Multiple sclerosis (MS) is an immune-mediated process in which an abnormal response of the immune system is directed against myelin, causing focal demyelination and axonal loss to variable extent. Diffusion tensor imaging (DTI)[1] detects microstructural white matter (WM) damage in MS[2]. Differences in fractional anisotropy (FA), mean diffusivity (MD), axial (L1) and radial diffusivity (RD) between patients and healthy controls (HC) can be observed at the focal lesion site as well as in the normal appearing WM (NAWM). Although DTI is extremely sensitive to changes in microstructure, it has limited pathologic specificity[3]. More advanced diffusion techniques can better inform about the pathophysiologic processes taking place in MS [4,5]. Here, we apply for the first time two multi-shell diffusion MRI approaches: the composite hindered and restricted model of diffusion (CHARMED)[6] and the stretched exponential model (SEM)[7]. CHARMED provides maps of the restricted water fraction (FR), a proxy for the axonal density, while SEM generates maps of the heterogeneity index (MA), sensitive to different diffusivity domains inside the voxel. Our aim was to assess the ability of these methods to differentiate between MS lesions, NAWM and HC as compared to the natural variability found in control tissue.7 MS patients and 5 age/gender-matched HC underwent a comprehensive MRI protocol at 3T/7T, comprising: an MPRAGE(3T), multi-shell diffusion with b=700/2000s/mm2 and 27/45 gradient orientations at 1.5mm isotropic resolution (3T/7T), two double inversion recovery (DIR) acquisitions[8] targeted to retain grey or WM signal respectively (3T/7T) and a high resolution MP2RAGE(7T). Diffusion data were pre-processed using FSL-EDDY [9] and analysed in native space using conventional DTI[10] (using only b=700s/mm2), using CHARMED[2] and using SEM[3], to extract maps of FA,MD,L1,RD,FR,MA. All the maps were linearly registered to the high resolution anatomical. For each subject, the anatomical map was then nonlinearly warped to the FSL template using ANTs[11] and the obtained transformation was applied to all diffusion maps. MS lesions were manually segmented using the DIR maps(Fig.a); the lesion masks were then nonlinearly warped to the template. The tracts affected by the MS lesion were identified by using WM labels in standard space [12]. Then, for each subject and for each lesion whose volume occupied more than 5% of the ROI, mean and standard deviation(SD) of diffusion indices were calculated in the intersection between the lesion area and the ROI, in the contralateral ROI (when present and not affected by lesions) and in the same ROI across the healthy population, for a total of 13 lesions monitored(Fig.b). For each index, the % change between HC and lesion and between HC and contralateral NAWM (when present) was calculated and normalized to the SD of the parameter in the HC. Then, the changes were averaged casting the lesions in two groups: those outside the corpus callosum (CC), for which also the difference between HC and NAWM was calculated, and those affecting the CC. The differential performance between 3T and 7T was also evaluated in terms of differences between values in the lesions and in the healthy corresponding ROI.

For lesions affecting WM areas outside the CC(Fig.c), FR and MA are the most sensitive indices at 3T, while FR is the most sensitive index at 7T, when comparing the lesion site with healthy tissue. Conversely, in the CC(Fig.d), which is characterized by a highly coherent fiber organization, DTI indices (FA) have the best performance in detecting differences between HC and lesion, closely followed by FR. This can be explained by the fact that in tissue characterized by coherent fiber organization, the tensor model is appropriate to describe the diffusion dynamics, while in areas of complex fiber architecture, more advanced models are required[13]. FR shows consistent discriminating performance across both cases. All indices show that NAWM has an intermediate behavior between focal lesion and HC. FR has best performance at 3T in discriminating between healthy and NAWM, while MD has the best performance at 7T(Fig.c). Interestingly, the performance of all indices is 1-3 times better at 7T than at 3T, except for MA (little or no impact of field strength)(Fig.e).At 7T, CHARMED-FR gives the best discriminability between MS lesions and HC across the WM. In areas of low fiber dispersion such as the CC, DTI-FA&L1 were found to discriminate well the diffusion dynamics of the tissue, but showed less lesion discrimination outside the CC. We therefore conclude that CHARMED (more than DTI) at 7T (more than at 3T) can improve detection of MS lesions. Further statistical quantification over a larger number of lesion sites can serve to validate these findings.