Séminaire du 10/04/2014

jeudi 10 avril 2014, 14h00, C218

General linear models for group studies in diffusion tensor imaging

Conférencier : Alix Bouchon

Résumé : The degenerative diseases of the human central nervous system are an important clinical issue.Neurodegenerative diseases affect the cerebral white matter integrity, which may now be probed in vivo using diffusion tensor imaging (DT-MRI) [1]. Characterizing those pathologies and its evolution provides new information for the diagnosis and the prognosis of doctors. It can be done by comparing a group of healthy subjects vs cohorts of patients, according to neuroimaging data, clinical data and cognitive scores. It usually performed on scalar images derived from DT-MRI such as Fractional Anisotropy (FA) or Mean Diffusivity (MD) using either region of interest-based analysis, the voxel-based analysis framework provided in SPM [3] or the tract-based spatial statistics (TBSS) method [4] provided in the FSL library. However, none of these methods can handle all the information contained in the tensor images and thus cannot detect all kind of changes, such as modification of the diffusion orientation. Some methods proposed to perform the comparison using multivariate statistical tests on several scalar indices simultaneously or to compare eigenvalues or eigenvectors of diffusion tensors [5]. However, these approaches cannot include clinical data in the statistical analysis. To the best of the author’s knowledge, only the work of Zhu et al. [6] has carried out multivariate regression analysis and statistical testing on diffusion tensors. The framework proposed in [6] is versatile and accounts for the positive definiteness of diffusion tensor by estimating a generalized linear model on a Riemannian manifold. Unfortunately, the regression step suffers from being computationally intensive.

In this context, we extended the general linear model, as implemented in SPM [3], to tensor images. This model has the advantage of being both versatile and computationally effective. We investigated the relevance of the whole tensor information compared to scalar indices only by a simulation framework based on DT-MRI acquisitions of healthy subjects in which different kinds of lesions have been introduced. Results on a cohort of patients suffering from neuromyelitis optica (NMO) are also analysed [7].

• [1] C. Yu, F.Lin, K. Li, T. Jiang, W. Qin, H. Sun and P. Chan, « Pathologenesis of normal-appearing white matter damage in neuromyelitis optica : Diffusion-Tensor MR Imaging », Radiology, vol. 246, no. 1, pp. 222-228, 2008.
• [2] M. Cercignani, Strategies for Patient-Control Comparison of Diffusion MR Data, Oxford University Press, 2010.
• [3] W.D Penny, K.J Friston, J.T Ashburner, S.J Kirbel and T.E Nichols, Statistical Parametric Mapping : The Analysis of Functional Brain Images, Elsevier LTD, Oxford, 2006.
• [4] S.M Smith, M. Jenkinson, H. Johansen-Berg, D. Rueckert, T.E Nichols, C.E Mackay, K.E Watkins, O. Ciccarelli, M.Z Cader, P.M Matthews, T.E.J Behrens, « Tract based spatial satistics : voxelwise analysis of multi-subjects diffusion data », NeuroImage, vol. 31, no. 4, pp. 1487-1505, 2006.
• [5] A. Schwartzman, R.F Dougherty, J.E Taylor, « Group comparison of eigenvalues and eigenvectors of diffusion tensors », Journal of the American Statistical Association, vol. 105, no. 490, pp. 588-599, 2010.
• [6] H. Zhu, Y. Chen, J.G Ibrahim, Y. Li, C. Hall, W. Lin, « Intrinsic regression models for positive-definite matrices with applications to Diffusion Tensor Imaging », Journal of the American Statistical Association, vol. 104, no. 487, pp. 1203-1212, 2009.
• [7] A. Bouchon, V. Noblet, F. Heitz, J. Lamy, F. Blanc, J.P Armspach, « General linear models for group studies in diffusion tensor imaging », ISBI 2013 in press.