MFCSC: Novel method to calculate mismatch between functional and structural brain connectomes, and its application for detecting hemispheric functional specialisations

We introduce a novel connectomics method, MFCSC, that integrates information on structural connectivity (SC) from diffusion MRI tractography and functional connectivity (FC) from functional MRI, at individual subject level. The MFCSC method is based on the fact that SC only broadly predicts FC, and for each connection in the brain, the method calculates a value that quantifies the mismatch that often still exists between the two modalities. To capture underlying physiological properties, MFCSC minimises biases in SC and addresses challenges with the multimodal analysis, including by using a data-driven normalisation approach. We ran MFCSC on data from the Human Connectome Project and used the output to detect pairs of left and right unilateral connections that have distinct relationship between structure and function in each hemisphere; we suggest that this reflects cases of hemispheric functional specialisation. In conclusion, the MFCSC method provides new information on brain organisation that may not be inferred from an analysis that considers SC and FC separately.


Supplementary Information Text
We analysed 50 pre-processed adult datasets from the Human Connectome Project (HCP) 1 , which were acquired on a customized Siemens Magnetom Skyra 3T MRI system using a multiband pulse sequence 2-5 .
To further process the "minimally pre-processed" HCP data consisting of vowel-wise diffusion weighted measurements, we first applied bias-field correction 16 . This was followed by multi-shell multi-tissue constrained spherical deconvolution (MSMT-CSD) 17,18,19 to model white matter, grey matter and cerebrospinal fluid 19 , with a maximum spherical harmonic degree Lmax = 8.

Connectome generation
For each subject, tractogram construction included several steps: generation of 10 million probabilistic streamlines using the 2 nd -order Integration over Fibre Orientation Distributions algorithm (iFOD2) 20 and anatomically-constrained tractography (ACT) 21 , with dynamic seeding 22 : FOD amplitude threshold 0.06, step size was half of voxel size, length of 5-300 mm, and backtracking, i.e. possibility for tracks to be truncated and re-tracked if a poor structural termination is encountered 21 . In the next step, each streamline was assigned a weight computed using SIFT2 (ref. 22 ); to achieve that, SIFT2 runs global optimisation to select weights that would fit the tractogram to the underlying data best. Connection strengths were calculated by summing the weights of the streamlines that connect each pair of parcellated regions-of-interest. To find which two regions-of-interest a streamline connects (if any), we performed a radial search of 2 mm from each of its endpoints.

Functional data
Acquisition and pre-processing The resting-state functional MRI protocol of this dataset used the following parameters: TR = 720 ms, using a multiband factor of 8; TE = 33.1 ms; flip angle = 52°, 2 mm isotropic resolution (FOV: 208 mm × 180 mm, Matrix: 104 × 90 with 72 slices). The four available runs of 14 min and 33 sec each (1200 volumes) were analysed separately in this study 7,8 .
Pre-processing steps are described in detail elsewhere 8 . Briefly, pre-processing included gradient distortion correction, motion correction, fieldmap-based EPI distortion correction, brainboundary-based registration of EPI to structural T1-weighted volume, non-linear registration into MNI space, and intensity normalization. The images are then denoised using the ICA-FIX method (FMRIB's ICA-based Xnoiseifier) 23,24 .

Connectome generation
The pre-processed data consisted of voxel-wise time series representing denoised BOLD signal fluctuations. We analysed these data separately for each subject k. Using the same 84-nodes parcellation as previously described, we calculated the time-course of each region-of-interest by averaging the BOLD signals of its voxels, separately for each of the four runs. The functional connectome of each run was computed using Pearson's correlation coefficient between the timecourse of each pair of regions-of-interest. We did not take the absolute of the correlation coefficients, and therefore the functional connectomes included negative values. For each subject, the four runs are then averaged to form a single mean functional connectome. This averaged out run-specific effects (range of cognitive and attentional states, influence of head motion, quality of registration to structural scan, etc.), reducing inter-subject variability in FC (ref. 25,26 ). Processing of the functional data was performed with Matlab (MathWorks, Natick, MA).  insula-banks of the superior temporal sulcus insula-pars opercularis insula-precentral insula-precuneus insula-superior frontal insula-supramarginal banks of the superior temporal sulcus-precentral precentral-superior temporal precentral-supramarginal superior frontal-cerebellum a Bilateral pairs are labelled according to the two brain regions that the unilateral connections connect in each hemisphere. Note that the order of the two regions is arbitrary because the unilateral connections may include axons that project either way. b The bilateral pair is shown in Fig. 2 of the main text for selected subjects.  Table S1.