Beyond the Tensor Model
Evren Ozarslan1

1Bogazici University

Synopsis

Diffusion tensor imaging (DTI) is widely employed to characterize diffusion anisotropy in multi-directional diffusion MR acquisitions. However, the DTI model has well-known limitations primarily because it assumes diffusion to be Gaussian. In this talk, DTI’s limitations will be discussed for three cases: (i) the presence of orientational complexity, (ii) nonlinearity of the signal decay curves, and (iii) dependence on the timing parameters of the sequence. Several alternative approaches will be outlined and it will be argued that a cost-benefit analysis has to be performed before abandoning the diffusion tensor model.

Target Audience

Scientists with elementary physics background interested in the main approaches taken to model and represent diffusion-weighted MR data.

Outcome / Objectives

· The goal is to introduce models that go beyond diffusion tensor imaging.

· The need for more sophisticated models will be justified.

· The audience will find out about the different directions in which the diffusion tensor model is generalized.

Purpose

As the first analytical model used to describe the anisotropic diffusional processes that can be observed via magnetic resonance methods, diffusion tensor imaging (DTI) [1,2] has attracted widespread interest from many different fields of investigation. Several new MRI indices, akin to different “stains” in histological examinations, have been introduced that exploits the diffusional characteristics of the medium based on the DTI model [3,4]. Perhaps equally importantly, DTI has made it possible to map the major neuronal connections between different regions of the central nervous system. DTI features a relatively simple mathematical structure, and is practical also from the point of view of data collection. However, there are some shortcomings of DTI, stemming from its assumption that diffusion is Gaussian. This assumption may have serious ramifications related to fiber-tract mapping and microstructure elucidation. Thus, more general representations applicable to a wider range of diffusional scenarios are commonly employed.

Methods

Models that go beyond DTI can be grouped into different classes. DTI’s inability to describe diffusion MRI data becomes most apparent when one considers variations in different parameters of a pulse sequence. ·

When the dependence of the signal on the gradient orientation is concerned, DTI assumes a unimodal profile on the hemisphere. Such a profile is sufficient when there is a single underlying orientation. However, since there are as many as 100,000 neurons within a voxel, it is not unusual to encounter orientationally heterogeneous structures. In such complicated regions, the principal eigenvector of the estimated diffusion tensor yields only an average direction, which typically does not coincide with any of the distinct fiber pathways. As such, DTI may lead to erroneous connections and pathways in anatomical connectivity studies. To overcome this limitation, numerous methods have been proposed to date. For example, by increasing the angular resolution of the diffusion measurements [5], one can infer the distinct orientations along which diffusion is favored [6-12].

For the case of traditional diffusion measurements performed by applying a pair of gradients [13], the sensitivity of the signal to diffusion is determined partially by a quantity called the q-value, which has the dimension of inverse length. When the logarithm of the signal is plotted against this quantity, one obtains a linear dependence for the case of Gaussian diffusion. However, real experiments indicate a deviation from this behavior [14]; this deviation becomes substantial as the q-value is increased. Therefore, when data with large q-values are to be used, the DTI model becomes insufficient. Thus, more general representations are commonly employed [15-16]. Note that this deviation is not limited to one-dimensional sampling of the q-space. For characterizing diffusion anisotropy, one typically samples the three-dimensional q-space. In this case, one could either use a “model-free” approach to transform the entire data set directly into a meaningful quantity called the average propagator [17] or employ a generalization of DTI’s signal expression [18-19] to represent the data.

Another, often overlooked limitation of DTI is apparent when one considers the dependence of the MR signal on the timing parameters of the sequence. Gaussian diffusion suggests a particular dependence, expressed by the Stejskal-Tanner expression [13], which is quite different than what one obtains for the case of restricted diffusion [20,21]. The difference is quite significant even at small q-values. Thus, referring to the small diffusion sensitivity regime as the “DTI regime” may be problematic.

Discussion & Conclusion

Depending on the acquisition scheme and the goal of the study, different models tailored to that problem could be favored if the diffusion tensor model fails. We note that some of the shortcomings of DTI could be overcome if the tensor model is employed as the building block of a composite model. For example, the signal can be expressed as the sum of two expressions, each having a separate diffusion tensor [22]. Such representation produces the nonlinearity of the semi-logarithmic signal vs. q-value plot, and could be employed to resolve two fiber orientations. Similarly, the signal from the voxel can be envisioned to emerge from a Wishart distribution of diffusion tensors [23], which yields, remarkably, a power-law decay in the tail of the signal decay curve.

Last but not least, it should be mentioned that the more sophisticated models could be costly in terms of their computational burden and typically require more time-consuming data acquisitions. In certain cases, it may be necessary to compromise on the spatial resolution in acquisitions, which could exacerbate the problem of fiber crossings. Thus, when the cost of employing a more general model outweighs its benefits, DTI may still be the method of choice.

Acknowledgements

No acknowledgement found.

References

1. P.J. Basser, J. Mattiello, D. LeBihan, Estimation of the effective self-diffusion tensor from the NMR spin echo, J. Magn. Reson., B 103 (3) (1994) 247– 254.

2. P.J. Basser, J. Mattiello, D. LeBihan, MR diffusion tensor spectroscopy and imaging, Biophys. J. 66 (1), (1994) 259–267.

3. P.J. Basser, Inferring microstructural features and the physiological state of tissue from diffusion weighted images, NMR Biomed. 8 (1995) 333-344.

4. C. Pierpaoli, P.J. Basser, Toward a quantitative assessment of diffusion anisotropy. Magn. Reson. Med. 36 (1996) 893-906.

5. D.S. Tuch, T.G. Reese, M.R. Wiegell, N. Makris, J.W. Belliveau, V.J. Wedeen, High angular resolution diffusion imaging reveals intravoxel white matter fiber heterogeneity, Magn. Reson. Med. 48 (4) (2002) 577–582.

6. D.S. Tuch, Q-ball imaging, Magn. Reson. Med. 52 (2004) 1358–1372.

7. E. Özarslan, T.H. Mareci, Generalized diffusion tensor imaging and analytical relationships between diffusion tensor imaging and high angular resolution diffusion imaging, Magn. Reson. Med. 50 (5) (2003), 955–965.

8. J.D. Tournier, F. Calamante, D.G. Gadian, A. Connelly, Direct estimation of the fiber orientation density function from diffusion-weighted MRI data using spherical deconvolution, Neuroimage 23, (2004), 1176–1185.

9. E. Özarslan, T.M. Shepherd, B.C. Vemuri, S.J. Blackband, T.H. Mareci, Resolution of complex tissue microarchitecture using the diffusion orientation transform (DOT), Neuroimage 31 (3) (2006) 1086–1103.

10. M. Descoteaux, E. Angelino, S. Fitzgibbons, R. Deriche, Regularized, fast, and robust analytical q-ball imaging, Magn. Reson. Med. 58 (3) (2007) 497–510.

11. I. Aganj, C. Lenglet, G. Sapiro, E. Yacoub, K. Ugurbil, N. Harel, Reconstruction of the orientation distribution function in single- and multiple-shell q-ball imaging within constant solid angle, Magn. Reson. Med. 64 (2) (2010) 554–566.

12. F. Dell'Acqua, P. Scifo, G. Rizzo, M. Catani, A. Simmons, G. Scotti, F. Fazio, A modified damped Richardson–Lucy algorithm to reduce isotropic background effects in spherical deconvolution, Neuroimage 49 (2) (2010) 1446–1458.

13. E.O. Stejskal, J.E. Tanner, Spin diffusion measurements: Spin echoes in the presence of a time-dependent field gradient, J. Chem. Phys. 42 (1) (1965) 288–292.

14. R.V. Mulkern, H. Gudbjartsson, C.-F. Westin, H.P. Zengingönül, W. Gartner, C.R. Guttmann, R.L. Robertson, W. Kyriakos, R. Schwartz, D. Holtzman, F.A. Jolesz, S.E. Maier, Multi-component apparent diffusion coefficients in human brain, NMR Biomed. 12 (1) (1999) 51–62.

15. J.H. Jensen, J.A. Helpern, A. Ramani, H. Lu, K. Kaczynski, Diffusional kurtosis imaging: the quantification of non-gaussian water diffusion by means of magnetic resonance imaging, Magn. Reson. Med. 53 (2005) 1432–1440.

16. E. Özarslan, C.G. Koay, P.J. Basser, Simple harmonic oscillator based estimation and reconstruction for one-dimensional q-space MR, in Proc. Intl. Soc. Mag. Reson. Med. 16 (2008) p. 35.

17. V.J. Wedeen, P. Hagmann, W.-Y.I. Tseng, T.G. Reese, R.M. Weisskoff, Mapping complex tissue architecture with diffusion spectrum magnetic resonance imaging, Magn. Reson. Med. 54 (6) (2005) 1377–1386.

18. C.L. Liu, R. Bammer, M.E. Moseley, Generalized diffusion tensor imaging (GDTI): a method for characterizing and imaging diffusion anisotropy caused by non-Gaussian diffusion, Isr. J. Chem. 43 (1–2) (2003) 145–154.

19. E. Özarslan, C.G. Koay, T.M. Shepherd, M.E. Komlosh, M.O. Irfanoglu, C. Pierpaoli, P.J. Basser, Mean apparent propagator (MAP) MRI: a novel diffusion imaging method for mapping tissue microstructure, NeuroImage 78 (2013) 16–32.

20. E. Özarslan, P.J. Basser, Microscopic anisotropy revealed by NMR double pulsed field gradient experiments with arbitrary timing parameters, J. Chem. Phys. 128 (15) (2008) 154511.

21. E. Özarslan, C.-F. Westin, T.H. Mareci, Characterizing magnetic resonance signal decay due to Gaussian diffusion: the path integral approach and a convenient computational method, Concept. Magn. Reson. A, 44(A) (2015) 203–213.

22. B.A. Inglis, E.L. Bossart, D.L. Buckley, E.D. Wirth, T.H. Mareci, Visualization of neural tissue water compartments using biexponential diffusion tensor MRI, Magn. Reson. Med. 45 (4) (2001) 580–587.

23. B. Jian, B.C. Vemuri, E. Özarslan, P.R. Carney, T.H. Mareci, A novel tensor distribution model for the diffusion-weighted MR signal, Neuroimage 37 (1) (2007) 164–176.



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