Robust fusion of Jacobian maps for Deformation-Based Morphometry

Nicolas Guizard (1) , Pierrick Coupé (1) , Vladimir Fonov (1) , Douglas Arnold (1) , Louis D. Collins (1)

1 MNI - McConnell Brain Imaging Centre

In population studies with Magnetic Resonance Imaging (MRI), Deformation-Based Morphometry (DBM) [Ashburner, 1998] has received more and more attention as a tool to investigate and identify anatomical differences between groups in cross-sectional studies or anatomical changes over time in longitudinal studies. In DBM methods, statistical analysis of the parameters required to normalize all subjects into a common space enables the localization of even subtle brain shape changes. However, DBM is highly dependant on the non-linear registration accuracy used for normalization. Registration errors due to anatomical singularities such as non-homologous gyri, the presence of lesions or image artifacts may result in outliers in the analysis. While smoothing may reduce the resulting adverse effects of these errors, it may also lead to over-smoothing of the results and reduce localization power. Therefore, we propose a method to improve the robustness of the DBM method by using a robust patch-based aggregation of Jacobian maps in order to preserve fine local brain changes. The method is evaluated using MRI data from 20 patients with multiple sclerosis.

Domaines: Imagerie médicale

Nicolas Guizard, Pierrick Coupé, Vladimir Fonov, Douglas Arnold, Louis D. Collins. Robust fusion of Jacobian maps for Deformation-Based Morphometry. Organization for Human Brain Mapping 2011 Annual Meeting, Jun 2011, Canada. ⟨hal-00645486⟩