Showing papers by "Bertrand Thirion published in 2008"

Connectivity-based parcellation of the cortical mantle using q-ball diffusion imaging

Abstract: This paper exploits the idea that each individual brain region has a specific connection profile to create parcellations of the cortical mantle using MR diffusion imaging. The parcellation is performed in two steps. First, the cortical mantle is split at a macroscopic level into 36 large gyri using a sulcus recognition system. Then, for each voxel of the cortex, a connection profile is computed using a probabilistic tractography framework. The tractography is performed from q-ball fields using regularized particle trajectories. Fiber ODF are inferred from the q-balls using a sharpening process focusing the weight around the q-ball local maxima. A sophisticated mask of propagation computed from a T1-weighted image perfectly aligned with the diffusion data prevents the particles from crossing the cortical folds. During propagation, the particles father child particles in order to improve the sampling of the long fascicles. For each voxel, intersection of the particle trajectories with the gyri lead to a connectivity profile made up of only 36 connection strengths. These profiles are clustered on a gyrus by gyrus basis using a K-means approach including spatial regularization. The reproducibility of the results is studied for three subjects using spatial normalization.

Mutual information-based feature selection enhances fMRI brain activity classification

Abstract: In this paper, we address the question of decoding cognitive information from functional magnetic resonance (MR) images using classification techniques. The main bottleneck for accurate prediction is the selection of informative features (voxels). We develop a multivariate approach based on a mutual information criterion, estimated by nearest neighbors. This method can handle a large number of dimensions and is able to detect the non-linear correlations between the features and the label. We show that, by using Mi-based feature selection, we can achieve better performance together with sparse feature selection, and thus a better understanding of information coding within the brain than the reference method which is a mass univariate selection (ANOVA).

Probabilistic Anatomo-Functional Parcellation of the Cortex: How Many Regions?

Abstract: Understanding brain structure and function entails the inclusion of anatomical and functional information in a common space, in order to study how these different informations relate to each other in a population of subjects. In this paper, we revisit the parcellation model and explicitly combine anatomical features, i.e. a segmentation of the cortex into gyri, with a functional information under the form of several cortical maps, which are used to further subdivide the gyri into functionally consistent regions. A probabilistic model is introduced, and the parcellation model is estimated using a Variational Bayes approach. The number of regions in the model is validated based on cross-validation. It is found that about 250 patches of cortex can be delineated both in the left and right hemisphere based on this procedure.

Sensitivity analysis of parcellation in the joint detection-estimation of brain activity in fMRI

Abstract: Within-subject analysis in fMRI relies on both (i) a detection step to localize which parts of the brain are activated by a given stimulus type, and on (ii) an estimation step to recover the underlying brain dynamics. In [1], a Bayesian detection- estimation approach that jointly addresses (i)-(ii) has been proposed. In the latter, a functionally homogeneous parcel- lation of the brain is required prior to this analysis. If tools exist to produce suitable parcellations [2], the question remains open of its impact on both activation detection and dynamics estimation. Here, we present a sensitivity analysis of this Bayesian model regarding the parcellation. We show that some activating clusters are stable regarding parcellation while others are highly variable. The overall procedure is quite sensitive to the input parcellation as the uncertainty of the estimated effect is correlated to its size. The perspective is to extend our model with an adaptive parcellation combined with the detection-estimation.

DEALING WITH SPATIAL NORMALIZATION ERRORS IN fMRI GROUP INFERENCE USING HIERARCHICAL MODELING

Abstract: An important challenge in neuroimaging multi-subject studies is to take into account that different brains cannot be aligned perfectly. To this end, we extend the classical mass univariate model for group analysis to incorporate uncer- tainty on localization by introducing, for each subject, a spatial "jitter" variable to be marginalized out. We derive a Bayes factor to test for the mean popula- tion effect's sign in each voxel of a search volume, and discuss a Gibbs sampler to compute it. This Bayes factor, which generalizes the classical t-statistic, may be combined with a permutation test in order to control the frequentist false positive rate. Results on both simulated and experimental data suggest that this test may outperform conventional mass univariate tests in terms of detection power, while limiting the problem of overestimating the size of activity clusters.

Defining cortical sulcus patterns using partial clustering based on bootstrap and bagging

Abstract: The cortical folding patterns are very different from one individual to another. Here we try to find folding patterns automatically using large-scale datasets by non-supervised clustering analysis. The sulci of each brain are detected and identified using the brain VIS A open software. The 3D moment invariants are calculated and used as the shape descriptors of the sulci identified. A partial clustering algorithm using bootstrap sampling and bagging (PCBB) is devised for cortical pattern mining. Partial clusters are found using a modified hierarchical clustering method constrained by an objective function which looks for the most compact and dissimilar clusters. Bagging is used to increase stability. Experiments on simulated and real datasets are used to demonstrate the strength and stability of this algorithm compared to other standard approaches. Some cortical patterns are found using our method. In particular, the patterns found for the left cingulate sulcus are consistent with the patterns described in the atlas of Ono.

Triangulating cortical functional networks with anatomical landmarks

Abstract: Defining precisely the position of active regions obtained from functional neuroimaging studies is challenging due to the functional and anatomical variability across subjects. Traditional volumetric normalization techniques ignore the geometry of the cortex and use a relatively imprecise three-dimensional coordinate system. In this study we propose an alternative method that relates the position of functional regions on the cortical surface to the positions of the main macro-anatomical structures, the sulci. Our approach consists of using the nearest sulci to build a local referential in which the position of a region is defined. This triangulation approach improves the localization of brain regions involved in various cognitive tasks.