontaneous activity during a task-free state. Both task-related and task-free network activity has been associated with individual task performance and behavior under certain conditions. Therefore, our aim was to determine whether sex differences exist during a task-free resting state for two networks associated with cognitive task performance (executive control network (ECN), salience network (SN)) and the default mode network (DMN). Forty-nine healthy subjects (26 females, 23 males) underwent a 5-min task-free fMRI scan in a 3T MRI. An independent components analysis (ICA) was performed to identify the best-fit IC for each network based on specific spatial nodes defined in previous studies. To determine the consistency of these networks across subjects we performed self-organizing group-level ICA analyses. There were no significant differences between sexes in the functional connectivity of the brain areas within the ECN, SN, or the DMN. These important findings highlight the robustness of intrinsic connectivity of these resting state networks and their similarity between sexes. Furthermore, our findings suggest that resting state fMRI studies do not need to be controlled for sex.

Multimodal neuroimaging in patients with disorders of consciousness showing "functional hemispherectomy"

Beside behavioral assessment of patients with disorders of consciousness, neuroimaging modalities may offer objective paraclinical markers important for diagnosis and prognosis. They provide information on the structural location and extent of brain lesions (e.g., morphometric MRI and diffusion tensor imaging (DTI-MRI) assessing structural connectivity) but also their functional impact (e.g., metabolic FDG-PET, hemodynamic fMRI, and EEG measurements obtained in "resting state" conditions). We here illustrate the role of multimodal imaging in severe brain injury, presenting a patient in unresponsive wakefulness syndrome (UWS; i.e., vegetative state, VS) and in a "fluctuating" minimally conscious state (MCS). In both cases, resting state FDG-PET, fMRI, and EEG showed a functionally preserved right hemisphere, while DTI showed underlying differences in structural connectivity highlighting the complementarities of these neuroimaging methods in the study of disorders of consciousness.

Semiblind spatial ICA of fMRI using spatial constraints

Independent component analysis (ICA) utilizing prior information, also called semiblind ICA, has demonstrated considerable promise in the analysis of functional magnetic resonance imaging (fMRI). So far, temporal information about fMRI has been used in temporal ICA or spatial ICA as additional constraints to improve estimation of task-related components. Considering that prior information about spatial patterns is also available, a semiblind spatial ICA algorithm utilizing the spatial information was proposed within the framework of constrained ICA with fixed-point learning. The proposed approach was first tested with synthetic fMRI-like data, and then was applied to real fMRI data from 11 subjects performing a visuomotor task. Three components of interest including two task-related components and the "default mode" component were automatically extracted, and atlas-defined masks were used as the spatial constraints. The default mode network, a set of regions that appear correlated in particular in the absence of tasks or external stimuli and is of increasing interest in fMRI studies, was found to be greatly improved when incorporating spatial prior information. Results from simulation and real fMRI data demonstrate that the proposed algorithm can improve ICA performance compared to a different semiblind ICA algorithm and a standard blind ICA algorithm.

Topographic electrophysiological signatures of FMRI Resting State Networks

Background

fMRI Resting State Networks (RSNs) have gained importance in the present fMRI literature. Although their functional role is unquestioned and their physiological origin is nowadays widely accepted, little is known about their relationship to neuronal activity. The combined recording of EEG and fMRI allows the temporal correlation between fluctuations of the RSNs and the dynamics of EEG spectral amplitudes. So far, only relationships between several EEG frequency bands and some RSNs could be demonstrated, but no study accounted for the spatial distribution of frequency domain EEG.

Methodology/Principal Findings

In the present study we report on the topographic association of EEG spectral fluctuations and RSN dynamics using EEG covariance mapping. All RSNs displayed significant covariance maps across a broad EEG frequency range. Cluster analysis of the found covariance maps revealed the common standard EEG frequency bands. We found significant differences between covariance maps of the different RSNs and these differences depended on the frequency band.

Conclusions/Significance

Our data supports the physiological and neuronal origin of the RSNs and substantiates the assumption that the standard EEG frequency bands and their topographies can be seen as electrophysiological signatures of underlying distributed neuronal networks.

Whole brain resting-state analysis reveals decreased functional connectivity in major depression

Recently, both increases and decreases in resting-state functional connectivity have been found in major depression. However, these studies only assessed functional connectivity within a specific network or between a few regions of interest, while comorbidity and use of medication was not always controlled for. Therefore, the aim of the current study was to investigate whole-brain functional connectivity, unbiased by a priori definition of regions or networks of interest, in medication-free depressive patients without comorbidity. We analyzed resting-state fMRI data of 19 medication-free patients with a recent diagnosis of major depression (within 6 months before inclusion) and no comorbidity, and 19 age- and gender-matched controls. Independent component analysis was employed on the concatenated data sets of all participants. Thirteen functionally relevant networks were identified, describing the entire study sample. Next, individual representations of the networks were created using a dual regression method. Statistical inference was subsequently done on these spatial maps using voxel-wise permutation tests. Abnormal functional connectivity was found within three resting-state networks in depression: (1) decreased bilateral amygdala and left anterior insula connectivity in an affective network, (2) reduced connectivity of the left frontal pole in a network associated with attention and working memory, and (3) decreased bilateral lingual gyrus connectivity within ventromedial visual regions. None of these effects were associated with symptom severity or gray matter density. We found abnormal resting-state functional connectivity not previously associated with major depression, which might relate to abnormal affect regulation and mild cognitive deficits, both associated with the symptomatology of the disorder.

Unbiased group-level statistical assessment of independent component maps by means of automated retrospective matching

This report presents and validates a method for the group-level statistical assessment of independent component analysis (ICA) outcomes. The method is based on a matching of individual component maps to corresponding aggregate maps that are obtained from concatenated data. Group-level statistics are derived that include an explicit correction for selection bias. Outcomes were validated by means of calculations with artificial null data. Although statistical inferences were found to be incorrect if bias was neglected, the use of the proposed bias correction sufficed to obtain valid results. This was further confirmed by extensive calculations with artificial data that contained known effects of interest. While uncorrected statistical assessments systematically violated the imposed confidence level thresholds, the corrected method was never observed to exceed the allowed false positive rate. Yet, bias correction was found to result in a reduced sensitivity and a moderate decrease in discriminatory power. The method was also applied to analyze actual fMRI data. Various effects of interest that were detectable in the aggregate data were similarly revealed by the retrospective matching method. In particular, stimulus-related responses were extensive. Nevertheless, differences were observed regarding their spatial distribution. The presented findings indicate that the proposed method is suitable for neuroimaging analyses. Finally, a number of generalizations are discussed. It is concluded that the proposed method provides a framework that may supplement many of the currently available group ICA methods with validated unbiased group inferences.

Analysis of functional networks involved in motor execution and motor imagery using combined hierarchical clustering analysis and independent component analysis

Cognitive experiments involving motor execution (ME) and motor imagery (MI) have been intensively studied using functional magnetic resonance imaging (fMRI). However, the functional networks of a multitask paradigm which include ME and MI were not widely explored. In this article, we aimed to investigate the functional networks involved in MI and ME using a method combining the hierarchical clustering analysis (HCA) and the independent component analysis (ICA). Ten right-handed subjects were recruited to participate a multitask experiment with conditions such as visual cue, MI, ME and rest. The results showed that four activation clusters were found including parts of the visual network, ME network, the MI network and parts of the resting state network. Furthermore, the integration among these functional networks was also revealed. The findings further demonstrated that the combined HCA with ICA approach was an effective method to analyze the fMRI data of multitasks.

Persistent genital arousal disorder associated with functional hyperconnectivity of an epileptic focus

Persistent Genital Arousal Disorder (PGAD) refers to the experience of persistent sensations of genital arousal that are felt to be unprovoked, intrusive and unrelieved by one or several orgasms. It is often mistaken for hypersexuality since PGAD often results in a high frequency of sexual behaviour. At present little is known with certainty about the etiology of this condition. We described a woman with typical PGAD symptoms and orgasmic seizures that we found to be related to a specific epileptic focus. We performed a EEG/MEG and fMRI spontaneous activity study during genital arousal symptoms and after the chronic administration of 300 mg/day of topiramate. From MEG data an epileptic focus was localized in the left posterior insular gyrus (LPIG). FMRI data evidenced that sexual excitation symptoms with PGAD could be correlated with an increased functional connectivity (FC) between different brain areas: LPIG (epileptic focus), left middle frontal gyrus, left inferior and superior temporal gyrus and left inferior parietal lobe. The reduction of the FC observed after antiepileptic therapy was more marked in the left than in the right hemisphere in agreement with the lateralization identified by MEG results. Treatment completely abolished PGAD symptoms and functional hyperconnectivity. The functional hyperconnectivity found in the neuronal network including the epileptic focus could suggest a possible central mechanism for PGAD.

Functional segmentation of the brain cortex using high model order group PICA

Baseline activity of resting state brain networks (RSN) in a resting subject has become one of the fastest growing research topics in neuroimaging. It has been shown that up to 12 RSNs can be differentiated using an independent component analysis (ICA) of the blood oxygen level dependent (BOLD) resting state data. In this study, we investigate how many RSN signal sources can be separated from the entire brain cortex using high dimension ICA analysis from a group dataset. Group data from 55 subjects was analyzed using temporal concatenation and a probabilistic independent component analysis algorithm. ICA repeatability testing verified that 60 of the 70 computed components were robustly detectable. Forty-two independent signal sources were identifiable as RSN, and 28 were related to artifacts or other noninterest sources (non-RSN). The depicted RSNs bore a closer match to functional neuroanatomy than the previously reported RSN components. The non-RSN sources have significantly lower temporal intersource connectivity than the RSN (P < 0.0003). We conclude that the high model order ICA of the group BOLD data enables functional segmentation of the brain cortex. The method enables new approaches to causality and connectivity analysis with more specific anatomical details.

Random subspace ensembles for FMRI classification

Classification of brain images obtained through functional magnetic resonance imaging (fMRI) poses a serious challenge to pattern recognition and machine learning due to the extremely large feature-to-instance ratio. This calls for revision and adaptation of the current state-of-the-art classification methods. We investigate the suitability of the random subspace (RS) ensemble method for fMRI classification. RS samples from the original feature set and builds one (base) classifier on each subset. The ensemble assigns a class label by either majority voting or averaging of output probabilities. Looking for guidelines for setting the two parameters of the method-ensemble size and feature sample size-we introduce three criteria calculated through these parameters: usability of the selected feature sets, coverage of the set of "important" features, and feature set diversity. Optimized together, these criteria work toward producing accurate and diverse individual classifiers. RS was tested on three fMRI datasets from single-subject experiments: the Haxby data (Haxby, 2001.) and two datasets collected in-house. We found that RS with support vector machines (SVM) as the base classifier outperformed single classifiers as well as some of the most widely used classifier ensembles such as bagging, AdaBoost, random forest, and rotation forest. The closest rivals were the single SVM and bagging of SVM classifiers. We use kappa-error diagrams to understand the success of RS.

Default network connectivity reflects the level of consciousness in non-communicative brain-damaged patients

The 'default network' is defined as a set of areas, encompassing posterior-cingulate/precuneus, anterior cingulate/mesiofrontal cortex and temporo-parietal junctions, that show more activity at rest than during attention-demanding tasks. Recent studies have shown that it is possible to reliably identify this network in the absence of any task, by resting state functional magnetic resonance imaging connectivity analyses in healthy volunteers. However, the functional significance of these spontaneous brain activity fluctuations remains unclear. The aim of this study was to test if the integrity of this resting-state connectivity pattern in the default network would differ in different pathological alterations of consciousness. Fourteen non-communicative brain-damaged patients and 14 healthy controls participated in the study. Connectivity was investigated using probabilistic independent component analysis, and an automated template-matching component selection approach. Connectivity in all default network areas was found to be negatively correlated with the degree of clinical consciousness impairment, ranging from healthy controls and locked-in syndrome to minimally conscious, vegetative then coma patients. Furthermore, precuneus connectivity was found to be significantly stronger in minimally conscious patients as compared with unconscious patients. Locked-in syndrome patient's default network connectivity was not significantly different from controls. Our results show that default network connectivity is decreased in severely brain-damaged patients, in proportion to their degree of consciousness impairment. Future prospective studies in a larger patient population are needed in order to evaluate the prognostic value of the presented methodology.

Modulation by EEG features of BOLD responses to interictal epileptiform discharges

INTRODUCTION

EEG-fMRI of interictal epileptiform discharges (IEDs) usually assumes a fixed hemodynamic response function (HRF). This study investigates HRF variability with respect to IED amplitude fluctuations using independent component analysis (ICA), with the goal of improving the specificity of EEG-fMRI analyses.

METHODS

We selected EEG-fMRI data from 10 focal epilepsy patients with a good quality EEG. IED amplitudes were calculated in an average reference montage. The fMRI data were decomposed by ICA and a deconvolution method identified IED-related components by detecting time courses with a significant HRF time-locked to the IEDs (F-test, p<0.05). Individual HRF amplitudes were then calculated for each IED. Components with a significant HRF/IED amplitude correlation (Spearman test, p<0.05) were compared to the presumed epileptogenic focus and to results of a general linear model (GLM) analysis.

RESULTS

In 7 patients, at least one IED-related component was concordant with the focus, but many IED-related components were at distant locations. When considering only components with a significant HRF/IED amplitude correlation, distant components could be discarded, significantly increasing the relative proportion of activated voxels in the focus (p=0.02). In the 3 patients without concordant IED-related components, no HRF/IED amplitude correlations were detected inside the brain. Integrating IED-related amplitudes in the GLM significantly improved fMRI signal modeling in the epileptogenic focus in 4 patients (p<0.05).

CONCLUSION

Activations in the epileptogenic focus appear to show significant correlations between HRF and IED amplitudes, unlike distant responses. These correlations could be integrated in the analysis to increase the specificity of EEG-fMRI studies in epilepsy.

Detection of irregular, transient fMRI activity in normal controls using 2dTCA: comparison to event-related analysis using known timing

When events occur spontaneously during the acquisition of a series of images, traditional modeling methods for detecting functional MRI activation detection cannot be employed. The two-dimensional temporal clustering algorithm, 2dTCA, has been shown to accurately detect random, transient activations in computer simulations without the use of known event timings. In this study we applied the 2dTCA technique to detect the timings and spatial locations of sparse, irregular, transient activations of the visual, auditory, and motor cortices in 12 normal controls. Experiments with one and two independent types of stimuli were employed. Event-related activation using known timing was compared with event-related activation using 2dTCA-detected timing in individuals and across groups. The 2dTCA algorithm detected the activation from all presented stimuli in every subject. When compared with block-design results using a measure of correlation between activation maps, no significant difference was found between the 2dTCA activation maps and the event-related maps using known timing across all subjects. Therefore, 2dTCA has the potential to be an accurate and more practical method for detection of spontaneous, transient events using fMRI.

Applications of multivariate pattern classification analyses in developmental neuroimaging of healthy and clinical populations

Analyses of functional and structural imaging data typically involve testing hypotheses at each voxel in the brain. However, it is often the case that distributed spatial patterns may be a more appropriate metric for discriminating between conditions or groups. Multivariate pattern analysis has been gaining traction in neuroimaging of adult healthy and clinical populations; studies have shown that information present in neuroimaging data can be used to decode intentions and perceptual states, as well as discriminate between healthy and diseased brains. While few studies to date have applied these methods in pediatric populations, in this review we discuss exciting potential applications for studying both healthy, and aberrant, brain development. We include an overview of methods and discussion of challenges and limitations.

Noxious somatosensory stimulation affects the default mode of brain function: evidence from functional MR imaging

PURPOSE

To investigate whether default mode network (DMN) spatial properties can be directly affected by pain, with a comparison of painful and nonpainful conditions.

MATERIALS AND METHODS

The authors performed a functional magnetic resonance (MR) imaging study, approved by the local institutional ethics committee, involving 10 healthy male subjects (age range, 18-45 years) who gave written informed consent. The subjects underwent two experimental sessions of median nerve electrical stimulation at painful and nonpainful levels. Independent component analysis of the functional MR imaging data was performed to determine the DMN spatiotemporal pattern. Group-level DMN connectivity maps for painful and nonpainful conditions were obtained (P < .001, corrected with false discovery rate). The contrast between the connectivity maps in the two conditions was also computed (P < .05, corrected with false discovery rate).

RESULTS

The DMN maintained its typical temporal properties but was subject to modifications in connectivity pattern during painful stimulation, affecting the brain areas associated with pain processing. Increased connectivity in painful conditions was found mainly in the left prefrontal cortex and posterior cingulate cortex-precuneus, and decreased connectivity was found in the lateral parietal cortex.

CONCLUSION

Study findings were in line with the impairments of the DMN reported in patients with chronic pain. They support the hypothesis that alteration of the DMN connectivity pattern localized in specific brain areas during acute pain, if repeated across time, might induce permanent changes that could disrupt the DMN functional architecture.

Independent component analysis as a model-free approach for the detection of BOLD changes related to epileptic spikes: a simulation study

EEG-fMRI in epileptic patients is commonly analyzed using the general linear model (GLM), which assumes a known hemodynamic response function (HRF) to epileptic spikes in the EEG. In contrast, independent component analysis (ICA) can extract Blood-Oxygenation Level Dependent (BOLD) responses without imposing constraints on the HRF. This technique was evaluated on data generated by superimposing artificial responses on real background fMRI signals. Simulations were run using a wide range of EEG spiking rates, HRF amplitudes, and activation regions. The data were decomposed by spatial ICA into independent components. A deconvolution method then identified component time courses significantly related to the simulated spikes, without constraining the shape of the HRF. Components matching the simulated activation regions ("concordant components") were found in 84.4% of simulations, while components at discordant locations were found in 12.2% of simulations. These false activations were often related to large artifacts that coincidentally occurred simultaneously with some of the random simulated spikes. The performance of the method depended closely on the simulation parameters; when the number of spikes was low, concordant components could only be identified when HRF amplitudes were large. Although ICA did not depend on the shape of the HRF, data processed with the GLM did not reveal the appropriate activation region when the HRF varied slightly from the canonical shape used in the model. ICA may thus be able to extract BOLD responses from EEG-fMRI data in epileptic patients, in a way that is robust to uncertainty and variability in the shape of the HRF.

Two systems of resting state connectivity between the insula and cingulate cortex

The insula and cingulate cortices are implicated in emotional, homeostatic/allostatic, sensorimotor, and cognitive functions. Non-human primates have specific anatomical connections between sub-divisions of the insula and cingulate. Specifically, the anterior insula projects to the pregenual anterior cingulate cortex (pACC) and the anterior and posterior mid-cingulate cortex (aMCC and pMCC); the mid-posterior insula only projects to the posterior MCC (pMCC). In humans, functional neuroimaging studies implicate the anterior insula and pre/subgenual ACC in emotional processes, the mid-posterior insula with awareness and interoception, and the MCC with environmental monitoring, response selection, and skeletomotor body orientation. Here, we tested the hypothesis that distinct resting state functional connectivity could be identified between (1) the anterior insula and pACC/aMCC; and (2) the entire insula (anterior, middle, and posterior insula) and the pMCC. Functional connectivity was assessed from resting state fMRI scans in 19 healthy volunteers using seed regions of interest in the anterior, middle, and posterior insula. Highly correlated, low-frequency oscillations (< 0.05 Hz) were identified between specific insula and cingulate subdivisions. The anterior insula was shown to be functionally connected with the pACC/aMCC and the pMCC, while the mid/posterior insula was only connected with the pMCC. These data provide evidence for a resting state anterior insula-pACC/aMCC cingulate system that may integrate interoceptive information with emotional salience to form a subjective representation of the body; and another system that includes the entire insula and MCC, likely involved in environmental monitoring, response selection, and skeletomotor body orientation.

Hierarchical functional modularity in the resting-state human brain

Functional magnetic resonance imaging (fMRI) studies have shown that anatomically distinct brain regions are functionally connected during the resting state. Basic topological properties in the brain functional connectivity (BFC) map have highlighted the BFC's small-world topology. Modularity, a more advanced topological property, has been hypothesized to be evolutionary advantageous, contributing to adaptive aspects of anatomical and functional brain connectivity. However, current definitions of modularity for complex networks focus on nonoverlapping clusters, and are seriously limited by disregarding inclusive relationships. Therefore, BFC's modularity has been mainly qualitatively investigated. Here, we introduce a new definition of modularity, based on a recently improved clustering measurement, which overcomes limitations of previous definitions, and apply it to the study of BFC in resting state fMRI of 53 healthy subjects. Results show hierarchical functional modularity in the brain.

BOLD correlates of EEG alpha phase-locking and the fMRI default mode network

Phase locking or synchronization of brain areas is a key concept of information processing in the brain. Synchronous oscillations have been observed and investigated extensively in EEG during the past decades. EEG oscillations occur over a wide frequency range. In EEG, a prominent type of oscillations is alpha-band activity, present typically when a subject is awake, but at rest with closed eyes. The spectral power of alpha rhythms has recently been investigated in simultaneous EEG/fMRI recordings, establishing a wide-range cortico-thalamic network. However, spectral power and synchronization are different measures and little is known about the correlations between BOLD effects and EEG synchronization. Interestingly, the fMRI BOLD signal also displays synchronous oscillations across different brain regions. These oscillations delineate so-called resting state networks (RSNs) that resemble the correlation patterns of simultaneous EEG/fMRI recordings. However, the nature of these BOLD oscillations and their relations to EEG activity is still poorly understood. One hypothesis is that the subunits constituting a specific RSN may be coordinated by different EEG rhythms. In this study we report on evidence for this hypothesis. The BOLD correlates of global EEG synchronization (GFS) in the alpha frequency band are located in brain areas involved in specific RSNs, e.g. the 'default mode network'. Furthermore, our results confirm the hypothesis that specific RSNs are organized by long-range synchronization at least in the alpha frequency band. Finally, we could localize specific areas where the GFS BOLD correlates and the associated RSN overlap. Thus, we claim that not only the spectral dynamics of EEG are important, but also their spatio-temporal organization.

Spatial and temporal reproducibility-based ranking of the independent components of BOLD fMRI data

Independent component analysis (ICA) decomposes fMRI data into spatially independent maps and their corresponding time courses. However, distinguishing the neurobiologically and biophysically reasonable components from those representing noise and artifacts is not trivial. We present a simple method for the ranking of independent components, by assessing the resemblance between components estimated from all the data, and components estimated from only the odd- (or even-) numbered time points. We show that the meaningful independent components of fMRI data resemble independent components estimated from downsampled data, and thus tend to be highly ranked by the method.

Altered resting state in diabetic neuropathic pain

Background

The spontaneous component of neuropathic pain (NP) has not been explored sufficiently with neuroimaging techniques, given the difficulty to coax out the brain components that sustain background ongoing pain. Here, we address for the first time the correlates of this component in an fMRI study of a group of eight patients suffering from diabetic neuropathic pain and eight healthy control subjects. Specifically, we studied the functional connectivity that is associated with spontaneous neuropathic pain with spatial independent component analysis (sICA).

Principal Findings

Functional connectivity analyses revealed a cortical network consisting of two anti-correlated patterns: one includes the left fusiform gyrus, the left lingual gyrus, the left inferior temporal gyrus, the right inferior occipital gyrus, the dorsal anterior cingulate cortex bilaterally, the pre and postcentral gyrus bilaterally, in which its activity is correlated negatively with pain and positively with the controls; the other includes the left precuneus, dorsolateral prefrontal, frontopolar cortex (both bilaterally), right superior frontal gyrus, left inferior frontal gyrus, thalami, both insulae, inferior parietal lobuli, right mammillary body, and a small area in the left brainstem, in which its activity is correlated positively with pain and negatively with the controls. Furthermore, a power spectra analyses revealed group differences in the frequency bands wherein the sICA signal was decomposed: patients' spectra are shifted towards higher frequencies.

Conclusion

In conclusion, we have characterized here for the first time a functional network of brain areas that mark the spontaneous component of NP. Pain is the result of aberrant default mode functional connectivity.

An ICA-based method for the identification of optimal FMRI features and components using combined group-discriminative techniques

Extraction of relevant features from multitask functional MRI (fMRI) data in order to identify potential biomarkers for disease, is an attractive goal. In this paper, we introduce a novel feature-based framework, which is sensitive and accurate in detecting group differences (e.g. controls vs. patients) by proposing three key ideas. First, we integrate two goal-directed techniques: coefficient-constrained independent component analysis (CC-ICA) and principal component analysis with reference (PCA-R), both of which improve sensitivity to group differences. Secondly, an automated artifact-removal method is developed for selecting components of interest derived from CC-ICA, with an average accuracy of 91%. Finally, we propose a strategy for optimal feature/component selection, aiming to identify optimal group-discriminative brain networks as well as the tasks within which these circuits are engaged. The group-discriminating performance is evaluated on 15 fMRI feature combinations (5 single features and 10 joint features) collected from 28 healthy control subjects and 25 schizophrenia patients. Results show that a feature from a sensorimotor task and a joint feature from a Sternberg working memory (probe) task and an auditory oddball (target) task are the top two feature combinations distinguishing groups. We identified three optimal features that best separate patients from controls, including brain networks consisting of temporal lobe, default mode and occipital lobe circuits, which when grouped together provide improved capability in classifying group membership. The proposed framework provides a general approach for selecting optimal brain networks which may serve as potential biomarkers of several brain diseases and thus has wide applicability in the neuroimaging research community.

Large-scale brain networks account for sustained and transient activity during target detection

Target detection paradigms have been widely applied in the study of human cognitive functions, particularly those associated with arousal, attention, stimulus processing and memory. In EEG recordings, the detection of task-relevant stimuli elicits the P300 component, a transient response with latency around 300 ms. The P300 response has been shown to be affected by the amount of mental effort and learning, as well as habituation. Furthermore, trial-by-trial variability of the P300 component has been associated with inter-stimulus interval, target-to-target interval or target probability; however, understanding the mechanisms underlying this variability is still an open question. In order to investigate whether it could be related to the distinct cortical networks in which coherent intrinsic activity is organized, and to understand the contribution of those networks to target detection processes, we carried out a simultaneous EEG-fMRI study, collecting data from 13 healthy subjects during a visual oddball task. We identified five large-scale networks, that largely overlap with the dorsal attention, the ventral attention, the core, the visual and the sensory-motor networks. Since the P300 component has been consistently associated with target detection, we concentrated on the first two brain networks, the time-course of which showed a modulation with the P300 response as detected in simultaneous EEG recordings. A trial-by-trial EEG-fMRI correlation approach revealed that they are involved in target detection with different functional roles: the ventral attention network, dedicated to revealing salient stimuli, was transiently activated by the occurrence of targets; the dorsal attention network, usually engaged during voluntary orienting, reflected sustained activity, possibly related to search for targets.

Noncanonical spike-related BOLD responses in focal epilepsy

Till now, most studies of the Blood Oxygen Level‐Dependent (BOLD) response to interictal epileptic discharges (IED) have assumed that its time course matches closely to that of brief physiological stimuli, commonly called the canonical event‐related haemodynamic response function (canonical HRF). Analyses based on that assumption have produced significant response patterns that are generally concordant with prior electroclinical data. In this work, we used a more flexible model of the event‐related response, a Fourier basis set, to investigate the presence of other responses in relation to individual IED in 30 experiments in patients with focal epilepsy. We found significant responses that had a noncanonical time course in 37% of cases, compared with 40% for the conventional, canonical HRF‐based approach. In two cases, the Fourier analysis suggested activations where the conventional model did not. The noncanonical activations were almost always remote from the presumed generator of epileptiform activity. In the majority of cases with noncanonical responses, the noncanonical responses in single‐voxel clusters were suggestive of artifacts. We did not find evidence for IED‐related noncanonical HRFs arising from areas of pathology, suggesting that the BOLD response to IED is primarily canonical. Noncanonical responses may represent a number of phenomena, including artefacts and propagated epileptiform activity. Hum Brain Mapp 2008. © 2007 Wiley‐Liss, Inc.

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.

Fetal cortical activation to sound at 33 weeks of gestation: a functional MRI study

Hearing already functions before birth, but little is known about the neural basis of fetal life experiences. Recent imaging studies have validated the use of functional magnetic resonance imaging (fMRI) in pregnant women at 38-weeks of gestation. The aim of the present study was to examine fetal brain activation to sound, using fMRI at the beginning of the third trimester of pregnancy. 6 pregnant women between 28- and 34-weeks of gestation were scanned using a magnetic strength of 1.5 T, with an auditory stimulus applied to their abdomen. 3 fetuses with a gestational age of 33 weeks, showed significant activation to sound in the left temporal lobe, measured using a new data-driven approach (Independent Component Analysis for fMRI time series). Only 2 of these fetuses showed left temporal activation, when the standard voxel-wise analysis method was used (p=0.007; p=0.001). Moreover, motion parameters added as predictors of the General Linear Model confirmed that motion cannot account for the signal variance in the fetal temporal cortex (p=0.01). Comparison between the statistical maps obtained from MRI scans of the fetuses with those obtained from adults, made it possible to confirm our hypothesis, that there is brain activation in the primary auditory cortex in response to sound. Measurement of the fetal hemodynamic response revealed an average fMRI signal change of +3.5%. This study shows that it is possible to use fMRI to detect early fetal brain function, but also confirms that sound processing occurs beyond the reflexive sub-cortical level, at the beginning of the third trimester of pregnancy.

Independent component analysis of interictal fMRI in focal epilepsy: Comparison with general linear model-based EEG-correlated fMRI

The general linear model (GLM) has been used to analyze simultaneous EEG-fMRI to reveal BOLD changes linked to interictal epileptic discharges (IED) identified on scalp EEG. This approach is ineffective when IED are not evident in the EEG. Data-driven fMRI analysis techniques that do not require an EEG derived model may offer a solution in these circumstances. We compared the findings of independent components analysis (ICA) and EEG-based GLM analyses of fMRI data from eight patients with focal epilepsy. Spatial ICA was used to extract independent components (IC) which were automatically classified as either BOLD-related, motion artefacts, EPI-susceptibility artefacts, large blood vessels, noise at high spatial or temporal frequency. The classifier reduced the number of candidate IC by 78%, with an average of 16 BOLD-related IC. Concordance between the ICA and GLM-derived results was assessed based on spatio-temporal criteria. In each patient, one of the IC satisfied the criteria to correspond to IED-based GLM result. The remaining IC were consistent with BOLD patterns of spontaneous brain activity and may include epileptic activity that was not evident on the scalp EEG. In conclusion, ICA of fMRI is capable of revealing areas of epileptic activity in patients with focal epilepsy and may be useful for the analysis of EEG-fMRI data in which abnormalities are not apparent on scalp EEG.

Automatic independent component labeling for artifact removal in fMRI

Blood oxygenation level dependent (BOLD) signals in functional magnetic resonance imaging (fMRI) are often small compared to the level of noise in the data. The sources of noise are numerous including different kinds of motion artifacts and physiological noise with complex patterns. This complicates the statistical analysis of the fMRI data. In this study, we propose an automatic method to reduce fMRI artifacts based on independent component analysis (ICA). We trained a supervised classifier to distinguish between independent components relating to a potentially task-related signal and independent components clearly relating to structured noise. After the components had been classified as either signal or noise, a denoised fMR time-series was reconstructed based only on the independent components classified as potentially task-related. The classifier was a novel global (fixed structure) decision tree trained in a Neyman-Pearson (NP) framework, which allowed the shape of the decision regions to be controlled effectively. Additionally, the conservativeness of the classifier could be tuned by modifying the NP threshold. The classifier was tested against the component classifications by an expert with the data from a category learning task. The test set as well as the expert were different from the data used for classifier training and the expert labeling the training set. The misclassification rate was between 0.2 and 0.3 for both the event-related and blocked designs and it was consistent among variety of different NP thresholds. The effects of denoising on the group-level statistical analyses were as expected: The denoising generally decreased Z-scores in the white matter, where extreme Z-values can be expected to reflect artifacts. A similar but weaker decrease in Z-scores was observed in the gray matter on average. These two observations suggest that denoising was likely to reduce artifacts from gray matter and could be useful to improve the detection of activations. We conclude that automatic ICA-based denoising offers a potentially useful approach to improve the quality of fMRI data and consequently increase the accuracy of the statistical analysis of these data.

Dissecting cognitive stages with time-resolved fMRI data: a comparison of fuzzy clustering and independent component analysis

In combination with cognitive tasks entailing sequences of sensory and cognitive processes, event-related acquisition schemes allow using functional MRI to examine not only the topography but also the temporal sequence of cortical activation across brain regions (time-resolved fMRI). In this study, we compared two data-driven methods--fuzzy clustering method (FCM) and independent component analysis (ICA)--in the context of time-resolved fMRI data collected during the performance of a newly devised visual imagery task. We analyzed a multisubject fMRI data set using both methods and compared their results in terms of within- and between-subject consistency and spatial and temporal correspondence of obtained maps and time courses. Both FCM and spatial ICA allowed discriminating the contribution of distinct networks of brain regions to the main cognitive stages of the task (auditory perception, mental imagery and behavioural response), with good agreement across methods. Whereas ICA worked optimally on the original time series, averaging with respect to the task onset (and thus introducing some a priori information on the stimulation protocol) was found to be indispensable in the case of FCM. On averaged time series, FCM led to a richer decomposition of the spatio-temporal patterns of activation and allowed a finer separation of the neurocognitive processes subserving the mental imagery task. This study confirms the efficacy of the two examined methods in the data-driven estimation of hemodynamic responses in time-resolved fMRI studies and provides empirical guidelines to their use.

Modelling large motion events in fMRI studies of patients with epilepsy

EEG-correlated fMRI can provide localisation information on the generators of epileptiform discharges in patients with focal epilepsy. To increase the technique's clinical potential, it is important to consider ways of optimising the yield of each experiment while minimizing the risk of false-positive activation. Head motion can lead to severe image degradation and result in false-positive activation and is usually worse in patients than in healthy subjects. We performed general linear model fMRI data analysis on simultaneous EEG-fMRI data acquired in 34 cases with focal epilepsy. Signal changes associated with large inter-scan motion events (head jerks) were modelled using modified design matrices that include 'scan nulling' regressors. We evaluated the efficacy of this approach by mapping the proportion of the brain for which F-tests across the additional regressors were significant. In 95% of cases, there was a significant effect of motion in 50% of the brain or greater; for the scan nulling effect, the proportion was 36%; this effect was predominantly in the neocortex. We conclude that careful consideration of the motion-related effects in fMRI studies of patients with epilepsy is essential and that the proposed approach can be effective.