Introduction

Diffusion MRI (dMRI) at high b-values might be sensitive to cellular modifications. But how much increase of b-value is conveying additional biological meaning remains unclear. This work aims to determine the impact of very high b-values on DTI metrics measured within the cortex and studies how much the b-values increase is helpful to capture changes in tissue microstructure as assessed by the cortical myelination from the T1WI/T2WI map.

Materials and methods

We scanned a set of 9 healthy volunteers with the dedicated protocol at 3 increasing b-values of 1000, 3000 and 5000 s/mm² together with T1-WI and T2-WI. Averaged maps across the 9 subjects superimposed to surface were obtained for estimated myelin content, cortical thickness, mean diffusivity (MD) and fractional anisotropy (FA). We used the Human connectome project (HCP) multi-modal parcellation version 1.0 (HCP_MMP1.0 210P MPM version)¹ to separate 180 cortical parcels by hemisphere and to measured mean values in each parcel of the averaged. Averaged cortical surface map of myelin (T1-WI/T2-WI) was compared to surface maps of MD computed from each b-value (named MD₁₀₀₀, MD₃₀₀₀ and MD₅₀₀₀) in 360 cortical parcels using Spearman correlations, multiple linear regressions and Akaike information criteria (AIC).

Results

Surface MD maps averaged across the 9 subjects are first presented in Figure-1. From a qualitative inspection, we found that MD values progressively decreased with b-values, which was expected² (Figure-1a). Interestingly, when window/level was adjusted individually, variations of MD values were observed from one cortical area to another with different patterns according to b-values (Figure-1b).
Figure-2 displays the lateral and medial views of the same inflated averaged surface maps in which the boundaries from the HCP parcellation have been overlaid. It showed that MD₃₀₀₀ and MD₅₀₀₀ maps inversely mirrored myelin map. Overall, MD_3000-or-5000 appeared low in high myelinated cortical areas (such as the primary motor cortex, the primary auditory cortex or the primary visual cortex) whereas MD_3000-or-5000 appeared high in less myelinated cortical areas (such as the fronto-polar cortex). However, this pattern was not found on MD₁₀₀₀ maps in which higher MD values were mainly detected within the thinnest cortical parcels (such as the primary somatosensory cortex 3b). To illustrate this observation, the normalized signal was plotted versus b-value in a highly myelinated parcel (primary motor cortex) and in a lightly myelinated parcel (fronto-polar cortex) in figure 2-f. The curves showed that signal decay was similar between the 2 regions at b=1000s/mm², which implies an absence of contrast on MD₁₀₀₀ maps, while dissociation of the two curves was observed at b=3000s/mm² and even more at b=5000s/mm². We observed a negative correlation between estimated myelin and MD that became stronger as long as b-value increased with Spearman’s rank correlation rho=-0.112 (p=0.03), -0.508 (p< 0.001), and -0.518 (p< 0.001) for MD₁₀₀₀, MD₃₀₀₀ and MD₅₀₀₀ respectively. Concerning MD and cortical thickness, we found a negative correlation for MD₁₀₀₀ (rho=-0.228; p< 0.001) while positive and weaker correlations were found for MD₃₀₀₀ and MD₅₀₀₀ (rho=0.131, p< 0.001; and 0.174, p< 0.001 respectively).
Because cortical thickness and myelin content are correlated, we performed multiple regression statistical models to be unsensitive to the cortical thickness confounding factor. We obtained a significant improvement of the adjusted R² between myelin and MD from MD₁₀₀₀ to MD₃₀₀₀ and MD₅₀₀₀ (R²=0.33, p<0.001; R²=0.43, p<0.001; and R²=0.50, p<0.001) with regression coefficients β that became more and more negative (figure 3). Comparisons of the 3 models showed better performances with MD₅₀₀₀ (AIC_^MD5000< AIC_^MD3000 < AIC_^MD1000).

Discussion

The current study found that increasing b-values incrementally improves the sensitivity of MD fitted from a DTI model to capture the fine variations of cortical myelin content. A complexity with high b-values is that diffusion becomes non-Gaussian^3,4. A large number of alternatives to the DTI model has been developed to address the high b-value component, and many of them showed an improved sensitivity compared to DTI model at standard b-value to capture microstructural alterations associated with different conditions^3,4. However, these methods all require collecting multiple shells to fit the signal which limits their dissemination because of scan time constraint. Interestingly, we used a simple DTI model to fit the data even if DTI-assumption of Gaussian distribution of the diffusion probability is violated at high b-value. However, such model is efficient and robust and the decrease of cortical MD that we observed along b-values, as reported by others⁵ is accounted for by the non-monoexponentiality of the signal driven by slow moving water molecules. Thus, even if this approach cannot comprehensively characterize the signal, we showed it could be sensitive enough even in grey matter.

Conclusion

The current study found that increasing b-values from standard (1000 s/mm²) to high (3000 s/mm²), and up to very high values (5000 s/mm²), incrementally improves the sensitivity of MD fitted from a DTI model to capture the fine variations of cortical myelin content. The averaged surface maps of MD seemed to inversely mirror the averaged myelin map (assessed by T1-WI/T2-WI) more and more strikingly as long as b-values increased. Finally, we confirmed a significant improvement of the adjusted R² between myelin and MD with a maximum for MD₅₀₀₀.

Acknowledgements

No acknowledgement found.