as first author...
 

Finer parcellation reveals intricate correlational structure of steady-state fMRI signals

Joao V. Dornas,  Jochen Braun

British Neuroscience Association, Birmingham 2017

Anatomical and functional parcellations of the human brain are widely used, for example, ‘automated anatomical labelling’ into 90 cortical and subcortical regions (‘AAL90’, Tzourio-Mazoyer et al., 2002), spatially constrained clustering of functional correlations (‘C400’, Cradock et al., 2013), or multi-modal parcellation from the Human Connectome Project (‘HPC360’, Glasser et al., 2016).  However, only a modest amount of correlational information can be retrieved at these comparatively coarse resolutions (and only about half of the pairwise functional correlations between resting-state signals are consistently significant). We propose a finer parcellation (‘MD758’) which increases the bivariate mutual information retrieved by functional correlations approximately 100-fold (and the multivariate mutual information approximately 10-fold).  Subdividing each AAL area separately on the basis of local functional correlations, we define 758 highly inter-correlated and spatially largely contiguous volumes (‘functional clusters’).  At this finer resolution, a large majority of pairwise functional correlations is consistently significant (86% with p<.01, cv<1.0). Moreover, fibre tracking reveals consistent anatomical connectivity between these ‘functional clusters', echoing the global pattern of functional correlations.  In fact, even local patterns of cluster-to-cluster correlations often mirror cluster-to-cluster connectivity in detail and with high significance (p<.00001).  The global and local correspondence of functional correlations and anatomical connectivity at the level of ‘functional clusters’ further validates the proposed parcellation. We conclude that a finer parcellation, which combines both anatomical and functional criteria, unlocks a treasure trove of intricate correlational structure in resting-state BOLD signals.

Computational Neuroscience Society, Antwerp 2017

Birmingham, 2017

Antwerp, 2017

How attentional states change brain activity has been studied extensively for decades.  Attention shifts triggered by salient stimuli activate cortical regions known as the VAN, whereas volitional shifts activate regions of the DAN.  Sustained attention changes average activity at certain cortical sites, as well as temporal correlations between activity at these sites. The challenge for any comprehensive analysis of the functional correlations accompanying sustained attention is the overwhelming amount of information: O(105) voxels imply  O(1010) pairwise correlations.  To reduce this complexity, previous studies have resorted to spatial averaging (to O(102) ROIs) or have restricted themselves to correlation densities (FCD).  

We have developed a novel approach and have reduced, in an information-preserving manner,  the high-resolution correlation matrix of O(105) voxels to an intermediate resolution matrix of  758 ‘functional clusters’.  

We subdivided the anatomically-defined AAL regions, into 758 ‘functional clusters’ averaging 212 voxels each, on the basis of functional connectivity in the resting state.   This parcellation, which we call ‘MD758’,  combines voxels with similar local correlation profiles such as to remove correlational redundancy without losing correlational information.  

The ‘attention contrast’ shows a preponderance of decorrelation within, respectively,  occipital, frontal superior, and temporal medial cortex, as well as within cuneus and precuneus, in addition to strong recorrelation within parietal cortex.  In addition, there is massive decorrelation between occipital, parietal and precuneus, on the one hand, and frontal-medial, frontal-superior, and temporal-medial cortex, on the other hand.  Finally, there is massive recorrelation between  occipital cortex, on the one hand, and parietal, mid-orbitotfrontal cortex and precuneus.

The ‘stimulus contrast’ shows decorrelation within and between occipital and lingual cortex, as well as recorrelation between occipital cortex, on the one hand, and cingulum, insula, and frontal inferior and orbital cortex.  The most dramatic effect, however, is a massive recorrelation between thalamus, caudate, and putamen, on the one hand, and calcarine, insula, occipital, cuneus, and fusiform cortex.

European Conference in Visual Perception, Berlin 2017

DETAILED CHANGES IN GLOBAL FUNCTIONAL CONNECTIVITY DURING ATTENTIONAL TRACKING

Joao V. Dornas,  Jochen Braun

 

Berlin, 2017

 
 
 
Bernstein Conference, Berlin 2016
Oxford Autumn Conference, Oxford 2016

Global Changes in Connectivity due to Attention

Joao V. Dornas, Katharina Glomb, Jochen Braun

Local Functional Connectivity Density showed a widespread Increase in Negative correlations and a high amount of voxels with significant Decrease in Positive correlations (results showed on FORUM FENS, Copenhagen, 2016). In this new study, we analyzed the Global correlation using a more coarse spatial resolution but more fine grained than usual Anatomical parcellations. For each AAL parcel, we defined clusters of around 200 voxels which are highly correlated to each other in Resting State condition.  We then calculated the Functional Connectivity and Granger Causality among all clusters using the mean time series of all voxels inside each cluster for each run in each condition (Resting State, Passive Viewing, Attentive Tracking). Doing a contrast between Passive Viewing and Resting State, and Attentive Tracking and Passive Viewing, for both analysis, we established when there were Increases or Decreases for Functional and Granger Connectivity due to Stimulus and Attention, individually, in a Global level. 1. There is much less Increase in FC for Attention and much more Decrease while for the Stimulus contrast we have a widespread Increase and a slightly smaller Decrease. 2. If you look carefully, you will see that the seeds are not random. On Attention Summary plot with Increase in FC there are several, the majority, of Dorsal Attention Network seeds, which are related to Top-Down Attention. 3. If you compare the Decrease in Attention with the Increase in Stimulus you will find out that there are several seeds that are present on both plots. So, those seeds in clusters where there is an Increase due to Stimulus are almost the same for those that have a Decrease due to Attention. And on the other two plots the seeds are different. 4. The Increase due to Stimulus is present on all AAL parcels, while the Increase due to Attention is more concentrated in fewer parcels. 5. The Decrease due to Attention is very strong in almost all clusters, the strength of Decrease in FC is very big for everybody, while on the Decrease due to Stimulus the strength is less prominent. 6. The Increase due to Attention has a few hubs, based on FC strength. Particularly Insula-R. On the Increase due to Stimulus, there is a prominence in FC strength on subcortical areas. 

 

 

Oxford, 2016

Berlin, 2016

Copenhagen, 2016

FORUM FENS, Copenhagen 2016

Density of local correlations varies extensively with attentive tracking

Joao V. Dornas, Katharina Glomb, Jochen Braun

 Attentive tracking of multiple moving targets is known to engage widely distributed networks of cortical regions [1], leading to significant functional correlations between time-varying average MRI signals in these regions. We studied more fine-grained functional correlations between voxel pairs in eight observers under three conditions (resting, passive viewing, attentive   

tracking). For each voxel, we established the density of significant correlations (positive or negative) with others in the same region (FCD, [2]). Attentive tracking induced far more extensive and consistent changes in FCD than did passive viewing the same stimulus. Within the ventral frontoparietal network (VAN), FCD tended to decrease. Within the dorsal      

frontoparietal network (DAN), positive FCD tended to decrease and negative FCD to increase. Outside the attention networks, significant changes were evident in all four cortical lobes as well as subcortically, with increases predominating in occipital and temporal cortices, but decreases   

subcortically. Adjacent to DAN, frontal superior medial cortex exhibited increased FCD and angular gyrus decreased positive, but increased negative FCD. In summary, we found different changes for positive and negative correlations as well as for areas within and adjacent to frontoparietal networks. We conclude that attentive tracking extensively changes the fine-grained cooperative dynamics of brain activity, as indexed by time-varying MRI signals from individual voxels, confirming and extending traditional studies of coarse-grained cooperative dynamics.

 

[1] Corbetta & Shulman (2002). Control of goal-directed and stimulus-driven attention in the brain. Nature Reviews Neuroscience, 3(3), 201–215.

 

[2] Tomasi & Volkow (2010) Functional connectivity density mapping. PNAS 107(21): 9885-9890. 

 
  • Twitter Social Icon
  • LinkedIn Social Icon
  • Instagram Social Icon