Event Abstract

CSF partial volume modeling in diffusion kurtosis imaging: a comparative parameter estimation study

  • 1 University of Antwerp, Belgium
  • 2 New York University Langone Medical Center, United States
  • 3 Delft University of Technology, Netherlands

Diffusion kurtosis imaging (DKI) model [1] parameters and derived metrics are known to be more sensitive to brain physiological changes compared to diffusion tensor imaging (DTI) parameters [2]. However, in diffusion MR imaging, many voxels suffer from partial volume effects with cerebrospinal fluid (CSF). Ignoring these effects can lead to large biases in the estimation of diffusion properties of brain tissue [3]. For DTI, much research has already been done on extending the model to a bi-exponential model that incorporates a free water signal fraction. Although results are promising, the DTI + CSF model is ill-conditioned, making it necessary to use more complex parameter estimation methods that include some sort of prior information [4-7]. In this work, the DKI model is extended to include a CSF fraction. We propose to use a Bayesian approach to estimate the model parameters. This approach includes a Gaussian shrinkage prior on the model parameters, a Rician distributed data model and a Markov chain Monte Carlo approach to compute the posterior estimates [8]. The prior distributions are completely determined by the data, omitting the need for any user-defined parameters. Simulations show that the Bayesian shrinkage prior (BSP) estimator achieves a higher precision and accuracy compared to both a constrained non-linear least squares (NLS) and a constrained maximum likelihood (MLE) estimation. Because of the ill-conditioned model, NLS and MLE estimates will sometimes be close to the (user-defined) bounds. This is not the case for the BSP estimates. These findings are also confirmed in a real data experiment (see Figure) where the FA map from the BSP estimation look much more natural than those from the NLS and MLE estimations.

Figure 1

References

1. Jensen, J. H., Helpern, J. a, Ramani, A., Lu, H., & Kaczynski, K. (2005). Diffusional kurtosis imaging: the quantification of non-gaussian water diffusion by means of magnetic resonance imaging. Magnetic Resonance in Medicine, 53(6), 1432–40;
2. Cheung, M. M., Hui, E. S., Chan, K. C., Helpern, J. A., Qi, L. & Wu, E. X. (2009). Does diffusion kurtosis imaging lead to better neural tissue characterization? A rodent brain maturation study. NeuroImage, 45(2), 386-92;
3. Metzler-Baddeley, C., O’Sullivan, M. J., Bells, S., Pasternak, O., & Jones, D. K. (2012). How and how not to correct for CSF-contamination in diffusion MRI. NeuroImage, 59(2), 1394–403;
4. Pierpaoli, C., & Jones, D. K. (2004). Removing CSF Contamination in Brain DT-MRIs by Using a Two-Compartment Tensor Model. In Proc. Intl. Soc. Mag. Reson. Med. (Vol. c, p. 1215);
5. Pasternak, O., Sochen, N., Gur, Y., Intrator, N., & Assaf, Y. (2009). Free water elimination and mapping from diffusion MRI. Magnetic Resonance in Medicine, 62(3), 717–30;
6. Collier, Q., Veraart, J., Jeurissen, B., den Dekker, A. J., & Sijbers, J. (2015). Theoretical study of the free water elimination model. In Proc. Intl. Soc. Mag. Reson. Med. (p. 2757);
7. Vallée, E., Douaud, G., Monsch, A. U., Gass, A., Wu, W., Smith, S. & Jbabdi, S. (2015). Modelling free water in diffusion MRI. In Proc. Intl. Soc. Mag. Reson. Med. (p. 0474);
8. Orton, M. R., Collins, D. J., Koh, D.-M., & Leach, M. O. (2014). Improved intravoxel incoherent motion analysis of diffusion weighted imaging by data driven Bayesian modeling. Magnetic Resonance in Medicine, 71(1), 411–20;

Keywords: parameter estimation, DKI, kurtosis, partial volume effects, Cerebrospinal Fluid, Bayesian Models, Prior Knowledge

Conference: Second Belgian Neuroinformatics Congress, Leuven, Belgium, 4 Dec - 4 Dec, 2015.

Presentation Type: Poster Presentation

Topic: Methods and Modeling

Citation: Collier Q, Veraart J, Den Dekker AJ, Jeurissen B and Sijbers J (2015). CSF partial volume modeling in diffusion kurtosis imaging: a comparative parameter estimation study. Front. Neuroinform. Conference Abstract: Second Belgian Neuroinformatics Congress. doi: 10.3389/conf.fninf.2015.19.00039

Received: 13 Nov 2015; Published Online: 17 Nov 2015.

* Correspondence: Mr. Quinten Collier, University of Antwerp, Antwerp, Belgium, quinten.collier@uantwerpen.be

© 2007 - 2017 Frontiers Media S.A. All Rights Reserved