An evaluation of methods for detecting brain activations from functional neuroimages

Effects of early midlife ovarian removal on sleep: Polysomnography-measured cortical arousal, homeostatic drive, and spindle characteristics

Bilateral salpingo-oophorectomy (BSO; removal of ovaries and fallopian tubes) prior to age 48 is associated with elevated risk for both Alzheimer's disease (AD) and sleep disorders such as insomnia and sleep apnea. In early midlife, individuals with BSO show reduced hippocampal volume, function, and hippocampal-dependent verbal episodic memory performance associated with changes in sleep. It is unknown whether BSO affects fine-grained sleep measurements (sleep microarchitecture) and how these changes might relate to hippocampal-dependent memory. We recruited thirty-six early midlife participants with BSO. Seventeen of these participants were taking 17β-estradiol therapy (BSO+ET) and 19 had never taken ET (BSO). Twenty age-matched control participants with intact ovaries (AMC) were also included. Overnight at-home polysomnography recordings were collected, along with subjective sleep quality and hot flash frequency. Multivariate Partial Least Squares (PLS) analysis was used to assess how sleep varied between groups. Compared to AMC, BSO without ET was associated with significantly decreased time spent in non-rapid eye movement (NREM) stage 2 sleep as well as increased NREM stage 2 and 3 beta power, NREM stage 2 delta power, and spindle power and maximum amplitude. Increased spindle maximum amplitude was negatively correlated with verbal episodic memory performance. Decreased sleep latency, increased sleep efficiency, and increased time spent in rapid eye movement sleep were observed for BSO+ET. Findings suggest there is an association between ovarian hormone loss and sleep microarchitecture, which may contribute to poorer cognitive outcomes and be ameliorated by ET.

Women's Brain Health: Midlife Ovarian Removal Affects Associative Memory

Women with early bilateral salpingo-oophorectomy (BSO; removal of ovaries and fallopian tubes) have greater Alzheimer's disease (AD) risk than women in spontaneous/natural menopause (SM), but early biomarkers of this risk are not well-characterized. Considering associative memory deficits may presage preclinical AD, we wondered if one of the earliest changes might be in associative memory and whether younger women with BSO had changes similar to those observed in SM. Women with BSO (with and without 17β-estradiol replacement therapy (ERT)), their age-matched premenopausal controls (AMC), and older women in SM completed a functional magnetic resonance imaging face-name associative memory task shown to predict early AD. Brain activation during encoding was compared between groups: AMC (n=25), BSO no ERT (BSO; n=15), BSO+ERT (n=16), and SM without hormone therapy (n=16). Region-of-interest analyses revealed AMC did not contribute to functional group differences. BSO+ERT had higher hippocampal activation than BSO and SM. This hippocampal activation correlated positively with urinary metabolite levels of 17β-estradiol. Multivariate partial least squares analyses showed BSO+ERT had a different network-level activation pattern than BSO and SM. Thus, despite being approximately 10 years younger, women with BSO without ERT had similar brain function to those with SM, suggesting early 17β-estradiol loss may lead to an altered functional brain phenotype which could influence late-life AD risk, making face-name encoding a potential biomarker for midlife women with increased AD risk. Despite similarities in activation, BSO and SM groups showed opposite within-hippocampus connectivity, suggesting menopause type is an important consideration when assessing brain function.

Associated functional network development and language abilities in children

During childhood, the brain is gradually converging to the efficient functional architecture observed in adults. How the brain's functional architecture evolves with age, particularly in young children, is however, not well understood. We examined the functional connectivity of the core language regions, in association with cortical growth and language abilities, in 175 young children in the age range of 4 to 9 years. We analyzed the brain's developmental changes using resting-state functional and T1-weighted structural magnetic resonance imaging data. The results showed increased functional connectivity strength with age between the pars triangularis of the left inferior frontal gyrus and left temporoparietal regions (cohen's d = 0.54, CI: 0.24 - 0.84), associated with children's language abilities. Stronger functional connectivity between bilateral prefrontal and temporoparietal regions was associated with better language abilities regardless of age. In addition, the stronger functional connectivity between the left inferior frontal and temporoparietal regions was associated with larger surface area and thinner cortical thickness in these regions, which in turn was associated with superior language abilities. Thus, using functional and structural brain indices, coupled with behavioral measures, we elucidate the association of functional language network development, language ability, and cortical growth, thereby adding to our understanding of the neural basis of language acquisition in young children.

Influence of sample size and analytic approach on stability and interpretation of brain-behavior correlations in task-related fMRI data

Limited statistical power due to small sample sizes is a problem in fMRI research. Most of the work to date has examined the impact of sample size on task‐related activation, with less attention paid to the influence of sample size on brain‐behavior correlations, especially in actual experimental fMRI data. We addressed this issue using two large data sets (a working memory task, N = 171, and a relational processing task, N = 865) and both univariate and multivariate approaches to voxel‐wise correlations. We created subsamples of different sizes and calculated correlations between task‐related activity at each voxel and task performance. Across both data sets the magnitude of the brain‐behavior correlations decreased and similarity across spatial maps increased with larger sample sizes. The multivariate technique identified more extensive correlated areas and more similarity across spatial maps, suggesting that a multivariate approach would provide a consistent advantage over univariate approaches in the stability of brain‐behavior correlations. In addition, the multivariate analyses showed that a sample size of roughly 80 or more participants would be needed for stable estimates of correlation magnitude in these data sets. Importantly, a number of additional factors would likely influence the choice of sample size for assessing such correlations in any given experiment, including the cognitive task of interest and the amount of data collected per participant. Our results provide novel experimental evidence in two independent data sets that the sample size commonly used in fMRI studies of 20–30 participants is very unlikely to be sufficient for obtaining reproducible brain‐behavior correlations, regardless of analytic approach.

Functional Connectivity within and beyond the Face Network Is Related to Reduced Discrimination of Degraded Faces in Young and Older Adults

Degrading face stimuli reduces face discrimination in both young and older adults, but the brain correlates of this decline in performance are not fully understood. We used functional magnetic resonance imaging to examine the effects of degraded face stimuli on face and nonface brain networks and tested whether these changes would predict the linear declines seen in performance. We found decreased activity in the face network (FN) and a decrease in the similarity of functional connectivity (FC) in the FN across conditions as degradation increased but no effect of age. FC in whole-brain networks also changed with increasing degradation, including increasing FC between the visual network and cognitive control networks. Older adults showed reduced modulation of this whole-brain FC pattern. The strongest predictors of within-participant decline in accuracy were changes in whole-brain network FC and FC similarity of the FN. There was no influence of age on these brain-behavior relations. These results suggest that a systems-level approach beyond the FN is required to understand the brain correlates of performance decline when faces are obscured with noise. In addition, the association between brain and behavior changes was maintained into older age, despite the dampened FC response to face degradation seen in older adults.

Changes in patterns of neural activity underlie a time-dependent transformation of memory in rats and humans

The dynamic process of memory consolidation involves a reorganization of brain regions that support a memory trace over time, but exactly how the network reorganizes as the memory changes remains unclear. We present novel converging evidence from studies of animals (rats) and humans for the time-dependent reorganization and transformation of different types of memory as measured both by behavior and brain activation. We find that context-specific memories in rats, and naturalistic episodic memories in humans, lose precision over time and activity in the hippocampus decreases. If, however, the retrieved memories retain contextual or perceptual detail, the hippocampus is engaged similarly at recent and remote timepoints. As the interval between the timepoint increases, the medial prefrontal cortex is engaged increasingly during memory retrieval, regardless of the context or the amount of retrieved detail. Moreover, these hippocampal-frontal shifts are accompanied by corresponding changes in a network of cortical structures mediating perceptually-detailed as well as less precise, schematic memories. These findings provide cross-species evidence for the crucial interplay between hippocampus and neocortex that reflects changes in memory representation over time and underlies systems consolidation.

5-HTTLPR differentially predicts brain network responses to emotional faces

The effects of the 5-HTTLPR polymorphism on neural responses to emotionally salient faces have been studied extensively, focusing on amygdala reactivity and amygdala-prefrontal interactions. Despite compelling evidence that emotional face paradigms engage a distributed network of brain regions involved in emotion, cognitive and visual processing, less is known about 5-HTTLPR effects on broader network responses. To address this, we evaluated 5-HTTLPR differences in the whole-brain response to an emotional faces paradigm including neutral, angry and fearful faces using functional magnetic resonance imaging in 76 healthy adults. We observed robust increased response to emotional faces in the amygdala, hippocampus, caudate, fusiform gyrus, superior temporal sulcus and lateral prefrontal and occipito-parietal cortices. We observed dissociation between 5-HTTLPR groups such that LA LA individuals had increased response to only angry faces, relative to neutral ones, but S' carriers had increased activity for both angry and fearful faces relative to neutral. Additionally, the response to angry faces was significantly greater in LA LA individuals compared to S' carriers and the response to fearful faces was significantly greater in S' carriers compared to LA LA individuals. These findings provide novel evidence for emotion-specific 5-HTTLPR effects on the response of a distributed set of brain regions including areas responsive to emotionally salient stimuli and critical components of the face-processing network. These findings provide additional insight into neurobiological mechanisms through which 5-HTTLPR genotype may affect personality and related risk for neuropsychiatric illness.

Comparing within-subject classification and regularization methods in fMRI for large and small sample sizes

In recent years, a variety of multivariate classifier models have been applied to fMRI, with different modeling assumptions. When classifying high-dimensional fMRI data, we must also regularize to improve model stability, and the interactions between classifier and regularization techniques are still being investigated. Classifiers are usually compared on large, multisubject fMRI datasets. However, it is unclear how classifier/regularizer models perform for within-subject analyses, as a function of signal strength and sample size. We compare four standard classifiers: Linear and Quadratic Discriminants, Logistic Regression and Support Vector Machines. Classification was performed on data in the linear kernel (covariance) feature space, and classifiers are tuned with four commonly-used regularizers: Principal Component and Independent Component Analysis, and penalization of kernel features using L₁ and L₂ norms. We evaluated prediction accuracy (P) and spatial reproducibility (R) of all classifier/regularizer combinations on single-subject analyses, over a range of three different block task contrasts and sample sizes for a BOLD fMRI experiment. We show that the classifier model has a small impact on signal detection, compared to the choice of regularizer. PCA maximizes reproducibility and global SNR, whereas Lp -norms tend to maximize prediction. ICA produces low reproducibility, and prediction accuracy is classifier-dependent. However, trade-offs in (P,R) depend partly on the optimization criterion, and PCA-based models are able to explore the widest range of (P,R) values. These trends are consistent across task contrasts and data sizes (training samples range from 6 to 96 scans). In addition, the trends in classifier performance are consistent for ROI-based classifier analyses.

Pattern classification of fMRI data: applications for analysis of spatially distributed cortical networks

The field of fMRI data analysis is rapidly growing in sophistication, particularly in the domain of multivariate pattern classification. However, the interaction between the properties of the analytical model and the parameters of the BOLD signal (e.g. signal magnitude, temporal variance and functional connectivity) is still an open problem. We addressed this problem by evaluating a set of pattern classification algorithms on simulated and experimental block-design fMRI data. The set of classifiers consisted of linear and quadratic discriminants, linear support vector machine, and linear and nonlinear Gaussian naive Bayes classifiers. For linear discriminant, we used two methods of regularization: principal component analysis, and ridge regularization. The classifiers were used (1) to classify the volumes according to the behavioral task that was performed by the subject, and (2) to construct spatial maps that indicated the relative contribution of each voxel to classification. Our evaluation metrics were: (1) accuracy of out-of-sample classification and (2) reproducibility of spatial maps. In simulated data sets, we performed an additional evaluation of spatial maps with ROC analysis. We varied the magnitude, temporal variance and connectivity of simulated fMRI signal and identified the optimal classifier for each simulated environment. Overall, the best performers were linear and quadratic discriminants (operating on principal components of the data matrix) and, in some rare situations, a nonlinear Gaussian naïve Bayes classifier. The results from the simulated data were supported by within-subject analysis of experimental fMRI data, collected in a study of aging. This is the first study that systematically characterizes interactions between analysis model and signal parameters (such as magnitude, variance and correlation) on the performance of pattern classifiers for fMRI.

Age differences in the intrinsic functional connectivity of default network subsystems

Recent work suggests that the default mode network (DMN) includes two core regions, the ventromedial prefrontal cortex and posterior cingulate cortex (PCC), and several unique subsystems that are functionally distinct. These include a medial temporal lobe (MTL) subsystem, active during remembering and future projection, and a dorsomedial prefrontal cortex (dmPFC) subsystem, active during self-reference. The PCC has been further subdivided into ventral (vPCC) and dorsal (dPCC) regions that are more strongly connected with the DMN and cognitive control networks, respectively. The goal of this study was to examine age differences in resting state functional connectivity within these subsystems. After applying a rigorous procedure to reduce the effects of head motion, we used a multivariate technique to identify both common and unique patterns of functional connectivity in the MTL vs. the dmPFC, and in vPCC vs. dPCC. All four areas had robust functional connectivity with other DMN regions, and each also showed distinct connectivity patterns in both age groups. Young and older adults had equivalent functional connectivity in the MTL subsystem. Older adults showed weaker connectivity in the vPCC and dmPFC subsystems, particularly with other DMN areas, but stronger connectivity than younger adults in the dPCC subsystem, which included areas involved in cognitive control. Our data provide evidence for distinct subsystems involving DMN nodes, which are maintained with age. Nevertheless, there are age differences in the strength of functional connectivity within these subsystems, supporting prior evidence that DMN connectivity is particularly vulnerable to age, whereas connectivity involving cognitive control regions is relatively maintained. These results suggest an age difference in the integrated activity among brain networks that can have implications for cognition in older adults.

Age differences in the frontoparietal cognitive control network: implications for distractibility

Current evidence suggests that older adults have reduced suppression of, and greater implicit memory for, distracting stimuli, due to age-related declines in frontal-based control mechanisms. In this study, we used fMRI to examine age differences in the neural underpinnings of attentional control and their relationship to differences in distractibility and subsequent memory for distraction. Older and younger adults were shown a rapid stream of words or nonwords superimposed on objects and performed a 1-back task on either the letters or the objects, while ignoring the other modality. Older adults were more distracted than younger adults by the overlapping words during the 1-back task, and they subsequently showed more priming for these words on an implicit memory task. A multivariate analysis of the imaging data revealed a set of regions, including the rostral PFC and inferior parietal cortex, that younger adults activated to a greater extent than older adults during the ignore-words condition, and activity in this set of regions was negatively correlated with priming for the distracting words. Functional connectivity analyses using right and left rostral PFC seeds revealed a network of putative control regions, including bilateral parietal cortex, connected to the frontal seeds at rest. Older adults showed reduced functional connectivity within this frontoparietal network, suggesting that their greater distractibility may be due to decreased activity and coherence within a cognitive control network that normally acts to reduce interference from distraction.

Familiarity modulates the functional relationship between theory of mind and autobiographical memory

Qualitative and quantitative reviews of the neuroimaging literature show that overlapping brain regions support theory of mind (ToM) and autobiographical memory (AM). This overlap has been taken to suggest that individuals draw on past personal experiences to infer others' mental states, but work with amnesic people shows that ToM does not always depend on AM. One variable that may determine the extent to which one relies on AM when inferring another's thoughts and feelings during ToM is whether that individual is personally known. To test this possibility, participants were scanned with fMRI as they remembered past experiences in response to personal photos ('AM' condition) and imagined others' experiences in response to photos of personally familiar ('pToM' condition) and unfamiliar ('ToM' condition) others. Spatiotemporal Partial Least Squares was used to identify the spatial and temporal characteristics of neural activation patterns associated with AM, pToM, and ToM. We found that the brain regions supporting pToM more closely resembled those supporting AM relative to ToM involving unfamiliar others, with the greatest degree of overlap within midline regions. A complementary finding was the observation of striking differences between pToM and ToM such that midline regions associated with AM predominated during pToM, whereas more lateral regions associated with social semantic memory predominated during ToM. Overall, this study demonstrates that ToM involves a dynamic interplay between AM and social semantic memory that is biased towards AM when a personally familiar other is the subject of the mental state inference.

Optimizing preprocessing and analysis pipelines for single-subject fMRI: 2. Interactions with ICA, PCA, task contrast and inter-subject heterogeneity

A variety of preprocessing techniques are available to correct subject-dependant artifacts in fMRI, caused by head motion and physiological noise. Although it has been established that the chosen preprocessing steps (or "pipeline") may significantly affect fMRI results, it is not well understood how preprocessing choices interact with other parts of the fMRI experimental design. In this study, we examine how two experimental factors interact with preprocessing: between-subject heterogeneity, and strength of task contrast. Two levels of cognitive contrast were examined in an fMRI adaptation of the Trail-Making Test, with data from young, healthy adults. The importance of standard preprocessing with motion correction, physiological noise correction, motion parameter regression and temporal detrending were examined for the two task contrasts. We also tested subspace estimation using Principal Component Analysis (PCA), and Independent Component Analysis (ICA). Results were obtained for Penalized Discriminant Analysis, and model performance quantified with reproducibility (R) and prediction metrics (P). Simulation methods were also used to test for potential biases from individual-subject optimization. Our results demonstrate that (1) individual pipeline optimization is not significantly more biased than fixed preprocessing. In addition, (2) when applying a fixed pipeline across all subjects, the task contrast significantly affects pipeline performance; in particular, the effects of PCA and ICA models vary with contrast, and are not by themselves optimal preprocessing steps. Also, (3) selecting the optimal pipeline for each subject improves within-subject (P,R) and between-subject overlap, with the weaker cognitive contrast being more sensitive to pipeline optimization. These results demonstrate that sensitivity of fMRI results is influenced not only by preprocessing choices, but also by interactions with other experimental design factors. This paper outlines a quantitative procedure to denoise data that would otherwise be discarded due to artifact; this is particularly relevant for weak signal contrasts in single-subject, small-sample and clinical datasets.

Routes to the past: neural substrates of direct and generative autobiographical memory retrieval

Models of autobiographical memory propose two routes to retrieval depending on cue specificity. When available cues are specific and personally-relevant, a memory can be directly accessed. However, when available cues are generic, one must engage a generative retrieval process to produce more specific cues to successfully access a relevant memory. The current study sought to characterize the neural bases of these retrieval processes. During functional magnetic resonance imaging (fMRI), participants were shown personally-relevant cues to elicit direct retrieval, or generic cues (nouns) to elicit generative retrieval. We used spatiotemporal partial least squares to characterize the spatial and temporal characteristics of the networks associated with direct and generative retrieval. Both retrieval tasks engaged regions comprising the autobiographical retrieval network, including hippocampus, and medial prefrontal and parietal cortices. However, some key neural differences emerged. Generative retrieval differentially recruited lateral prefrontal and temporal regions early on during the retrieval process, likely supporting the strategic search operations and initial recovery of generic autobiographical information. However, many regions were activated more strongly during direct versus generative retrieval, even when we time-locked the analysis to the successful recovery of events in both conditions. This result suggests that there may be fundamental differences between memories that are accessed directly and those that are recovered via the iterative search and retrieval process that characterizes generative retrieval.

Identification of psychopathic individuals using pattern classification of MRI images

BACKGROUND

Psychopathy is a disorder of personality characterized by severe impairments of social conduct, emotional experience, and interpersonal behavior. Psychopaths consistently violate social norms and bring considerable financial, emotional, or physical harm to others and to society as a whole. Recent developments in analysis methods of magnetic resonance imaging (MRI), such as voxel-based-morphometry (VBM), have become major tools to understand the anatomical correlates of this disorder. Nevertheless, the identification of psychopathy by neuroimaging or other neurobiological tools (e.g., genetic testing) remains elusive.

METHODS/PRINCIPAL FINDINGS

The main aim of this study was to develop an approach to distinguish psychopaths from healthy controls, based on the integration between pattern recognition methods and gray matter quantification. We employed support vector machines (SVM) and maximum uncertainty linear discrimination analysis (MLDA), with a feature-selection algorithm. Imaging data from 15 healthy controls and 15 psychopathic individuals (7 women in each group) were analyzed with SPM2 and the optimized VBM preprocessing routines. Participants were scanned with a 1.5 Tesla MRI system. Both SVM and MLDA achieved an overall leave-one-out accuracy of 80%, but SVM mapping was sparser than using MLDA. The superior temporal sulcus/gyrus (bilaterally) was identified as a region containing the most relevant information to separate the two groups.

CONCLUSION/SIGNIFICANCE

These results indicate that gray matter quantitative measures contain robust information to predict high psychopathy scores in individual subjects. The methods employed herein might prove useful as an adjunct to the established clinical and neuropsychological measures in patient screening and diagnostic accuracy.

Age differences in FMRI adaptation for sound identity and location

We explored age differences in auditory perception by measuring fMRI adaptation of brain activity to repetitions of sound identity (what) and location (where), using meaningful environmental sounds. In one condition, both sound identity and location were repeated allowing us to assess non-specific adaptation. In other conditions, only one feature was repeated (identity or location) to assess domain-specific adaptation. Both young and older adults showed comparable non-specific adaptation (identity and location) in bilateral temporal lobes, medial parietal cortex, and subcortical regions. However, older adults showed reduced domain-specific adaptation to location repetitions in a distributed set of regions, including frontal and parietal areas, and to identity repetition in anterior temporal cortex. We also re-analyzed data from a previously published 1-back fMRI study, in which participants responded to infrequent repetition of the identity or location of meaningful sounds. This analysis revealed age differences in domain-specific adaptation in a set of brain regions that overlapped substantially with those identified in the adaptation experiment. This converging evidence of reductions in the degree of auditory fMRI adaptation in older adults suggests that the processing of specific auditory “what” and “where” information is altered with age, which may influence cognitive functions that depend on this processing.

Dimensionality estimation for optimal detection of functional networks in BOLD fMRI data

Estimation of the intrinsic dimensionality of fMRI data is an important part of data analysis that helps to separate the signal of interest from noise. We have studied multiple methods of dimensionality estimation proposed in the literature and used these estimates to select a subset of principal components that was subsequently processed by linear discriminant analysis (LDA). Using simulated multivariate Gaussian data, we show that the dimensionality that optimizes signal detection (in terms of the receiver operating characteristic (ROC) metric) goes through a transition from many dimensions to a single dimension as a function of the signal-to-noise ratio. This transition happens when the loci of activation are organized into a spatial network and the variance of the networked, task-related signals is high enough for the signal to be easily detected in the data. We show that reproducibility of activation maps is a metric that captures this switch in intrinsic dimensionality. Except for reproducibility, all of the methods of dimensionality estimation we considered failed to capture this transition: optimization of Bayesian evidence, minimum description length, supervised and unsupervised LDA prediction, and Stein's unbiased risk estimator. This failure results in sub-optimal ROC performance of LDA in the presence of a spatially distributed network, and may have caused LDA to underperform in many of the reported comparisons in the literature. Using real fMRI data sets, including multi-subject group and within-subject longitudinal analysis we demonstrate the existence of these dimensionality transitions in real data.

The default network and processing of personally relevant information: converging evidence from task-related modulations and functional connectivity

Despite a growing interest in the default network (DN), its composition and function are not fully known. Here we examined whether the DN, as a whole, is specifically active during a task involving judgments about the self, or whether this engagement extends to judgments about a close other. We also aimed to provide converging evidence of DN involvement from across-task functional connectivity, and resting-state functional connectivity analyses, to provide a more comprehensive delineation of this network. Using functional MRI we measured brain activity in young adults during tasks and rest, and utilized a multivariate method to assess task-related changes as well as functional connectivity. An overlapping set of regions showed increased activity for judgments about the self, and about a close other, and strong functional connectivity with the posterior cingulate, a critical node of the DN. These areas included ventromedial prefrontal cortex, posterior parietal cortex, and medial temporal regions, all thought to be part of the DN. Several additional regions, such as the left inferior frontal gyrus and bilateral caudate, also showed the same pattern of activity and connectivity. These results provide evidence that the default network, as an integrated whole, supports internally oriented cognition involving information that is personally relevant, but not limited specifically to the self. They also suggest that the DN may be somewhat more extensive than currently thought.

An evaluation of spatial thresholding techniques in fMRI analysis

Many fMRI experiments have a common objective of identifying active voxels in a neuroimaging dataset. This is done in single subject experiments, for example, by performing individual voxel-wise tests of the null hypothesis that the observed time course is not significantly related to an assigned reference function. A voxel activation map is then constructed by applying a thresholding rule to the resulting statistics (e.g., t-statistics). Typically the task-related activation is expected to occur in clusters of voxels rather than in isolated single voxels. A variety of spatial thresholding techniques have been proposed to reflect this belief, including smoothing the raw t-statistics, cluster size inference, and spatial mixture modeling. We study two aspects of these spatial thresholding procedures applied to single subject fMRI analysis through simulation. First, we examine the performance of these procedures in terms of sensitivity to detect voxel activation, using receiver operating characteristic curves. Second, we consider the accuracy of these spatial thresholding procedures in estimation of the size of the activation region, both in terms of bias and variance. The findings indicate that smoothing has the highest sensitivity to modest magnitude signals, but tend to overestimate the size of the activation region. Spatial mixture models estimate the size of a spatially distributed activation region well, but may be less sensitive to modest magnitude signals, indicating that additional research into more sensitive spatial mixture models is needed. Finally, the methods are illustrated with a real bilateral finger-tapping fMRI experiment.

Evaluation and optimization of fMRI single-subject processing pipelines with NPAIRS and second-level CVA

In functional magnetic resonance imaging (fMRI) analysis, although the univariate general linear model (GLM) is currently the dominant approach to brain activation detection, there is growing interest in multivariate approaches such as principal component analysis, canonical variate analysis (CVA), independent component analysis and cluster analysis, which have the potential to reveal neural networks and functional connectivity in the brain. To understand the effect of processing options on performance of multivariate model-based fMRI processing pipelines with real fMRI data, we investigated the impact of commonly used fMRI preprocessing steps and optimized the associated multivariate CVA-based, single-subject processing pipelines with the NPAIRS (nonparametric prediction, activation, influence and reproducibility resampling) performance metrics [prediction accuracy and statistical parametric image (SPI) reproducibility] on the Fiswidgets platform. We also compared the single-subject SPIs of univariate GLM with multivariate CVA-based processing pipelines from SPM, FSL.FEAT, NPAIRS.GLM and NPAIRS.CVA software packages (or modules) using a novel second-level CVA. We found that for the block-design data, (a) slice timing correction and global intensity normalization have little consistent impact on the fMRI processing pipeline, but spatial smoothing, temporal detrending or high-pass filtering, and motion correction significantly improved pipeline performance across all subjects; (b) the combined optimization of spatial smoothing, temporal detrending and CVA model parameters on average improved between-subject reproducibility; and (c) the most important pipeline choices include univariate or multivariate statistical models and spatial smoothing. This study suggests that considering options other than simply using GLM with a fixed spatial filter may be of critical importance in determining activation patterns in BOLD fMRI studies.

Evaluation and comparison of GLM- and CVA-based fMRI processing pipelines with Java-based fMRI processing pipeline evaluation system

Activation patterns identified by fMRI processing pipelines or fMRI software packages are usually determined by the preprocessing options, parameters, and statistical models used. Previous studies that evaluated options of GLM (general linear model)--based fMRI processing pipelines are mainly based on simulated data with receiver operating characteristics (ROC) analysis, but evaluation of such fMRI processing pipelines on real fMRI data is rare. To understand the effect of processing options on performance of GLM-based fMRI processing pipelines with real fMRI data, we investigated the impact of commonly-used fMRI preprocessing steps; optimized the associated GLM-based single-subject processing pipelines; and quantitatively compared univariate GLM (in FSL.FEAT and NPAIRS.GLM) and multivariate CVA (canonical variates analysis) (in NPAIRS.CVA)-based analytic models in single-subject analysis with a recently developed fMRI processing pipeline evaluation system based on prediction accuracy (classification accuracy) and reproducibility performance metrics. For block-design data, we found that with GLM analysis (1) slice timing correction and global intensity normalization have little consistent impact on fMRI processing pipelines, spatial smoothing and high-pass filtering or temporal detrending significantly increases pipeline performance and thus are essential for robust fMRI statistical analysis; (2) combined optimization of spatial smoothing and temporal detrending improves pipeline performance; and (3) in general, the prediction performance of multivariate CVA is higher than that of the univariate GLM, while univariate GLM is more reproducible than multivariate CVA. Because of the different bias-variance trade-offs of univariate and multivariate models, it may be necessary to consider a consensus approach to obtain more accurate activation patterns in fMRI data.

Group analysis and the subject factor in functional magnetic resonance imaging: analysis of fifty right-handed healthy subjects in a semantic language task

Before considering a given fMRI paradigm as a valid clinical tool, one should first assess the reliability of functional responses across subjects by establishing a normative database and defining a reference activation map that identifies major brain regions involved in the task at hand. However, the definition of such a reference map can be hindered by inter-individual functional variability. In this study, we analysed functional data obtained from 50 healthy subjects during a semantic language task to assess the influence of the number of subjects on the reference map and to characterise inter-individual functional variability. We first compared different group analysis approaches and showed that the extent of the activated network depends not only on the choice of the analysis approach but also on the statistical threshold used and the number of subjects included. This analysis suggested that, while the RFX analysis is suitable to detect confidently true positive activations, the other group approaches are useful for exploratory investigations in small samples. The application of quantitative measures at the voxel and regional levels suggested that while approximately 15-20 subjects were sufficient to reveal reliable and robust left hemisphere activations, >30 subjects were necessary for revealing more variable and weak right hemisphere ones. Finally, to visualise inter-individual variability, we combined two similarity indices that assess the percentages of true positive and false negative voxels in individual activation patterns relative to the group map. We suggest that these measures can be used for the estimation of the degree of 'normality' of functional responses in brain-damaged patients, where this question is often raised, and recommend the use of different quantifications to appreciate accurately the inter-individual functional variability that can be incorporated in group maps.

Domain-specific data sharing in neuroscience: what do we have to learn from each other?

Molecular biology and genomics have made notable strides in the sharing of primary data and resources. In other domains of neuroscience research, however, there has been resistance to adopting formalized strategies for data exchange, archiving, and availability. In this article, we discuss how neuroscience domains might follow the lead of molecular biology on what has been successful and what has failed in active data sharing. This considers not only the technical challenges but also the sociological concerns in making it possible. Though, not a pain-free process, with increased data availability, scientists from multiple fields can enjoy greater opportunity for novel discoveries about the brain in health and disease.

Bayesian kernel methods for analysis of functional neuroimages

We propose an approach to analyzing functional neuroimages in which 1) regions of neuronal activation are described by a superposition of spatial kernel functions, the parameters of which are estimated from the data and 2) the presence of activation is detected by means of a generalized likelihood ratio test (GLRT). Kernel methods have become a staple of modern machine learning. Herein, we show that these techniques show promise for neuroimage analysis. In an on-off design, we model the spatial activation pattern as a sum of an unknown number of kernel functions of unknown location, amplitude, and/or size. We employ two Bayesian methods of estimating the kernel functions. The first is a maximum a posteriori (MAP) estimation method based on a Reversible-Jump Markov-chain Monte-Carlo (RJMCMC) algorithm that searches for both the appropriate model complexity and parameter values. The second is a relevance vector machine (RVM), a kernel machine that is known to be effective in controlling model complexity (and thus discouraging overfitting). In each method, after estimating the activation pattern, we test for local activation using a GLRT. We evaluate the results using receiver operating characteristic (ROC) curves for simulated neuroimaging data and example results for real fMRI data. We find that, while RVM and RJMCMC both produce good results, RVM requires far less computation time, and thus appears to be the more promising of the two approaches.

Bayesian decoding of brain images

This paper introduces a multivariate Bayesian (MVB) scheme to decode or recognise brain states from neuroimages. It resolves the ill-posed many-to-one mapping, from voxel values or data features to a target variable, using a parametric empirical or hierarchical Bayesian model. This model is inverted using standard variational techniques, in this case expectation maximisation, to furnish the model evidence and the conditional density of the model's parameters. This allows one to compare different models or hypotheses about the mapping from functional or structural anatomy to perceptual and behavioural consequences (or their deficits). We frame this approach in terms of decoding measured brain states to predict or classify outcomes using the rhetoric established in pattern classification of neuroimaging data. However, the aim of MVB is not to predict (because the outcomes are known) but to enable inference on different models of structure-function mappings; such as distributed and sparse representations. This allows one to optimise the model itself and produce predictions that outperform standard pattern classification approaches, like support vector machines. Technically, the model inversion and inference uses the same empirical Bayesian procedures developed for ill-posed inverse problems (e.g., source reconstruction in EEG). However, the MVB scheme used here extends this approach to include a greedy search for sparse solutions. It reduces the problem to the same form used in Gaussian process modelling, which affords a generic and efficient scheme for model optimisation and evaluating model evidence. We illustrate MVB using simulated and real data, with a special focus on model comparison; where models can differ in the form of the mapping (i.e., neuronal representation) within one region, or in the (combination of) regions per se.

Implications of the Rician distribution for fMRI generalized likelihood ratio tests

In functional magnetic resonance imaging (fMRI), the general linear model test (GLMT) is widely used for brain activation detection. However, the GLMT relies on the assumption that the noise corrupting the data is Gaussian distributed. Because the majority of fMRI studies employ magnitude image reconstructions, which are Rician distributed, this assumption is invalid and has significant consequences in case the signal-to-noise ratio (SNR) is low. In this study, we show that the GLMT should not be used at low SNR. Furthermore, we propose a generalized likelihood ratio test for magnitude MR data that has the same performance compared to the GLMT for high SNR, but performs significantly better than the GLMT for low SNR.

Functional connectivity during Stroop task performance

Using covariance-based multivariate analysis, we examined patterns of functional connectivity in rCBF on a practice-extended version of the Stroop color-word paradigm. Color-word congruent and incongruent conditions were presented in six AB trials to healthy subjects during 12 H2(15)O PET scans. Analyses identified two reproducible canonical eigenimages (CE) from the PET data, which were converted to a standard Z score scale after cross-validation resampling and correction for random subject effects. The first CE corresponded to practice-dependent changes in covarying rCBF that occurred over early task repetitions and correlated with improved behavioral performance. This included many regions previously implicated by PET and fMRI studies of this task, which we suggest may represent two "parallel" networks: (i) a cingulo-frontal system that was initially engaged in selecting and mapping a task-relevant response (color naming) when the attentional demands of the task were greatest; and (ii) a ventral visual processing stream whose concurrent decrease in activity represented the task-irrelevant inhibition of word reading. The second CE corresponded to a consistent paradigmatic effect of Stroop interference on covarying rCBF. Coactivations were located in dorsal and ventral prefrontal regions as well as frontopolar cortex. This pattern supports existing evidence that prefrontal regions are involved in maintaining attentional control over conflicting response systems. Taken together, these findings may be more in line with theoretical models that emphasize a role for practice in the emergence of Stroop phenomena. These findings may also provide some additional insight into the nature of anterior cingulate- and prefrontal cortical contributions to implementing cognitive control in the brain.

Optimizing the fMRI data-processing pipeline using prediction and reproducibility performance metrics: I. A preliminary group analysis

We argue that published results demonstrate that new insights into human brain function may be obscured by poor and/or limited choices in the data-processing pipeline, and review the work on performance metrics for optimizing pipelines: prediction, reproducibility, and related empirical Receiver Operating Characteristic (ROC) curve metrics. Using the NPAIRS split-half resampling framework for estimating prediction/reproducibility metrics (Strother et al., 2002), we illustrate its use by testing the relative importance of selected pipeline components (interpolation, in-plane spatial smoothing, temporal detrending, and between-subject alignment) in a group analysis of BOLD-fMRI scans from 16 subjects performing a block-design, parametric-static-force task. Large-scale brain networks were detected using a multivariate linear discriminant analysis (canonical variates analysis, CVA) that was tuned to fit the data. We found that tuning the CVA model and spatial smoothing were the most important processing parameters. Temporal detrending was essential to remove low-frequency, reproducing time trends; the number of cosine basis functions for detrending was optimized by assuming that separate epochs of baseline scans have constant, equal means, and this assumption was assessed with prediction metrics. Higher-order polynomial warps compared to affine alignment had only a minor impact on the performance metrics. We found that both prediction and reproducibility metrics were required for optimizing the pipeline and give somewhat different results. Moreover, the parameter settings of components in the pipeline interact so that the current practice of reporting the optimization of components tested in relative isolation is unlikely to lead to fully optimized processing pipelines.

Partial least squares analysis of neuroimaging data: applications and advances

Partial least squares (PLS) analysis has been used to characterize distributed signals measured by neuroimaging methods like positron emission tomography (PET), functional magnetic resonance imaging (fMRI), event-related potentials (ERP) and magnetoencephalography (MEG). In the application to PET, it has been used to extract activity patterns differentiating cognitive tasks, patterns relating distributed activity to behavior, and to describe large-scale interregional interactions or functional connections. This paper reviews the more recent extension of PLS to the analysis of spatiotemporal patterns present in fMRI, ERP, and MEG data. We present a basic mathematical description of PLS and discuss the statistical assessment using permutation testing and bootstrap resampling. These two resampling methods provide complementary information of the statistical strength of the extracted activity patterns (permutation test) and the reliability of regional contributions to the patterns (bootstrap resampling). Simulated ERP data are used to guide the basic interpretation of spatiotemporal PLS results, and examples from empirical ERP and fMRI data sets are used for further illustration. We conclude with a discussion of some caveats in the use of PLS, including nonlinearities, nonorthogonality, and interpretation difficulties. We further discuss its role as an important tool in a pluralistic analytic approach to neuroimaging.

A post-processing/region of interest (ROI) method for discriminating patterns of activity in statistical maps of fMRI data

To combine functional neuroimaging studies across subjects, anatomical and functional data are typically either transformed to a common space or averaged across regions of interest (ROIs). However, if there are (1) anatomical variations within the subject pool (as in clinical or aging populations), (2) non-Gaussian distributions of task-related activity within a typical ROI or, (3) more ROIs than subjects, neither spatial transformation of the data to a common space nor averaging across all subjects' ROIs is suitable for standard discriminant analysis. To solve these problems, we describe a post-processing method that uses voxel-based statistics representing task-related activity (pooled within ROIs) to establish combinations of ROIs that maximally differentiate tasks across all subjects. The method involves randomized resampling from multiple ROIs within each subject, multivariate linear discriminant analysis across all subjects and validation with bootstrapping techniques. When applied to experimental data from healthy subjects performing two motor tasks, the method detected some brain regions, including the supplementary motor area (SMA), that participated in a distributed network differentially active between tasks. However there was not a significant difference in SMA activity when this region was examined in isolation. We suggest this method is a practical means to combine voxel-based statistics within anatomically defined ROIs across subjects.

Evaluating subject specific preprocessing choices in multisubject fMRI data sets using data-driven performance metrics

This study investigated the possible benefit of subject specific optimization of preprocessing strategies in functional magnetic resonance imaging (fMRI) experiments. The optimization was performed using the data-driven performance metrics developed recently [Neuroimage 15 (2002), 747]. We applied numerous preprocessing strategies and a multivariate statistical analysis to each of the 20 subjects in our two example fMRI data sets. We found that the optimal preprocessing strategy varied, in general, from subject to subject. For example, in one data set, optimum smoothing levels varied from 16 mm (4 subjects), 10 mm (5 subjects), to no smoothing at all (1 subject). This strongly suggests that group-specific preprocessing schemes may not give optimum results. For both studies, optimizing the preprocessing for each subject resulted in an increased number of suprathresholded voxels in within-subject analyses. Furthermore, we demonstrated that we were able to aggregate the optimized data with a random effects group analysis, resulting in improved sensitivity in one study and the detection of interesting, previously undetected results in the other.