Background

NEXI¹ (or SMEX²) is a gray matter (GM) two-compartment diffusion model that accounts for inter-compartment exchange. It has first been implemented in a preclinical setting in vivo¹ and ex vivo². Here, we aimed to test the performance of NEXI in the human cortex in vivo, exploiting the Connectom strong gradient system to achieve short diffusion times and high b-values. We report NEXI estimates across various GM ROIs based on three implementations: the 2-compartment model in the narrow pulse approximation (NEXI), the two-compartment model accounting for gradient pulse duration² (NEXI finite pulses, or NEXI_fp), and a three-compartment extension, allowing for an extra parameter² capturing a possible Rician bias or ‘dot’ compartment (NEXI_dot). We also compare the NEXI estimates of inter-compartment exchange time t_ex with the one from the Kärger model time-dependent kurtosis^1,3 . Finally, we test the reliability of NEXI estimates by comparing intra-subject (scan-rescan) to inter-subject variability.

Methods

Four healthy volunteers were scanned (3 of whom rescanned on a different day). Acquisition: An MPRAGE was acquired for anatomical reference (1-mm isotropic resolution). Diffusion-weighted images were acquired on a 3T Siemens Connectom system using a PGSE EPI sequence with combinations of b-values of 1 (13 directions), 2.5 (25 dir.), 4 (25 dir.), 6 (32 dir.) and 7.5ms/µm² (65 dir.), and Δ=20, 29, 39 and 49ms; δ=9ms, in addition to 15 b=0 images per Δ, at 1.8-mm isotropic resolution, TE/TR=76/ 3700 ms, total scan time: 45min. Processing: Multi-shell multi-diffusion time data was preprocessed jointly; steps included MP-PCA magnitude denoising⁴, Gibbs ringing correction⁵, distortion and eddy current correction⁶. DKI tensors were calculated to extract Mean Diffusivity (MD) and Mean Kurtosis (MK) for each diffusion time, using b-values up to 2.5ms/µm². was then fit to MK to yield an estimation of ^1,3. The grey matter region of interests (ROIs) from the Desikan-Killiany-Tourville atlas⁷ were segmented on the MPRAGE image using FastSurfer⁸ and projected onto the diffusion native space using linear registration⁹ of b=0 images to MPRAGE images. The ROI-averaged powder-averaged signals were then extracted to fit the three model implementations: NEXI, NEXI_fp and NEXI_dot. The models were fit using non-linear least squares, initialized following a grid search, and yielding estimates for t_ex, the intra and extra-neurite apparent diffusivities D_i and D_e and the intra-neurite signal fraction f, as well as the potential ‘dot’ compartment signal fraction f_dot. NEXI ROI estimates were then compared between scan-rescan and across subjects.

Results and discussion

MD was almost independent of the diffusion time, with a slope of -7.5x10^-4 µm²/ms², and MK decreased more markedly with time (Fig. 1A), consistent with previous studies^1,10 and with a regime dominated by exchange rather than structural disorder. The estimated was ~30 ms, (Fig. 1B) but with a broad distribution across the cortex. NEXI parametric maps were consistent with expected neuroanatomy in the human brain (Fig. 2). However, D_i was the most challenging parameter to estimate and often hit the upper bound of 3.5 ms/µm². NEXI estimates calculated across GM ROIs (Fig. 3A) were consistent with previous studies of rat cortex in vivo¹, with however longer t_ex (median 87 ms). NEXI_fp resulted in shorter t_ex estimates (median 19 ms, more consistent with rat cortex in vivo) and higher f (0.53), possibly a signature of more myelinated and denser human cortex vs rat, or more partial volume with WM. NEXI_dot reduced the t_ex estimate further, while changing compartment diffusivities completely. The corrected Akaike information criterion (AICc) attributed poorest fitting performance to NEXI_dot, followed by NEXI_fp, and NEXI as best (Fig. 3A-B). Broad interquartile intervals overall suggest nonetheless substantial impact from partial volume with CSF and/or WM, or GM microstructure variability across the cortex. Distributions of NEXI estimates across the cortex display good inter-subject consistency (Fig. 4A) and scan-rescan reproducibility (Fig. 4B), also shown in the ROI analysis (Fig. 5), with the mean inter-subject confidence interval being larger than the mean scan-rescan confidence intervals by a factor of 1.8 for t_ex, 1.5 for D_i, 1.3 for D_e and 2.3 for f. The lower intra-subject vs inter-subject variability suggests NEXI is reproducible while sensitive to individual differences, notably with f (Figs. 4-5). Largest inter-subject variability is recorded for t_ex and f, the former likely due to fit uncertainty but the latter possibly due to genuine individual differences, which remain to be confirmed.

Conclusion

For the first time, we report NEXI model parameters estimates in human cortex in vivo. The estimated neurite fraction and compartment diffusivities are consistent with previous findings in the rat cortex in vivo, while the exchange time is longer. Our data do not favor the addition of a dot compartment to the NEXI model of two exchanging compartments. The correction for wide pulses NEXI_fp however brings t_ex estimates closer to the ones initially reported in the rat cortex (albeit with higher f), but results in poorer fit quality, which requires further investigation. Notably, the typical exchange time is in all cases longer (20 – 90ms) than reported in a previous similar study¹¹. Importantly, NEXI estimates displayed good scan-rescan reproducibility while retaining sensitivity to inter-subject differences. D_i and t_ex remain the most challenging parameter to estimate. Future efforts will focus on possible improvements, such as more advanced acquisition schemes.