Multiple sclerosis (MS) lesions are well known to alter tissue segmentation, shifting tissue boundaries between grey and white matter regions (GM and WM). Despite evidence of these errors occurring when working with anatomical images, little is known about the possible effects of MS lesions on the functional MRI results. Here, we addressed this question by simulating the presence of MS lesions on resting state functional MRI data from healthy controls. Subsequently, we tested whether lesion filling functional MRI data is useful to prevent artefactual results of functional connectivity alterations that are actually due to MS lesions.
19 healthy controls (HC) and 19 secondary progressive MS (SPMS) subjects underwent MRI examination using a 3T Philips Achieva scanner (Philips Healthcare, Best, The Netherlands) with a 32-channel head-coil.
MRI acquisition: For each subject, rs-fMRI scans were acquired using a fast field echo-echo planar imaging (FFE-EPI, TR/TE = 2600/35ms, flip angle=90°, voxel size=3 mm isotropic, FOV=192x192 mm2, 46 slices, 120 volumes). A 3DT1 volume was also collected (TR/TE=6.9/3.1ms; flip angle 5°;180 sagittal slices; voxel size=1x1x1mm3, FOV=256mm).
Fake lesions generation: For each HC subject, each volume of the rs-fMRI images was masked with the lesion mask obtained from one (distinct) SPMS subject. A synthetic lesion generation process on rs-fMRI was performed using the NiftySeg and NiftyReg software packages3 as described in Prados (2016)2 in order to obtain 19 faked-MS rs-fMRI series with a realistic distribution of WM lesions (Fig.1). The 3DT1 images of HCs were not modified in order to simulate lesion-filled 3DT1 volumes according to standard recommendations for MRI studies in MS4.
rs-fMRI lesion filling: The multi-time-point filling (mtpF) approach proposed by Prados (2016)2 was applied to fill the simulated lesions on the faked-MS rs-fMRI images, therefore obtaining filled-faked-MS rs-fMRI images (Fig.2).
rs-fMRI analysis: We hypothesised that WM lesions neighbouring GM boundaries may affect GM areas through the fMRI processing pipeline, which includes image smoothing and a normalisation step to move fMRI images into standard space. To test this hypothesis, for each subject we calculated the transformation matrix from fMRI to 3DT1 and then to MNI standard space using the real HC data (rs-fMRI and 3DT1). The resulting transformations were then applied to the subject’s faked-MS rs-fMRI and filled-faked-MS rs-fMRI images (Fig.3). This operation ensured that any difference in the subsequent analysis was therefore due to the WM lesions spill over. The pre-processing of the normalised rs-fMRI datasets was then completed using the FSL FEAT pipeline5.
Finally, the Independent Component Analysis (ICA) and dual regression steps were used to compare voxel-wise the FC between the real original (true-HC) and the faked-MS rs-fMRI as well as between the true-HC and the filled-faked-MS rs-fMRI to validate the performance of the lesion filling procedure.
1 – Castellazzi G (2018) Front Neurol 9:690
2 – Prados F (2016) NeuroImage 139:376–384
3 – NIFTYK, cmictig.cs.ucl.ac.uk/research/software/software-nifty
4 – González-Villà S (2017) Neuroimage Clin15:228-238
5 – FSL FEAT, fsl.fmrib.ox.ac.uk/fsl/fslwiki/FEAT