CocchiYZSHGM20173LCocchiZYangAZaleskyJStelzerLJHearneLLGolloJBMattingley2017-06-006383069–3080Functional magnetic resonance imaging (fMRI) studies have shown that neural activity fluctuates spontaneously between different states of global synchronization over a timescale of several seconds. Such fluctuations generate transient states of high and low correlation across distributed cortical areas. It has been hypothesized that such fluctuations in global efficiency might alter patterns of activity in local neuronal populations elicited by changes in incoming sensory stimuli. To test this prediction, we used a linear decoder to discriminate patterns of neural activity elicited by face and motion stimuli presented periodically while participants underwent time-resolved fMRI. As predicted, decoding was reliably higher during states of high global efficiency than during states of low efficiency, and this difference was evident across both visual and nonvisual cortical regions. The results indicate that slow fluctuations in global network efficiency are associated with variations in the pattern of activity across widespread cortical regions responsible for representing distinct categories of visual stimulus. More broadly, the findings highlight the importance of understanding the impact of global fluctuations in functional connectivity on specialized, stimulus driven neural processes.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/published-3069Neural decoding of visual stimuli varies with fluctuations in global network efficiency1501718821LohmannSMLBKGS20173GLohmannJStelzerKMuellerELacosseTBuschmannVJKumarWGroddKScheffler2017-03-00Reproducibility is generally regarded as a hallmark of scientific validity. It can be undermined by two very different factors, namely inflated false positive rates or inflated false negative rates. Here we investigate the role of the second factor, i.e. the degree to which true effects are not detected reliably. The availability of large public databases and also supercomputing allows us to tackle this problem quantitatively. Specifically, we estimated the reproducibility in task-based fMRI data over different samples randomly drawn from a large cohort of subjects obtained from the Human Connectome Project. We use the full cohort as a standard of reference to approximate true positive effects, and compute the fraction of those effects that was detected reliably using standard software packages at various smaller sample sizes. We found that with standard sample sizes this fraction was less than 25 percent. We conclude that inflated false negative rates are a major factor that undermine reproducibility. We introduce a new statistical inference algorithm based on a novel test statistic and show that it improves reproducibility without inflating false positive rates.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/submitted0Inflated False Negative Rates Undermine Reproducibility In Task-Based fMRI1501718821HoveSNTGMVTKM20153MJHoveJStelzerTNierhausSDThielCGundlachDSMarguliesKRAVan DijkRTurnerPEKellerBMerker2016-07-0072631163124Trance is an absorptive state of consciousness characterized by narrowed awareness of external surroundings and has long been used—for example, by shamans—to gain insight. Shamans across cultures often induce trance by listening to rhythmic drumming. Using functional magnetic resonance imaging (fMRI), we examined the brain-network configuration associated with trance. Experienced shamanic practitioners (n = 15) listened to rhythmic drumming, and either entered a trance state or remained in a nontrance state during 8-min scans. We analyzed changes in network connectivity. Trance was associated with higher eigenvector centrality (i.e., stronger hubs) in 3 regions: posterior cingulate cortex (PCC), dorsal anterior cingulate cortex (dACC), and left insula/operculum. Seed-based analysis revealed increased coactivation of the PCC (a default network hub involved in internally oriented cognitive states) with the dACC and insula (control-network regions involved in maintaining relevant neural streams). This coactivation suggests that an internally oriented neural stream was amplified by the modulatory control network. Additionally, during trance, seeds within the auditory pathway were less connected, possibly indicating perceptual decoupling and suppression of the repetitive auditory stimuli. In sum, trance involved coactive default and control networks, and decoupled sensory processing. This network reconfiguration may promote an extended internal train of thought wherein integration and insight can occur.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/published8Brain Network Reconfiguration and Perceptual Decoupling During an Absorptive State of Consciousness1501718821LohmannSZBMBS20153GLohmannJStelzerVZuberTBuschmannDMarguliesABartelsKScheffler2016-06-00611122The formation of transient networks in response to external stimuli or as a reflection of internal cognitive processes is a hallmark of human brain function. However, its identification in fMRI data of the human brain is notoriously difficult. Here we propose a new method of fMRI data analysis that tackles this problem by considering large-scale, task-related synchronisation networks. Networks consist of nodes and edges connecting them, where nodes correspond to voxels in fMRI data, and the weight of an edge is determined via task-related changes in dynamic synchronisation between their respective times series. Based on these definitions, we developed a new data analysis algorithm that identifies edges that show differing levels of synchrony between two distinct task conditions and that occur in dense packs with similar characteristics. Hence, we call this approach “Task-related Edge Density” (TED). TED proved to be a very strong marker for dynamic network formation that easily lends itself to statistical analysis using large scale statistical inference. A major advantage of TED compared to other methods is that it does not depend on any specific hemodynamic response model, and it also does not require a presegmentation of the data for dimensionality reduction as it can handle large networks consisting of tens of thousands of voxels. We applied TED to fMRI data of a fingertapping and an emotion processing task provided by the Human Connectome Project. TED revealed network-based involvement of a large number of brain areas that evaded detection using traditional GLM-based analysis. We show that our proposed method provides an entirely new window into the immense complexity of human brain function.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/published21Task-Related Edge Density (TED): A New Method for Revealing Dynamic Network Formation in fMRI Data of the Human Brain15017188211501715421JacobsenSFCLT20153J-HJacobsenJStelzerTHFritzGChételatRLa JoieRTurner2015-08-00813824382450Musical memory is considered to be partly independent from other memory systems. In Alzheimer’s disease and different types of dementia, musical memory is surprisingly robust, and likewise for brain lesions affecting other kinds of memory. However, the mechanisms and neural substrates of musical memory remain poorly understood. In a group of 32 normal young human subjects (16 male and 16 female, mean age of 28.0 ± 2.2 years), we performed a 7 T functional magnetic resonance imaging study of brain responses to music excerpts that were unknown, recently known (heard an hour before scanning), and long-known. We used multivariate pattern classification to identify brain regions that encode long-term musical memory. The results showed a crucial role for the caudal anterior cingulate and the ventral pre-supplementary motor area in the neural encoding of long-known as compared with recently known and unknown music. In the second part of the study, we analysed data of three essential Alzheimer’s disease biomarkers in a region of interest derived from our musical memory findings (caudal anterior cingulate cortex and ventral pre-supplementary motor area) in 20 patients with Alzheimer’s disease (10 male and 10 female, mean age of 68.9 ± 9.0 years) and 34 healthy control subjects (14 male and 20 female, mean age of 68.1 ± 7.2 years). Interestingly, the regions identified to encode musical memory corresponded to areas that showed substantially minimal cortical atrophy (as measured with magnetic resonance imaging), and minimal disruption of glucose-metabolism (as measured with 18F-fluorodeoxyglucose positron emission tomography), as compared to the rest of the brain. However, amyloid-β deposition (as measured with 18F-flobetapir positron emission tomography) within the currently observed regions of interest was not substantially less than in the rest of the brain, which suggests that the regions of interest were still in a very early stage of the expected course of biomarker development in these regions (amyloid accumulation → hypometabolism → cortical atrophy) and therefore relatively well preserved. Given the observed overlap of musical memory regions with areas that are relatively spared in Alzheimer’s disease, the current findings may thus explain the surprising preservation of musical memory in this neurodegenerative disease.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/published12Why musical memory can be preserved in advanced Alzheimer’s disease1501718821NierhausFPHKTLSMSV20153TNierhausNForschackSKPiperSHoltzeTKrauseBTaskinXLongJStelzerDSMarguliesJSteinbrinckAVillringer2015-04-00153559175925Most sensory input to our body is not consciously perceived. Nevertheless, it may reach the cortex and influence our behavior. In this study, we investigated noninvasive neural signatures of unconscious cortical stimulus processing to understand mechanisms, which (1) prevent low-intensity somatosensory stimuli from getting access to conscious experience and which (2) can explain the associated impediment of conscious perception for additional stimuli. Stimulation of digit 2 in humans far below the detection threshold elicited a cortical evoked potential (P1) at 60 ms, but no further somatosensory evoked potential components. No event-related desynchronization was detected; rather, there was a transient synchronization in the alpha frequency range. Using the same stimulation during fMRI, a reduced centrality of contralateral primary somatosensory cortex (SI) was found, which appeared to be mainly driven by reduced functional connectivity to frontoparietal areas. We conclude that after subthreshold stimulation the (excitatory) feedforward sweep of bottom-up processing terminates in SI preventing access to conscious experience. We speculate that this interruption is due to a predominance of inhibitory processing in SI. The increase in alpha activity and the disconnection of SI from frontoparietal areas are likely correlates of an elevated perception threshold and may thus serve as a gating mechanism for the access to conscious experience.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/published8Imperceptible Somatosensory Stimulation Alters Sensorimotor Background Rhythm and ConnectivityStelzerLMBT20143JStelzerGLohmannKMuellerTBuschmannRTurner2014-07-004628116Functional magnetic resonance imaging (fMRI) is the workhorse of imaging-based human cognitive neuroscience. The use of fMRI is ever-increasing; within the last 4 years more fMRI studies have been published than in the previous 17 years. This large body of research has mainly focused on the functional localization of condition- or stimulus-dependent changes in the blood-oxygenation-level dependent (BOLD) signal. In recent years, however, many aspects of the commonly practiced analysis frameworks and methodologies have been critically reassessed. Here we summarize these critiques, providing an overview of the major conceptual and practical deficiencies in widely used brain-mapping approaches, and exemplify some of these issues by the use of imaging data and simulations. In particular, we discuss the inherent pitfalls and shortcomings of methodologies for statistical parametric mapping. Our critique emphasizes recent reports of excessively high numbers of both false positive and false negative findings in fMRI brain mapping. We outline our view regarding the broader scientific implications of these methodological considerations and briefly discuss possible solutions.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/published15Deficient Approaches to Human Neuroimaging1501718821StelzerBLMTT20143JStelzerTBuschmannGLohmannDSMarguliesRTrampelRTurner2014-04-0066818Although ultra-high-field fMRI at field strengths of 7T or above provides substantial gains in BOLD contrast-to-noise ratio, when very high-resolution fMRI is required such gains are inevitably reduced. The improvement in sensitivity provided by multivariate analysis techniques, as compared with univariate methods, then becomes especially welcome. Information mapping approaches are commonly used, such as the searchlight technique, which take into account the spatially distributed patterns of activation in order to predict stimulus conditions. However, the popular searchlight decoding technique, in particular, has been found to be prone to spatial inaccuracies. For instance, the spatial extent of informative areas is generally exaggerated, and their spatial configuration is distorted. We propose the combination of a non-parametric and permutation-based statistical framework with linear classifiers. We term this new combined method Feature Weight Mapping (FWM). The main goal of the proposed method is to map the specific contribution of each voxel to the classification decision while including a correction for the multiple comparisons problem. Next, we compare this new method to the searchlight approach using a simulation and ultra-high-field 7T experimental data. We found that the searchlight method led to spatial inaccuracies that are especially noticeable in high-resolution fMRI data. In contrast, FWM was more spatially precise, revealing both informative anatomical structures as well as the direction by which voxels contribute to the classification. By maximizing the spatial accuracy of ultra-high-field fMRI results, global multivariate methods provide a substantial improvement for characterizing structure-function relationships.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/published7Prioritizing spatial accuracy in high-resolution fMRI data using multivariate feature weight mapping1501718821LohmannSNAT20133GLohmannJStelzerJNeumannNAyRTurner2013-06-0033223239Two aspects play a key role in recently developed strategies for functional magnetic resonance imaging (fMRI) data analysis: first, it is now recognized that the human brain is a complex adaptive system and exhibits the hallmarks of complexity such as emergence of patterns arising out of a multitude of interactions between its many constituents. Second, the field of fMRI has evolved into a data-intensive, big data endeavor with large databases and masses of data being shared around the world. At the same time, ultra-high field MRI scanners are now available producing data at previously unobtainable quality and quantity. Both aspects have led to shifts in the way in which we view fMRI data. Here, we review recent developments in fMRI data analysis methodology that resulted from these shifts in paradigm.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/published16"More Is Different" in Functional Magnetic Resonance Imaging: A Review of Recent Data Analysis TechniquesStelzerCT20133JStelzerYChenRTurner2013-01-006569–82An ever-increasing number of functional magnetic resonance imaging (fMRI) studies are now using information-based multi-voxel pattern analysis (MVPA) techniques to decode mental states. In doing so, they achieve a significantly greater sensitivity compared to when they use univariate frameworks. However, the new brain-decoding methods have also posed new challenges for analysis and statistical inference on the group level. We discuss why the usual procedure of performing t-tests on accuracy maps across subjects in order to produce a group statistic is inappropriate. We propose a solution to this problem for local MVPA approaches, which achieves higher sensitivity than other procedures. Our method uses random permutation tests on the single-subject level, and then combines the results on the group level with a bootstrap method. To preserve the spatial dependency induced by local MVPA methods, we generate a random permutation set and keep it fixed across all locations. This enables us to later apply a cluster size control for the multiple testing problem. More specifically, we explicitly compute the distribution of cluster sizes and use this to determine the p-values for each cluster. Using a volumetric searchlight decoding procedure, we demonstrate the validity and sensitivity of our approach using both simulated and real fMRI data sets. In comparison to the standard t-test procedure implemented in SPM8, our results showed a higher sensitivity. We discuss the theoretical applicability and the practical advantages of our approach, and outline its generalization to other local MVPA methods, such as surface decoding techniques.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/published-69Statistical inference and multiple testing correction in classification-based multi-voxel pattern analysis (MVPA): Random permutations and cluster size controlLohmannSZBES20147GLohmannJStelzerVZuberTBuschmannMErbKSchefflerTübingen, Germany2014-06-0014Traditionally fMRI data analysis aims at identifying brain areas in which the amplitude of the BOLD signal responds to experimental stimulations. However, since the brain acts as a network, we would expect differential effects on network topology. Therefore, the target of statistical inference should not only be individual voxels or brain areas but rather network connections. Here we introduce a new approach to correlation-based statistics in fMRI. At the heart of our approach is the concept of correlation bundles as a functional analogy to anatomical fibre bundles. Statistical tests are applied to these bundles using large-scale inference methods such as FDR. We call this approach correlation bundle statistics (CBS). In contrast to previous correlation-based approaches to fMRI statistics, CBS does not require a presegmentation or smoothing of the data so that anatomical specificity is preserved. The result of a CBS analysis is not a set of voxels or brain regions but rather a set of correlation bundles that are found to be significantly affected by some experimental manipulation.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/published3Correlation bundle statistics in fMRI data1501718821LohmannSMKGBS20177GLohmannJStelzerKMuellerVKumarWGroddTBuschmannKSchefflerVancouver, BC, Canada2017-06-29Introduction:
Statistical inference in fMRI data analysis remains a challenging problem. Current techniques were mostly developed for 3T data, but are often unsatisfactory in terms of spatial acuity and sensitivity when applied to ultrahigh field data (>=7T). Furthermore, a recent publication by Eklund et al.  has highlighted problems with inflated false positive rates which can be alleviated by using very stringent initial cluster-forming thresholds (p < 0.001), but possibly at the expense of inflated false negative rates. Here we propose a new method to address these problems. It is called "LISA" because it is inspired by hot spot analysis of geographical information systems where hot spots are identified using so-called Local Indicators of Spatial Association (LISA) . With LISA, every voxel receives a hot spot score which serves as a new test statistic and may be seen as a compromise between cluster-level and voxel-level inference.
If operated at the second (group) level, the algorithm LISA expects as input a set of contrast maps obtained from a first level GLM analysis. First a voxelwise t-test is applied yielding a map in which each voxel has a z-value uncorrected for multiple comparisons. We now apply a bilateral filter to this map which suppresses noise while preserving spatial acuity . The parameters of this filter were determined using simulated data and were kept constant for all experiments reported below. The filtered map highlights hot spots of activation which LISA aims to detect. Statistical inference is performed by controlling the false discovery rate (FDR). Note that the classical FDR algorithm  assumes that all data points are independent and under the null hypothesis z-values follow a standard Gaussian distribution. Both assumptions may be violated here. Therefore, we use a different FDR procedure which is based on a two-component model  in which we estimate the null distribution using random permutations of the contrast maps. The LISA algorithm can also be used at the first level (single subject analysis) in which case it expects as input a preprocessed fMRI data set and the experimental design information. To ensure exchangeability, the null distribution is obtained using random permutations of labels [6,7]. Otherwise, the algorithm works as described above.
We subjected LISA to a battery of tests.
Test 1: We analysed 127 data sets of the "Beijing" sample of , using the same experimental designs and preprocessing regimes as in  (6mm spatial smoothing). In each of the four designs (B1,B2,E1,E2) , we randomly drew 100 samples consisting of 40 data sets and obtained the following family-wise error rates: 3/100 (B1), 0/100 (B2), 0/100 (E1), 2/100 (E2).
Test 2: Simulated data. Comparison with FSL-TFCE , see fig.1.
Test 3: fMRI data of the "emotion task" of the Human Connectome Project (HCP) . We randomly selected 10 sets of 20 data sets each and compared the LISA results with results obtained by FSL-TFCE , see fig.2 (top).
Test 4: Single subject data acquired at a 9.4T human whole-body scanner (Siemens) using a custom-built 31-channel receive coil array. Gradient Echo EPI, 30 slices, Grappa 4, 6/8 partial fourier, PSF-based distortion correction, resolution 0.8mm isotropic, 30 slices, 405 volumes, TR/TE=1580/22ms, FOV 171mm, working memory task, 8+8 trials (2back/0back), fig.2 (bottom)
Lisa corrects for FDR, but under the Eklund test it produced even more conservative results than expected if corrected for the familywise error. LISA appears to be less conservative than FSL-TFCE, and shows a high spatial precision and sensitivity when applied to data acquired at an ultra-high field scanner. Applying an edge-preserving filter at a late stage in the analysis chain rather than during preprocessing.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/published0LISA: a new threshold-free and non-parametric statistical inference method for fMRI data1501718821LewisSSBESRP20177LLewisKSetsompopJStelzerJBausePEhsesKSchefflerBRosenJPolimeniVancouver, BC, Canada2017-06-26Introduction:
The temporal resolution of fMRI is ultimately limited by the slow dynamics of the hemodynamic response function (HRF) . Canonical HRF models predict that neural activity above ~0.3 Hz should not be detectable with fMRI . However, recent work has demonstrated that neural oscillations can be measured at up to 0.75 Hz at 7T  due to nonlinear responses to rapid neural activity [4-7]. High-frequency oscillations are challenging to detect, as signal amplitude is small and cardiac noise appears around ~1 Hz. To probe the upper limit of fast fMRI, we tested whether neurally driven BOLD signals at 1 Hz could be detected by taking advantage of improved signal strength at 9.4 T.
Two healthy volunteers provided informed consent and were scanned on a 9.4T scanner using a custom-built 31-channel receive coil array. Functional runs were single-shot gradient-echo blipped-CAIPI SMS-EPI with 15 oblique slices, 2 mm isotropic, targeting the calcarine sulcus (R=2 acceleration, MB=3, TR=227 ms, TE=24 ms, FA=30°). Stimuli consisted of a 12 Hz flickering radial checkerboard displayed for 4 minutes. The luminance contrast of the stimulus oscillated at either 0.2 Hz (localizer run) or 1 Hz (test runs). Data were slice-timing corrected, motion corrected, and high-pass filtered. Mean responses were computed by upsampling the BOLD timecourse and averaging every cycle of the oscillation. ICA was performed with FSL MELODIC. Statistical comparison used three control analyses: 1) a manually defined non-visual contiguous gray matter ROI; 2) a random draw of non-visual voxels with similar tSNR, resampled 1000 times; and 3) using jittered cycle times within the visual ROI, resampled 1000 times. Reported p-values are from resampling tests and reported amplitude is of the best fit sine wave.
The localizer identified 0.2 Hz oscillatory responses in visual cortex (Fig 1a). Response phase varied by hundreds of milliseconds across voxels (Fig. 1b), suggesting that selecting a subset of voxels with similar phases could reduce cancellation across the ROI. Voxels also exhibited ~1-1.3 Hz cardiac noise contamination (Fig. 1c). Given the ~3% magnitude of the 0.2 Hz oscillation, canonical models would predict a ~0.0003% response magnitude at 1 Hz (undetectable at current SNR levels) whereas extrapolating from previous results at 0.75 Hz would suggest a ~0.01% response. Analyzing the mean ROI signal during 1 Hz stimulation yielded a significant oscillation in one of the two subjects (S1 p=0.19; S2 p=0.03; Fig. 2a) when compared to a jittered control within the same ROI, but not significantly larger than other ROIs (p>0.05), suggesting that noise levels were too high to detect signals with the predicted magnitude. To reduce noise, we performed ICA with 0.2 Hz highpassing to constrain the components to be high-frequency, and observed that the first 15 components reflected cardiac contamination within slice groups (Fig. 2b). After excluding these components and selecting only voxels with a narrow range of phase responses in the localizer run, we observed significant 1 Hz oscillations in both subjects (Fig. 2c). These oscillations were significantly larger than in resampled control ROIs (S1 p=0.028; S2 p=0.015; Fig. 2d) and than jittered control analyses within the visual ROI (S1 p=0.010; S2 p=0.025; Fig. 2e) in each subject.
These results suggest it may be possible to measure 1 Hz neural oscillations within single subjects using fMRI at ultra-high fields. However, due to the extremely small signal magnitudes, additional subjects will need to be studied to confirm this finding. Our results highlight the need for analyses that can separate noise and signal at high frequencies, particularly when approaching the 1 Hz range due to strong cardiac contamination. We conclude that combining physiological noise correction, fast acquisition, single-voxel phase estimation, and ultra-high field strengths may enable detection of surprisingly rapid neural oscillations using fMRI.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/published0Identifying neural contributions to high frequency dynamics in the fMRI signal at 9.4 Tesla1501718821PolimeniZSBEWS20177JPolimeniNZaretskayaJStelzerJBausePEhsesLWaldKSchefflerHonolulu, HI, USA2017-04-27Several strategies have been proposed for maximizing neuronal specificity of fMRI by utilizing pulse sequences that are primarily sensitive to signal changes within microvasculature. Here we compare the microvascular sensitivity of high-resolution balanced SSFP and gradient-echo EPI at 9.4T using cortical depth analyses within human visual cortex. Because of the large draining vessels lying along the pial surface, the behavior of fMRI signals as a function of cortical depth can provide helpful insights into the vascular contributions. Our preliminary analyses suggest that, for the protocols used here, both balanced SSFP and EPI show similar cortical depth profiles of BOLD responses.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/published0Macrovascular contributions to high-resolution balanced SSFP- and GE-EPI-based fMRI at 9.4T evaluated using surface-based cortical depth analyses in human visual cortex1501718821StelzerEBSL20167JStelzerPEhsesJBauseKSchefflerGLohmannSan Diego, CA, USA2016-11-16The combination of ultra-high field fMRI with state-of-the art network approaches offers a unique window for studying human brain function at the mesoscopic level. In this study, we used a 9.4T MRI system to acquire functional data at submillimetre resolution, covering more than 500000 voxels in fronto-parietal areas. As experimental manipulation we tested a simple 2-back against an 0-back memory task, as in the human connectome project. We analysed the data with a network-based method which we specifically tailored for ultra-high-resolution data on the single subject level, named “task-induced edge density” or “TED”. Our method aims to detect task-dependent changes in synchronization across the entire brain. The algorithm operates on the voxel level and does not require any presegmentation or spatial smoothing of the data.
Our method reveals widespread changes in the network configuration across the two memory tasks. A large proportion of grey matter voxels changes its connectivity to the rest of the brain between the two tasks. Thus our findings suggest that vast parts of the cortex might subserve the underlying brain functions. Interestingly, a distributed subset of areas appears to change its connectivity to an especially large number of voxels, possibly indicating key areas or super-hubs within the network. We further discuss the fine structure of the connectivity patterns, such as the formation of subnetworks on smaller spatial scales and the relation to the underlying anatomical structure.
The present results distinctively favour a more integrative rather than segregative view of brain function, which appears to be wide-spread instead of sparse. However, our results also raise other issues of interpretability due to the sheer extent of the involved brain areas.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/published0Resolving large-scale networks in ultra-high field fMRI (9.4T) of the human brain1501718821LohmannSS20167GLohmannJStelzerKSchefflerGeneva, Switzerland2016-06-29Introduction:
The formation of transient networks in response to external stimuli or as a reflection of internal cognitive processes is a hallmark of human brain function. However, its identification in fMRI data of the human brain is notoriously difficult. While there exist many algorithms for resting state data, the same cannot be said for task-based fMRI. Here we propose a new method of fMRI data analysis that tackles this problem by considering large-scale, task-related synchronisation networks. Our new method identifies task-related changes in network configuration without requiring presegmentations so that the spatial resolution of the input data is preserved. Furthermore, it is free from any specific hemodynamic response model so that it is capable of detecting many different types of responses to task changes.
Networks consist of nodes and edges connecting them, where nodes correspond to voxels in fMRI data, and the weight of an edge is determined via task-related changes in dynamic synchronisation between their respective times series. Based on these definitions, we developed a new data analysis algorithm that is designed to identify time series of voxels in an fMRI image that collectively synchronize in response to a task. At the heart of our approach is the concept of spatially localized and task-induced edge density motivating us to call this algorithm "TED" (Task-induced Edge Density). In short, TED identifies edges in a brain network that differentially respond in unison to a task onset and that occur in dense packs of edges with similar characteristics. Figure 1 illustrates this concept: the TED measure counts the percentage of supra-threshold edges connecting local neighbourhoods. A detailed description of the algorithm can be found in Lohmann et al. (2015). We applied TED to task-based fMRI data provided by the Human Connectome Project focusing on the social recognition task, see Barch et al (2013). Minimally preprocessed data of 100 subjects were included. We contrasted two conditions called 'mental' and 'random'. Statistical inference was based on false discovery rates (FDR) using 1000 permutations to derive a null distribution.
TED identified several task-specific, large-scale patterns of task-related synchronization. Figure 2 shows the result as a hubness map, FDR corrected at p < 0.05. A voxel in the hubness map records the number of edges for which this voxel serves as an endpoint. Voxels in which many edges accumulate may be viewed as hubs in a task-specific network, and the number of edges meeting in a voxel is a measure of the voxel's hubness. Note that the left and right temporal poles appear as extremely strong hubs indicating that they may serve as integration areas, see e.g. Pascual et al. (2015). Figure 3 shows an alternative visualization. Here edges passing through a pre-defined ROI are shown.
We found TED to be a very strong marker for dynamic network formation that easily lends itself to statistical analysis using large scale statistical approaches such as the false discovery rate. A major advantage of TED compared to other network-based methods is that it does not require a presegmentation of the data for dimensionality reduction as it can handle large networks consisting of tens of thousands of voxels. Because its conceptual basis is task-induced synchronization it does not depend on a hemodynamic response model. We conclude that the new TED method provides us with an entirely new window into the immense complexity of human brain function.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/published0Task-induced edge density analysis applied to the HCP social recognition experiment1501718821LohmannSBZMBS20157GLohmannJStelzerTBuschmannVZuberDMarguliesABartelsKSchefflerChicago, IL, USA2015-10-21The formation of transient networks in response to external stimuli or as a reflection of internal cognitive processes is a hallmark of human brain function. However, in the past twenty years, task-based fMRI studies have primarily focused on signal amplitude changes or connectivity related to a few selected nodes. Shifting focus away from signal amplitudes or constraining connectivity patterns of a few selected nodes, we propose an alternative view on fMRI data analysis by considering large-scale, task-induced synchronization networks. Networks consist of nodes and edges connecting them, where nodes in our method correspond to voxels in fMRI data, and the weight of an edge between any two voxels is determined via task-induced changes in dynamic synchronization between their respective times series. Based on these definitions, we developed a new data analysis algorithm that is designed to identify time series of voxels in an fMRI image that collectively synchronize in response to a task. At the heart of our approach is the concept of spatially localized and task-induced edge density motivating us to call this algorithm "TED" (Task induced Edge Density). In short, TED identifies edges in a brain network that differentially respond in unison to a task onset and that occur in dense packs of edges with similar responses to tasks. We found TED to be a very strong marker for dynamic network formation that easily lends itself to statistical analysis using large scale statistical approaches such as the local false discovery rate (local fdr). A major advantage of TED compared to other network-based methods is that it does not require a presegmentation of the data for dimensionality reduction as it can handle large networks consisting of tens of thousands of voxels. Because its conceptual basis is task-induced synchronization it does not depend on a hemodynamic response model. We applied TED to task-based fMRI data provided by the Human Connectome Project focusing on the motor, social recognition and working memory tasks. In all cases, TED identified several task-specific, large-scale patterns of synchronization. We conclude that the new TED method provides us with an entirely new window into the immense complexity of human brain function.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/published0Task-induced edge density as a marker for dynamic network formation in fMRI15017188211501715421StelzerBT20157JStelzerCBauerAThielscherHonolulu, HI, USA2015-06-17Introduction:
Electrical fields passing through the human skull are affected by its low electrical conductivity and structural irregularity. This issue is relevant both for brain stimulation methods (transcranial direct current stimulation) and electroencephalography (EEG). However, usually such effects are not taken into account. Head models are often overly simplified and unrealistic, entailing a potentially severe loss of predictability to which brain areas are being stimulated in non-invasive brain stimulation (Windhoff et al., 2011) and an imprecise source localization in the case of EEG data (Dannhauer et al., 2011; Rullmann et al., 2009).
Our approach to overcome these limitations is the optimization of specific MR sequences, which allow for a clearer delineation of the human head and skull anatomy. We validate the MR images by acquiring high-resolution computed tomography (CT) data of the same subjects. On basis of the MR sequences, we reconstructed the skull using image processing and surface based methods.
We optimized several MR sequences to aid the reconstruction of the human head anatomy on a 3T Philips Achieva system. Particularly relevant were an ultra-short echo time (UTE) sequence (Robson et al., 2003) and a mDixon sequence (Ma, 2008). The mDixon sequence allows separating between fat and water content and prevents the occurrence of chemical shift artifacts. Among others, it is useful to delineate the cortical bone. The UTE images were mainly employed to distinguish between cortical bone and air-filled cavities. Additionally, we acquired a venogram sequence and also standard T1 and T2-weighted images for comparison. We acquired computed tomography images of the head region of the same subjects, using a Siemens Biograph scanner (115mAs, 80keV, 0.6mm slices).
We applied a N3 bias field removal to all MR images (Sled et al., 1998) and coregistered all MR (FSL flirt (Jenkinson et al., 2012)). Combining the T2 image with the venogramm allowed us to apply binary morphology methods (Dogdas et al., 2005) to obtain an inner outline of the skull, including the subdural space. Shrinking and smoothing the inner outline generated a starting volume for the surface growing algorithm, which was guaranteed to not intersect with the skull.
The (energy-based) surface growing algorithm used a constraint image, which we computed as the weighted addition of the mDixon in-phase and fat images, the venogram and air cavities derived from the UTE image. Only if a surface node resulted in a net decrease of its total energy, the surface was expanded at this node. We defined the (node-wise) energy as the sum of its potential energy (provided by the constraint image), its elastic energy (provided by an elasticity term motivated by magnetic spin-lattices) and lastly a repulsion term from other surfaces (in analogy to electrostatic repulsion). Furthermore we introduced node-wise convergence criterions.
We depict the results of the optimized MR sequences and the computed tomography scan of the same subject in Figure 1. The delineations of the compact and spongy bone are particularly well visible in the mDixon in-phase image. In Figure 2 we display the combination of all MR modalities for the potential energy term (the constraint image), furthermore we show the computed tomography scan. Lastly we present the results of the three-dimensional surface reconstruction algorithm of the inner skull in Figure 3.
We present and validate optimized MR sequences which can be used for the reconstruction of the human skull. Furthermore we present a surface-growing algorithm for reconstructing the surfaces of the cortical bone in the human skull.
Our work can be readily incorporated into existing surface-based head models, which find usage in both non-invasive brain stimulation simulations and source reconstruction of EEG data. Future work will focus on reducing the number of MR modalities that are needed to achieve accurate results for the skull segmentations.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/published0A generic approach for realistic head modelling for electrical field mapping and source localizationLohmannSZBMES20147GLohmannJStelzerVZuberTBuschmannKMuellerMErbKSchefflerHamburg, Germany2014-06-12Introduction:
The multiple comparison problem arises in the statistical analysis of fMRI data because independent statistical tests are performed at each voxel of an image. In order to reduce the number of tests, statistical inference is often performed at the cluster level where clusters are identified by thresholding the uncorrected map of z-values, and assuming that small clusters tend to be spurious while larger clusters are more reliable. The multiple comparison problem then still exists but is alleviated because the number of tests is reduced from many thousands to a few dozens. A common procedure is to control the 'false discovery rate' (BH-Fdr) (Benjamini and Hochberg, 1995) at the cluster level. Here we argue that BH-Fdr as implemented in major software systems (e.g. Chumbley et al 2009) is too conservative because it rests on assumptions which are unrealistic in this context. We propose a revised algorithm to solve this problem.
BH-Fdr assumes that the null distribution is continuous and uniform in [0,1]. However, as we found using simulations, the distribution of cluster sizes is not uniform with small clusters being much more frequent than large ones. Thus, the corresponding null distribution is neither uniform nor continuous (fig 1,2 and Stelzer et al. 2013). To respect the specific nature of this distribution, we propose to use an empirical null density based on simulations. We employ this new null density within the framework of 'local fdr' (Efron 2007). Local fdr assumes a two-class model with a mixture density f(x) = p0 f0(x) + p1 f1(x) where f0 f1 are the null and non-null densities, and p0,p1 their priors. BH-Fdr uses the same model with f0, f1 replaced by their cumulative distributions where f0 is assumed to be uniform. In the context of fMRI, local fdr was previously proposed by Schwartzman et al (2009) for statistical inference at the voxel level. Here, we derive the empirical null by recording the sizes of randomly generated clusters and define fdr(x) = f0(x) / f(x). We call this new algorithm ``clusterFDR''.
We analyzed an fMRI experiment featuring an auditory paradigm described in Mueller et al (2011). In this experiment, 20 subjects (7 females) were presented with pieces of music versus scrambled music. Scanning was done at a 3T MedSpec 30/100 scanner (Bruker, Ettlingen, Germany) using a standard EPI sequence. For details see Mueller et al (2011). Using standard GLM-based data analysis, it was found that real music showed a stronger activation in left and right auditory cortices than scrambled music. Here, we also computed the reverse contrast (scrambled > real music) and found two clusters when BH-Fdr correction with an initial cluster threshold of z > 2.33 was used (figure 3). Using SPM8, spatial smoothness was found to be 6.3 voxels. We then tested clusterFDR on these data. To derive an empirical null distribution, we generated 200 images simulating zmaps with the same spatial smoothness, and obtained a histogram of cluster sizes via thresholding at z=2.33 and a null density function of corresponding p-values (figures 1,2). With our new clusterFDR, we could replicate the first finding. But in addition, we found four more clusters in the reverse contrast that had previously been overlooked (figure 3).
As noted by Lieberman and Cunningham (2009), statistical procedures for analyzing fMRI data traditionally have been geared towards a rather strict exclusion of false positives. As a consequence, relevant aspects of the data may have been overlooked in the past (Gonzales et al 2012). Our new clusterFDR may help to remedy this problem. But let us note that statistical methodology can only be used to weed out truly random effects. It does not guard against false positives due to confounding effects.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/published0Improving statistical sensitivity for fMRI data by
clusterFDR1501718821EklundBSL20137AEklundMBjörnsdotterJStelzerSLaConteSalt Lake City, UT, USA2013-04-25The searchlight algorithm is a popular choice for locally-multivariate decoding of fMRI data. A substantial drawback of searchlight is the increase in computational complexity, compared to the univariate general linear model. This is especially true for large searchlight spheres, non-linear classifiers, cross validation schemes and statistical permutation testing. Here we therefore present a graphics processing unit (GPU) implementation of the searchlight algorithm, to enable fast locally-multivariate fMRI analysis. The GPU implementation is 21 times faster than a multithreaded Matlab implementation. This makes it possible to apply 10 000 permutations with leave-one-out cross-validation in about 19 minutes.nonotspecifiedhttp://www.kyb.tuebingen.mpg.de/published0Searchlight Goes GPU: Fast Multi-Voxel Pattern Analysis of fMRI DataWeberSSLJ201710MWeberESaucanJStelzerGLohmannJJostStelzer201610JStelzer