Q1. What is Brainnetome atlas? 

Brainnetome atlas is considered to be the cornerstone of basic neuroscience and clinical researches, but available atlases often lack finer grained parcellation results and do not provide the functional important connectivity information. Over the past thirty years, remarkable advances of multimodal neuroimaging techniques that are rapidly advancing our understanding of the form and function of the human brain. The introduction of the framework for identifying the brain subdivisions with in vivo connectivity architecture has opened the door to neuroanatomical studies at the macro-scale brain studies. One key element of Brainnetome project is focused on setting up and optimizing the framework for connectivity-based parcellation(CBP), and aim to produce a new human brain atlas, i.e. Brainnetome atlas, based on structural and connectivity features. Brainnetome atlas will be an in vivo, and not only with finer-grained brain subregions, but with anatomical and functional connection patterns.

The main features of Brainnetome atlas are: in vivo, more fine-grained parcels and with connections. 1) Fine-grained brain sub-regions: Brainnetome atlas based on connectivity architecture has not only confirmed some of the accepted structural differentiations in earlier cytoarchitectonic maps, also revealed numerous anatomical subdivisions that were missed previously. 2) Functional and structural connections: The diffusion MRI combined with tractography could allow reconstruction of the major fiber bundles, while functional connectivity analysis of resting state fMRI could also provide a noninvasive means to assess in vivo large-scale connectivity in the human brain. With such approaches we can obtain connectivity data and relate these data to the particular parcellation scheme. Therefore, along the brainnetome atlas, different types of connectivity information are presented.

The optimized integrating of connectivity information with other criteria is critical for acquiring accurate brain parcellation results, which could be served as appropriate solutions for the next-generation brain atlas. The next step of Brainnetome atlas will plan to achieve reliable and reproducible maps of the human brain, and then, to get robust mapping of individual reliable parcellations across different populations.

Q2. The framework of Brainnetome atlas construction

It consists of the following steps: 1) the seed region was extracted using the structural MRI; 2) with the multimodal MRI data including diffusion MRI and resting state fMRI data, the anatomical and functional connectivity information was acquired for each voxel in the seed region; 3) the connections were estimated between voxels in the brain region and all of the remaining voxels; 4) based on the native connectivity matrix, a cross-correlation matrix was calculated to quantify the similarity/dissimilarity between the connectivity profiles of the seed voxels; 5) the cross-correlation matrix was then processed using clustering algorisms and reordered, which could group data on the basis of its similarity, the more similar their connectivity profiles, the more likely the voxels will be grouped together; 6) the connectivity-defined clusters were acquired and mapped back onto the brain.

Q3. About the dataset for Brainnetome atlas construction

20 healthy, right-handed participants (10 males and 10 females; age range, 19-25) were recruited via advertisement. All subjects were examined using a Signa HDx 3.0 Tesla MR scanner (General Electric, Milwaukee, WI, USA). DTI: The DTI scheme contained a collection of 55 images with non-collinear diffusion gradients (b = 1000 s/mm2) and 3 non-diffusion-weighted images (b = 0 s/mm2) using a single-shot echo planar imaging sequence. From each participant, 45 slices were collected with field of view (FOV) = 256x256 mm, acquisition matrix = 128x128, flip angle (FA) = 90°, number of averages = 1, and slice thickness = 3 mm, with no gap. This method resulted in voxel-dimensions of 2x2x3 mm. The echo time (TE) was 64.2 ms and repetition time (TR) 10000 ms. Sagittal 3D T1-weighted images were also acquired with a brain volume (BRAVO) sequence (TR/TE = 8.1/3.1 ms; inversion time = 450 ms; FA = 13°; FOV = 256x256 mm; matrix = 256x256; slice thickness = 1 mm, no gap; 176 sagittal slices).

Q4. The methods for anatomical and functional connections

Firstly, for the probabilistic tractography, we used the brain subregions (thresholded at 50% probability) as seeds for probabilistic tractography using estimates of the (multiple) fiber orientations in each voxel.The connection probability between a seed and another voxel in the brain is given by the number of traces arriving at the target site. Furthermore, an individual-level threshold for the probabilistic fiber tracking combined with a group-level threshold for the fiber tracking success rate across all subjects was used . At the individual level, we used a conservative threshold of the connectivity probability value P ≥ 2.0 (i.e. ≥0.04% of the 5000 samples from the seed reaching target) to remove voxels with a very low connectivity probability. Next, at the group level, we maintained the consistently identified fibers and target brain areas across subjects with a success rate of ≥50%.

To reduce the number of false positives in fiber tracking, the raw tracts of each subject were first thresholded with a connectivity probability value P ≥ 2.0, that is, ≥0.04% of the 5000 samples generated from each seed voxel. The fiber tracts were then binarized and warped into the standard MNI space according to the corresponding estimated transformations. We subsequently averaged the warped fiber tracts across subjects to obtain population maps, which were then thresholded to display only those voxels that were present in at least 50% of the subjects.

Secondly, for the resting state functional connectivity, the preprocessing of the resting-state fMRI data was performed using the scripts provided by the 1000 Functional Connectomes Project (www.nitrc.org/projects/fcon_1000) (Biswal et al. 2010) with both the FSL(http://www.fmrib.ox.ac.uk/fsl/) and AFNI (Automated Functional NeuroImaging) (http://afni.nimh.nih.gov/afni) software. The preprocessing steps consisted of (1) discarding the first 10 volumes in each scan series to allow for signal equilibration, (2) performing slice-timing correction, (3) performing motion correction, (4) time series despiking, (5) spatial smoothing with a 6-mm full-width at half-maximum Gaussian kernel, (6) normalizing the mean-based intensity, (7) bandpass temporal filtering (0.01 Hz < f < 0.10 Hz), (8) removing linear and quadratic trends, (9) performing linear and nonlinear spatial normalization of the structural MR images to the MNI152 brain template (MNI152, and conducting other anatomical data preprocessing steps; including brain masking and tissue classification), (10) coregistering the anatomical volume with the mean functional volume, (11) performing nuisance signal regression (WM, cerebrospinal fluid, the global signal, and 6 motion parameters), and (12) resampling the functional data into the MNI space with the concatenated transformations. Finally, 4-dimensional (4D) residual time series data in the standard MNI space for each subject were acquired after the preprocessing. No participant had a head motion of >1.5 mm maximum translations in the x, y, and z directions or 1.5° in any angular rotation.

A 1-sample t-test (n = 20 subjects) on these maps was performed to test for areas where the averaged normalized correlation was significantly different from 0. Moreover, paired t-tests were used to identify the precise regions between each pair of TP subregions ipsilaterally that differed in their RSFC strengths. For the above voxel-wise comparisons, the false discovery rate (FDR) method was used for multiple comparison correction (P < 0.01), and only clusters containing a minimum of 30 voxels were reported here.