Simulating axon packing for investigating white matter tissue characteristics with diffusion MRI
Hamed Y. Mesri1, Dmitry S. Novikov2, Max A. Viergever1, and Alexander Leemans1

1Image Sciences Institute, University Medical Center Utrecht, Utrecht, Netherlands, 2Bernard and Irene Schwartz Center for Biomedical Imaging, Department of Radiology, New York University School of Medicine, New York, NY, United States

Synopsis

A novel algorithm for simulating axon packing in nerve bundles is proposed. Statistical analysis of the results demonstrates that, in contrast to conventional methods, the proposed method eliminates the bias in the estimated distribution and achieves higher packing densities, while preserving the random nature of the axon packing structure. The resultant tissue models can be used subsequently to study the Brownian motion of water molecules within nerve bundles. With our novel axon packing simulation framework, the effect of axon properties on the derived diffusion-weighted MR signal can be investigated more reliably now.

BACKGROUND AND PURPOSE

Simulation of Brownian motion of water molecules in WM tissue plays an important role in validating models that describe the link between the measured diffusion-weighted MR signal and the underlying tissue microstructure.1-5 In such a simulation framework, it is crucial that specific tissue characteristics mimic those observed in real tissue as closely as possible. Failing to do so may induce a considerable bias that could confound subsequent analyses.

In this work, we present a novel approach for simulating the axon packing in nerve bundles, which optimizes tissue features in terms of fiber configuration. We demonstrate that in comparison with the conventional techniques,3,6,7 our proposed method achieves higher packing densities and eliminates the bias in the radii distribution, while preserving the short-range disorder of the axon packing structure.

METHODS

The distribution of axon radii within a nerve bundle, can be approximated with a gamma distribution.8,9 In our proposed packing algorithm, axons are represented by circles with radii generated from a gamma distribution and are assumed to be confined in a square with width $$$W$$$. The circles do not overlap and are separated by interfaces with specific boundary thickness, $$$h=\eta r$$$, where $$$r$$$ is the axon radius and $$$\eta$$$ is the separation parameter.

Given a desired set of model parameters, the algorithm starts with generating a circle in the center of the square continuing in levels around the first circle. At each step, a new circle is placed such that its boundary is tangent to the boundaries of the two previously placed neighboring circles, i.e., $$\begin{cases}(x_i-x_{L_1})^2+(y_i-y_{L_1})^2=(r_{L_1}+h_{L_1}+r_i+h_i)^2\\(x_i-x_{L_2})^2+(y_i-y_{L_2})^2=(r_{L_2}+h_{L_2}+r_i+h_i)^2\end{cases},$$ where $$$(x_i,y_i)$$$ are the coordinates of the center of the $$$i$$$-th circle with radius $$$r_i$$$ and boundary thickness $$$h_i$$$, and $$$(x_{L_1},y_{L_1})$$$, $$$(x_{L_2},y_{L_2})$$$ are the coordinates of the center points of the first and second neighboring circles as depicted in Fig.1. The algorithm terminates when the number of placed circles reaches $$$N_A$$$. For a given packing density ($$$\phi$$$), $$$N_A$$$ can be calculated as $$N_A=\frac{W^2\phi}{(1+\eta)^{2}v_0},$$ where $$$v_0$$$ is the mean axon area. For radii generated from a gamma distribution, $$$Gamma(\alpha,\beta)$$$, where $$$\alpha$$$ and $$$\beta$$$ denote the shape and scale parameters, $$$v_0$$$ can be calculated as $$v_0=\pi\frac{\Gamma(\alpha+2)\beta^2}{\Gamma(\alpha)},$$ where $$$\Gamma(\alpha)$$$ is the gamma function. The flowchart of the algorithm is presented in Fig.2.

Monte-Carlo simulations (1000 repetitions) with model parameter settings as shown in Fig.3(a) were performed to evaluate performance in terms of bias in estimated mean radius ($$$\mu_r$$$), radius variance ($$$\sigma_r^2$$$) and packing density ($$$\phi$$$). In addition, the structure correlation function (SCF) in Fourier domain is used to describe the level of randomness of the axon packing results, i.e., $$C(k)=\iint\,e^{-i\mathbf{kr}}\rho(\mathbf{r}+\mathbf{r}_0)\rho(\mathbf{r}_0)d\mathbf{r}\frac{d\mathbf{r}_0}{W^{2}},$$ where $$$\rho(\mathbf{r})$$$ is the density of axons at point $$$\mathbf{r}$$$. $$$\rho(\mathbf{r})$$$ is 1 for points inside axons and 0 for outside. $$$\phi v_0$$$ and $$$\langle r_{ext}\rangle=\sqrt{v_0/\pi}$$$ (external object radius) are used for the SCF and $$$k$$$-axis normalization respectively.10,11

RESULTS

Fig.4 shows the outcome of both techniques and the lattice packing. The related estimated radii PDF is presented in Fig.3(b), showing a bias in the estimated mean radius. The analysis of estimated mean radius and variance, depicted in Fig.5(a-b), illustrates that: (i) the conventional technique underestimates the model parameters and (ii) the uncertainty of the estimations is smaller for the proposed technique. The bias in the mean radius and variance for the conventional method is due to limited free space for new axons and rejection of larger axons as the procedure evolves.

The analysis of simulated packing densities presented in Fig.5(c) demonstrates that for a given separation parameter, the proposed method results in higher packing densities. However, the analysis of SCF, Fig.5(d), shows that the plateau in $$$C(k)$$$ as $$$k\to0$$$ for the proposed method is smaller than for the conventional (indicating stronger spatial correlation), but the method still preserves short-range disorder property (distinctive from completely ordered packing, lattice in Fig.4(c)).

DISCUSSION AND CONCLUSIONS

A novel approach was proposed for simulating the placement of axons in nerve bundles. Comparison of the results to the conventional method demonstrated that our method (i) eliminates the bias in estimated radius mean and variance (ii) results in tissue models which are highly consistent with the desired characteristics of nerve bundles, (iii) can result in higher packing densities for a given set of model parameters, and (iv) while is more correlated at short scales, still preserves the short-range disorder property of the tissue. However, the exact degree of randomness in WM tissue and how nature packs the axons remains an open question.

The resultant axon packings can be used subsequently to study the Brownian motion of water molecules within nerve bundles and the effect of axon configuration properties on the measured diffusion MR signal.

Acknowledgements

This research is supported by VIDI Grant 639.072.411 from the Netherlands Organisation for Scientific Research (NWO).

References

1. Fieremans E, De Deene Y, Delputte S, Ozdemir M S, Achten E, and Lemahieu I, “The design of anisotropic diffusion phantoms for the validation of diffusion weighted magnetic resonance imaging.,” Phys. Med. Biol., vol. 53, no. 19, pp. 5405–19, Oct. 2008.

2. Fieremans E, De Deene Y, Delputte S, Ozdemir M S, D’Asseler Y, Vlassenbroeck J, Deblaere K, Achten E, and Lemahieu I, “Simulation and experimental verification of the diffusion in an anisotropic fiber phantom.,” J. Magn. Reson., vol. 190, no. 2, pp. 189–99, Feb. 2008.

3. Fieremans E, De Deene Y, and Lemahieu I, “Validation of models for the diffusion weighted MR signal in brain white matter,” in 2008 5th IEEE International Symposium on Biomedical Imaging: From Nano to Macro, 2008, pp. 915–918.

4. Fieremans E, Novikov D S, Jensen J H, and Helpern J A, “Monte Carlo study of a two-compartment exchange model of diffusion.,” NMR Biomed., vol. 23, no. 7, pp. 711–24, Aug. 2010.

5. Novikov D S, Fieremans E, Jensen J H, and Helpern J A, “Random walk with barriers.,” Nat. Phys., vol. 7, no.6, pp. 508–514, Jun. 2011.

6. Donev A, Stillinger F H, and Torquato S, “Unexpected density fluctuations in jammed disordered sphere packings.,” Phys. Rev. Lett., vol. 95, no. 9, p. 090604, Aug. 2005.

7. Donev A, Connelly R, Stillinger F H, and Torquato S, “Underconstrained jammed packings of nonspherical hard particles: Ellipses and ellipsoids,” Phys. Rev. E, vol. 75, no. 5, p. 051304, May 2007.

8. Assaf Y, Blumenfeld-Katzir T, Yovel Y, and Basser P J, “AxCaliber: a method for measuring axon diameter distribution from diffusion MRI.,” Magn. Reson. Med., vol. 59, no. 6, pp. 1347–1354, Jun. 2008.

9. Aboitiz F, Scheibel A B, Fisher R S, and Zaidel E, “Fiber composition of the human corpus callosum.,” Brain Res., vol. 598, no. 1–2, pp. 143–53, Dec. 1992.

10. Novikov D S, Jensen J H, Helpern J A, and Fieremans E, “Revealing mesoscopic structural universality with diffusion.,” Proc. Natl. Acad. Sci. U. S. A., vol. 111, no. 14, pp. 5088–93, Apr. 2014.

11. Burcaw L M, Fieremans E, and Novikov D S, “Mesoscopic structure of neuronal tracts from time-dependent diffusion.,” Neuroimage, vol. 114, pp. 18–37, Mar. 2015.

Figures

Fig.1: (a) Axons are represented by circles. Each circle has a boundary (with radius ri+hi) which is shown with dotted circles. (b) The boundaries of the circles are in contact with each other but do not overlap. (c) Placement of new circles with respect to the previously placed adjacent circles. The boundary of the new circle (orange) interfaces the boundaries of the previously placed neighboring circles (green and blue).

Fig.2: Flowchart of the proposed algorithm.

Fig.3: (a) Range of model parameters used for Monte-Carlo simulations. (b) Probability density function of the resulting axon radii for both methods with μr = 3 μm, σr2 = 1 μm2 and η = 0. The distribution of the radii using the conventional technique underestimates the mean axon radius while the proposed method fits the desired distribution.

Fig.4: The results of the axon packing algorithms (a) Proposed method (b) Conventional method (c) Lattice. The conventional method produces smaller axons as the process evolves..

Fig.5: Statistical analysis of the simulated axon packings. (a) Analysis of mean radius. (b) Analysis of radius variance. (c) Analysis of packing density. (d) Analysis of normalized structure correlation function.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
3084