Section: New Results

Image processing on Diffusion Weighted Magnetic Resonance Imaging

Diffusion Directions Imaging (DDI)

Participants : Aymeric Stamm, Christian Barillot.

Diffusion magnetic resonance imaging (dMRI) is the reference in vivo modality to study the connectivity of the brain white matter. Images obtained through dMRI are indeed related to the probability density function (pdf) of displacement of water molecules subject to restricted diffusion in the brain white matter. The knowledge of this diffusion pdf is therefore of primary importance. Several methods have been devised to provide an estimate of it from noisy dMRI signal intensities. They include popular diffusion tensor imaging (DTI) as well as higher-order methods. These approaches suffer from important drawbacks. Standard DTI cannot directly cope with multiple fiber orientations. Higher-order approaches can alleviate these limitations but at the cost of increased acquisition time. We have proposed, in the same vein as DTI, a new parametric model of the diffusion pdf with a reasonably low number of parameters, the estimation of which does not require acquisitions longer than those used in clinics for DTI. This model also accounts for multiple fiber orientations. It is based on the assumption that, in a voxel, diffusing water molecules are divided into compartments. Each compartment is representative of a specific fiber orientation (which defines two opposite directions). In a given compartment, we further assume that water molecules that diffuse along each direction are in equal proportions. We then focus on modeling the pdf of the displacements of water molecules that diffuse only along one of the two directions. Under this model, we derive an analytical relation between the dMRI signal intensities and the parameters of the diffusion pdf. We exploit it to estimate these parameters from noisy signal intensities. We carry out a cone-of-uncertainty analysis to evaluate the accuracy of the estimation of the fiber orientations and we evaluate the angular resolution of our method. Finally, we show promising results on real data and propose a visualization of the diffusion parameters which is very informative to the neurologist. This work was conducted in collaboration with Patrick Perez from Technicolor [56] .

Anatomy of the corticospinal tracts: evaluation of a deterministic tractography method

Participants : Romuald Seizeur, Nicolas Wiest-Daesslé, Sylvain Prima, Camille Maumet, Jean-Christophe Ferré, Xavier Morandi.

In this work, anatomical, diffusion-weighted and functional 3T MRI were acquired on 15 right-handed healthy subjects to analyse the portions of the corticospinal tract (CST) dedicated to hand motor and sensory functions. The three MR images were then registered and regions of interest were delineated i) in the mid-brain using 3D T1-weighted MRI, and ii) in the cortex using fMRI using hand motor and sensory tasks. Deterministic tractography was then performed using these two ROIs from diffusion-weighted MRI after the diffusion tensors were computed. The ventrolateral tract fibers of the CST were generally not properly identified, due to fiber crossing in the corona radiata [55] .

Tracking of the Hand Motor Fibers within the Corticospinal Tract Using Functional, Anatomical and Diffusion MRI

Participants : Romuald Seizeur, Nicolas Wiest-Daesslé, Olivier Commowick, Sylvain Prima, Aymeric Stamm, Christian Barillot.

In this work, we proposed to compare three diffusion models to track the portion of the corticospinal tract dedicated to the hand motor function (called hand motor fibers hereafter), using diffusion, functional and anatomical MRI. The clinical diffusion data have few gradient directions and low b-values. In this context, we show that a newly introduced model, called diffusion directions imaging (DDI) outperforms both the DTI and the ODF models. This new model allows to capture several diffusion directions within a voxel, with only a low number of parameters. Two important results are that i) the DDI model is the only one allowing consistent tracking from the mesencephalon to the most lateral part of the cortical motor hand area, and that ii) the DDI model is the only model able to show that the number of hand motor fibers in the left hemisphere is larger than in the contralateral hemisphere for right-handed subjects; the DDI model, as the other two models, fails to find such a difference for left-handed subjects. To the best of our knowledge, this is the first time such results are reported, at least on clinical data. [44] .

Multifiber Deterministic Streamline Tractography Based on a New Diffusion Model

Participants : Olivier Commowick, Romuald Seizeur, Nicolas Wiest-Daesslé, Sylvain Prima, Aymeric Stamm, Christian Barillot.

In this work, we have built upon a new model, describing the random motion of water molecules in fibrous tissues, to develop a multifiber deterministic tractography algorithm. We apply this algorithm to track the corticospinal tract of the human brain, in both controls and patients with tumors. [31] .

Automated detection of white matter fiber bundles

Participant : Olivier Commowick.

This work is part of a collaboration with the Computational Radiology Laboratoy headed by Simon Warfield in Boston, USA. For this topic, we have studied how white matter fiber bundles can be extracted in a reproducible way from diffusion tensor MRI. Usually, white matter (WM) fiber bundles of the brain can be delineated by diffusion tractography utilizing anatomical regions-of-interest (ROI). These ROIs can specify seed regions in which tract generation algorithms are initiated. Interactive identification of such anatomical ROIs enables the detection of the major WM fiber tracts, but suffers from inter-rater and intra-rater variability, and is time consuming. We developed and compared three techniques for automated delineation of ROIs for the detection of two major WM fiber tracts in 12 healthy subjects. Tracts identified automatically were compared quantitatively to reference standard tracts derived from carefully hand-drawn ROIs. Based on comparative performance of the experimental techniques, a multi-template label fusion algorithm was found to generate tracts most consistent with the reference standard. More details on this work are available in [43] .