To improve the NODDI estimates of CSF volume fractions by accounting for tissue-specific T2.Neurite orientation dispersion and density imaging (NODDI) has become a popular diffusion MRI technique to provide microstructural information of axons and dendrites. The technique is based on combining a three-compartment tissue model with a multi-shell high-angular-resolution diffusion imaging (HARDI) protocol. NODDI estimates the intra-cellular volume fraction (neurite density, ficvf), orientation dispersion of the neurites (odi), fibre orientation and CSF volume fractions (fiso) [1]. MR signal is modelled using the 3-compartment model as: $$S=(1-f_{iso})(f_{icvf}S_{ic}+(1-f_{icvf})S_{ec})+f_{iso}S_{iso} (1)$$ Where: S is the full normalised signal. S_ic, S_ec and S_iso are the normalised signals for intra-cellular, extra-cellular and CSF compartments, respectively. However, NODDI assumes a common T2 for all compartments; while this is acceptable for intra-cellular and extra-cellular compartments, the CSF compartment has a much higher T2, resulting in a significant overestimation of fiso, especially in white matter. We have studied the extent to which this overestimation affects the important parameters of odi and ficvf. Furthermore, we have explored the possibility of improving the NODDI parameters estimation by combining DESPOT [2] and NODDI. We propose using the acquired T2 maps for the isotropic component and the intra and extra-cellular component from DESPOT to account for the longer T2 times in the CSF and scale fiso to compensate for the CSF overestimation.We acquired NODDI diffusion data for 5 male subjects (between 25 and 30 years old, median age 27) using a Siemens 1.5T Avanto scanner (maximum gradient strength=44mT/m). One of the subjects was rescanned on a Siemens Prisma 3T scanner (maximum gradient strength=80mT/m) for comparison. Images were acquired using echo-planar imaging, TE/TR=95/8200ms, fov=192x192mm², matrix size 96x96, slices=60, thickness=2.5mm. Two diffusion shells were used, b=800 and 2000s/mm² with 30 and 60 non-collinear directions, respectively. We acquired T2 maps for the isotropic component and the intra and extra-cellular component using multi-component driven equilibrium single pulse observation (mcDESPOT)[2]. First, we run full NODDI analysis on the data to generate fiso, ficvf and odi parameter maps. Then, we generated corrected fiso values (fiso^corr) that account for the longer T2 times in the CSF, as follows: $$fiso^{corr}=\frac{fiso\cdot e^{-{TE}/{T2_{aniso}}}}{fiso\cdot e^{-{TE}/{T2_{aniso}}}+(1-fiso)\cdot e^{-{TE}/{T2_{iso}}}} (2)$$ Rescaling fiso was done using voxel-based T2_iso and T2_aniso values from the CSF and the intra-extracellular T2 maps. And then we run a new NODDI analysis using the rescaled fiso to generate new parameter maps.

The mean values of NODDI maps, for a selection of brain regions from original NODDI and from using rescaled-fiso NODDI are shown in Table 1. We chose the frontal lobe and caudate for cortical and sub-cortical grey matter, and three different white matter tracts. Original NODDI overestimates CSF volume fractions, particularly in white matter regions with values reaching over 0.2. Similar observations can be made from ficvf maps for different brain regions, with unrealistic values (ficvf>0.7 being commonplace in WM). Scaling fiso using equation 2 reduces fiso to more expected physiological values. This is visibly clear on the fiso map image as shown in Fig.1.

Rescaling fiso with T2 maps generated low fiso measures consistent with those reported in literature,[3], with white matter fiso<10% and grey matter fiso<25%. We examined the change in NODDI maps after rescaling fiso, Fig.2. We noticed less effect on odi for different odi values. However, the lower the fiso, the more significant the fiso change was. Reduction in ficvf reached more than 20% of the original NODDI’s estimate resulting in fewer unrealistic values. There is a linear correlation of the change in ficvf with its magnitude in white matter. We compared the results from using 1.5T scanner with those using 3T, Fig.3; we found fiso rescaling brings the resulting fiso and ficvf measures closer, whereas odi was not sensitive to the change.

Using T2 maps provides voxel-based T2_iso and T2_aniso, but in their absence, we suggest using fixed values for all voxels in equation2. Fig.4 shows that rescaling with fixed T2_iso and T2_aniso is best achieved using values with a ratio T2_iso/T2_aniso>10. Another suggestion is to provide NODDI with pre-calculated fiso maps using other tools such as CSF map estimated by SPM or FSL’s segmentation.

NODDI overestimates CSF volume fractions (fiso), which is more apparent in white matter regions. This overestimation is due to assuming a single T2 for all compartments. High fiso is a likely cause for an inaccurate measure of ficvf and odi that has implications on NODDI studies. We have demonstrated a method to correct these errors by scaling fiso with voxel-based T2 maps in order to generate more physiologically believable measures.