Desktop version

Home arrow Computer Science arrow Computational Diffusion MRI: MICCAI Workshop, Athens, Greece, October 2016

Source

Discussion

In this article we explore the sensitivity of OGSE to microstructural dimensions of microcapillaries of unknown orientation on a clinical scanner. We find that 10 and 20 p,m micro-capillary diameters can be accurately and precisely estimated whereas

5 p,m estimates are neither accurate nor precise. We also find that low frequency OGSE sequences give the best results and are optimal for parameter estimation. In particular, N = 3 OGSE sequence can be used on its own to give estimates that are very similar to those of the combined OGSE frequencies (N = 1 to N = 9).

Our observations support the theoretical findings in [12,18] regarding the clinical scanner diameter resolution limit which, based on their calculations, for gradient strength of G = 62mT/m, is approximately 6 p,m for SNR ^ 50. We get excellent estimates for 10 and 20 p,m plates and can assume that the same would be true for the diameters of microcapillaries within this range (a e [10,20] p,m). On the other hand, 5 p,m diameters cannot be estimated as they fall bellow the resolution limit. In our study, we used idealised phantom plates (homogeneously and densely packed with microcapillaries), which were imaged with a HARDI type acquisition, pushed to the clinically feasible limits. We used a ‘long’ TE= 120 ms (in terms of standard clinical settings) in order to allow for larger diffusion weighting which is necessary to improve the sensitivity to the smaller diameter microcapillaries (5 p,m). We also maximised SNR(> 45) on the clinical scanner by imaging the phantom ensemble with a surface coil and using water as the substrate (long T2 relaxation time^ 1500ms). Yet for a gradient strength of 62mT/m, the diffusion weighted signal for the 5 |i,m microcapillaries could not be differentiated from a diffusion signal for 0|i,m microcapillaries. This highlights that diameters of 5 |i,m cannot be estimated on clinical scanners even under idealised conditions. On the other hand, when we place the same 5 |xm plates in a pre-clinical scanner with 800 mT/m gradients we estimate 5 |xm perfectly (data not shown), suggesting that the sole reason for the results is the insufficient gradient strength.

Our analysis of individual OGSE sequences shows that there is an optimal range of OGSE lobes, for estimation of diameters of microcapillaries and intrinsic diffusivity. The optimal OGSE shells are with low number of lobes, (Ne{2, 3, 4}) and their parameter estimates are accurate and precise, especially for N = 3. Our experimental findings are consistent with the recent ActiveAx simulation study [12] and spectroscopy study [17], which show that OGSE with lower N are optimal for the measurement of fibre diameters. The result highlights the importance of optimisation for microstructure indices estimation.

In this work we analysed the sensitivity of OGSE sequences to fibre diameter in micro-capillaries. Based on theoretical studies which compare OGSE and SDE sequences [12, 18], we do not expect SDE based techniques to provide better diameter estimates, with the same gradient constraints. Although we have not directly compared the sensitivity of other more complex sequences (e.g. NOGSE [26]), similar conclusions hold, as the sensitivity and resolution limit is driven by the maximum gradient strength and pulse duration [18].

The phantom we use in this study is much simpler than in vivo nerve tissue. However, the purpose of this work is to test the innate sensitivity of OGSE sequences to fibre diameters on a clinical scanner, which requires ideal diffusion substrates. We expect that results for in vivo nerve tissue to be similar or worse. For instance, resolution limit would be lower, i.e. since 5 |i,m diameter can not be estimated in an ideal phantom with extremely long T2 of pure water and simple parallel cylindrical capillaries, then its potential to be estimated in vivo is further reduced. As for the optimal frequency of the OGSE , the exact value would be different, however it is predictable that it would be low frequency. The benefits of using physical phantoms with known geometry and microstructural characteristics are numerous. They are not degradable over time and are easy to use in validating microstructure imaging protocols [27], even over multiple clinical trial sites. There are other ongoing development of more complex phantoms such as biomimetic phantoms [28] being developed for validating diffusion MR imaging. However, the simplicity of the plates used in this study is also ideal for validation and calibration purposes. In future, we plan to develop an integrated phantom with a more finely graded range of microcapillary diameters to explore the resolution limit with more accuracy.

Overall, our results suggest that imaging axon diameters in vivo in the brain using standard clinical scanners with gradient strength of 60-80 mT/m would be very challenging. This work, combined with the theoretical work by Drobnjak et al. [12] and Nillson et al. [18], provides further evidence for validity of models of brain nerve tissue where axons are represented as sticks and not as cylinders [29]. On the other hand, our work also demonstrates that axon diameter mapping is still a possibility in the peripheral nervous system, where axons are larger (1-8 |i,m [30]).

Axon diameter imaging of the peripheral nerves using clinical scanners can potentially play a crucial role in the understanding of nerve tissue regeneration— a mechanism unique to peripheral nerves and with correlation to axon diameters [31].

Acknowledgements We thank EPSRC for funding the research studentship of Lebina Shrestha Kakkar. EPSRC grants EP/I018700/1 and EP/H046410/1 also contributed to this work. The study was undertaken at UCH and UCL, both of whom are part-funded by the Department of Health NIHR Biomedical Research Centres funding scheme.

 
Source
Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >

Related topics