Hamied A Haroon^{1}, Ben R Dickie^{1}, Faezeh Sanaei Nezhad^{1,2}, Martyn McFarquhar^{1}, Stephen R Williams^{1}, Geoff JM Parker^{1,2}, and Laura M Parkes^{1}

Capturing the earliest signs of dementia with MR imaging relies on techniques that are sensitive to the subtle loss or disconnection of neurons before atrophy occurs. Models of multi-shell HARDI such as NODDI claim to quantify neurite density *in vivo* and non-invasively, but the specificity of these HARDI-based metrics remain unvalidated. This study aims to determine the sensitivity of NODDI’s neurite density and orientation dispersion index to regional variation of MRS markers of neuronal and glial cell density. We find that caution must be exercised when interpreting NODDI’s neurite density as related to neuronal density. Orientation dispersion instead appears to be a closer marker of neuronal density and may be a more sensitive marker of disease-related change.

**Data Acquisition**: 6 healthy volunteers (3 F, age: range 20-50 years, median 27 years) underwent MR imaging on a 3.0 T Philips Achieva (Philips Medical Systems, Best, The Netherlands). We acquired: (1) High-resolution T1-weighted volume, (2) Multi-shell HARDI, and (3) MRS. Acquisition parameters are given in Table 1(A).

**Analyses**: (1) Matlab (The MathWorks, Natick, MA, USA) scripts were used to overlay spectroscopy VOIs on SPM8 (http://www.fil.ion.ucl.ac.uk/spm/) probabilistic segmentations of the T1-weighted volumes in order to obtain percentages of CSF (CSFp), grey matter (GMp) and white matter (WMp) in each VOI. Tissue grey matter fraction (GMf) was calculated as GMf=GMp/(GMp + WMp). (2) HARDI data were pre-processed with FSL’s (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki) Topup and Eddy to correct distortions, and then analysed using FSL’s DTI-FIT to obtain maps of mean diffusivity (MD) and fractional anisotropy (FA) and also using the NODDI Matlab toolbox to obtain maps of neurite intracellular volume fraction (ICVF) and orientation dispersion index (ODI) (Figure 2). (3) The SPM8 segmentations and VOIs were registered to diffusion space using FSL’s FLIRT, and mean VOI values computed for MD, FA, ICVF (< 1) and ODI. (4) PRESS spectra were quantified in jMRUI v6.0 (http://www.jmrui.eu) using a simulated basis set of 8 metabolites including NAA and MI in QUEST. Metabolite concentrations were computed relative to water and corrected for CSFp.

**Statistics**: Linear mixed-effects models were fit as follows in R (version 3.4.2, R Foundation for Statistical Computing, Austria) using the nlme package^{5}. To account for repeated measurements across VOIs, subject ID was included as a random effect (random intercept) with GMf, [NAA] and [MI] included as fixed effects. Separate models were fit for ICVF, ODI, MD and FA as the response variable. Inference was conducted using asymptotic χ^{2} tests on the resultant fixed effect coefficients using the “car” package^{6}. Significance was assessed as p<0.01 to correct for multiple comparison tests.

1) Zhang H, et al. NODDI: practical in vivo neurite orientation dispersion and density imaging of the human brain. Neuroimage 2012; 61(4):1000-1016.

2) Urenjak J, et al. Specific expression of N-acetylaspartate in neurons, oligodendrocyte-type-2 astrocytes and immature oligodendrocytes in vitro. J Neurochem 1992; 59:55-61.

3) Nordengen K, et al. Localisation of N-acetylaspartate in oligodendrocytes/myelin. Brain Struct Funct 2015; 220(2):899-917.

4) Brand A, et al. Multinuclear NMR studies on the energy metabolism of glial and neuronal cells. Dev Neurosci 1993; 15:289-298.

5) Pinheiro JC and Bates DM. Mixed-effects models in S and S-Plus. Springer 2009.

6) Fox J and Weisberg S. An R companion to applied regression. 2nd edition. SAGE 2011.

7) Guimaraes AR, et al. Quantitative in vivo 1H nuclear magnetic resonance spectroscopic imaging of neuronal loss in rat brain. Neuroscience 1995; 69:1095-1101.

8) Tsai G and Coyle JT. N-acetylaspartate in neuropsychiatric disorders. Prog Neurobiol 1995; 46:531-540.