In a recent simulation study, Drobnjak et al¹ demonstrates that low-frequency oscillating gradient spin-echo (OGSE) sequence is more sensitive to axon diameter than conventional pulsed gradient spin-echo (PGSE) sequence when fibre orientation is unknown or when fibre dispersion exists. Here, we experimentally validate this claim. We image a live rat sciatic nerve tissue using both sequences and compare its agreement with histology. Our aim is to investigate whether OGSE provides more accurate, and more precise measurements, and how robust these estimates are to the number of diffusion gradient directions.

Tissue sample preparation: The sciatic nerve of a freshly sacrificed rat was carefully ligated, excised and then immediately immersed in gassed artificial cerebrospinal fluid (aCSF). The sample was transferred to a MRI compatible incubation chamber², where it was bathed in circulating aCSF at a constant temperature of 37°C, for the whole duration of the scan, to simulate an in-vivo environment. At the end of the scan, the nerve sample was fixed and histology of the same slice was later obtained with transmission electron microscopy (fig.1b,1c).

Optimised imaging protocol: We used the ActiveAx optimisation framework described previously by [3,4] to generate optimal OGSE and PGSE imaging protocols for 8, 16 and 32 gradient directions (fig.2). The algorithm minimizes the Cramer-Rao Lower Bound on parameters of a model to fit to a set of measurements. The tissue model used was a simplified composition of hindered and restricted diffusion compartments^3,5,6 with a single axon diameter. The apriori model parameters reflected the expected microstructure in rat sciatic nerve and were taken from literature^6,7,8: intrinsic diffusivity (D_i=1.7µm²/ms), intra-axonal volume fraction (f_icvf=0.6) and mixture of diameters (α=2.26µm,4.50µm,6.74µm). The protocols were run on a 9.4T Agilent Technologies, Inc. pre-clinical system with a diffusion gradient strength of 800mT/m. A single slice fast spin-echo readout with the following settings were used: echo train length=8, resolution=0.09×0.09×1mm³,8 repetitions, TR=1100ms and total acquisition time of 12 hours.

Model fitting: We used a three stage fitting model procedure^3,6 (grid search, gradient descent, Markov Chain Monte Carlo (MCMC) posterior sampling (2000 burn-in, 30000 iterations)) to fit the model parameters to the measured signals from each voxel. The final estimated parameters were from the gradient descent stage. Their precision was described by the width of posterior distribution of the parameter obtained from the MCMC stage. We used TortZeppelinCylinder (as described by [9]) as our MR signal model. The parallel cylinders of a single radius represented the axons and TortZeppelin accounted for contributions to the signal from anisotropic and tortuous movement of water molecules in the extracellular space.

Histological images were processed with an in-house segmentation algorithm to calculate the mean and standard deviation of axon diameter, α,3.64±0.34 µm, and intra-axonal volume fraction, f_icvf,0.51±0.09 (fig.1). The 32 gradient direction OGSE protocol provides more accurate diameter estimates (fig.3bii) than those from the equivalent PGSE protocol (fig.3aii). OGSEs also exhibit significantly lower uncertainties in their parameter estimates of α and f_icvf in comparison to PGSEs (fig.4). Furthermore, OGSE diameter estimates in the nerve changes significantly less than those from PGSE protocols when the number of directions were reduced from 32 to 16 and then to 8 directions (fig.5). A similar but not statistically significant pattern was observed for f_icvf.

Here we experimentally validated the simulation study in [1] using live nerve tissue and showed that low-frequency OGSE improves sensitivity to axon diameter in comparison to PGSE. We showed that OGSE provides more accurate and more precise axon diameter estimates compared to PGSE. Furthermore OGSE estimates are more robust to fewer gradient directions allowing for less measurements and faster scanning.

OGSE and PGSE sequences both seem to overestimate the true diameter and underestimate the intra-axonal volume fraction in the nerve tissue. This is possibly due to axon diameter shrinkage¹¹ and further water loss from the extracellular compartment which happens during the tissue fixation process.

Our OGSE protocols contained low and high frequency waveforms (fig.2b). As shown in [1], the low-frequency OGSE provides higher sensitivity to axon diameter compared to PGSE because the free diffusion component of the restricted signal attenuates less in OGSE due to their lower b values, giving rise to higher SNR, and facilitating better detection of the smaller components. The high frequency OGSE, on the other hand, enhances sensitivity to estimates of diffusivity^1,11.

This is the first time the comparison between OGSE and PGSE has been done in live nerve tissue using model based approach. Our study gives encouraging evidence for the superiority of low-frequency OGSE over PGSE protocols providing more accurate and precise axon diameter estimates.