EN FR
EN FR


Section: New Results

Image Segmentation, Registration and Analysis

Quantitative Analysis of Open Curves in Brain Imaging: Applications to White Matter Fibers and Sulci

Participants : Meena Mani, Christian Barillot.

Shape, scale, orientation and position, the four physical features associated with open curves, have different properties so the usual approach has been to design different metrics and spaces to treat them individually. We took an alternative approach using a comprehensive Riemannian framework where joint feature spaces allow for analysis of combinations of features. We can compare curves by using geodesic distances, which quantify their differences. We validated the metrics we used, demonstrated practical uses and applied the tools to important clinical problems. To begin, specific tract configurations in the corpus callosum are used to showcase clustering results that depend on the Riemannian distance metric used. This nicely argues for the judicious selection of metrics in various applications, a central premise in our work. The framework also provides tools for computing statistical summaries of curves. We represented fiber bundles with a mean and variance, which describes their essential characteristics. This is both a convenient way to work with a large volume of fibers and is a first step towards statistical analysis. Next, we designed and implemented methods to detect morphological changes, which can potentially track progressive white matter disease. With sulci, we addressed the specific problem of labeling. An evaluation of physical features and methods such as clustering leads us to a pattern matching solution in which the sulcal configuration itself is the best feature.

Trimmed-likelihood estimation for focal lesions and tissue segmentation in multisequence MRI for multiple sclerosis

Participants : Sylvain Prima, Christian Barillot.

Following Daniel Garcia-Lorenzo's PhD, we proposed a new automatic method for segmentation of multiple sclerosis (MS) lesions in magnetic resonance images. The method performs tissue classification using a model of intensities of the normal appearing brain tissues. In order to estimate the model, a trimmed likelihood estimator is initialized with a hierarchical random approach in order to be robust to MS lesions and other outliers present in real images. The algorithm was first evaluated with simulated images to assess the importance of the robust estimator in presence of outliers. The method was then validated using clinical data in which MS lesions were delineated manually by several experts. Our method obtains an average Dice similarity coefficient (DSC) of 0.65, which is close to the average DSC obtained by raters (0.66) [15] .

Segmentation of Multimodal Brain Images using Spectral Gradient and Graph Cut

Participants : Camille Maumet, Jean-Christophe Ferré, Christian Barillot.

Following Jeremy Lecoeur's PhD, we have introduced a new and original scale-space approach for segmenting normal and pathological tissue from multidimensional images. This method can perform a joint segmentation of three complementary imaging volumes at the same time by embedding a scale-space color invariant edge detector - i.e. the spectral gradient - as the boundary term in a graph cut optimization framework. Finally, we have proposed to extend this new scheme to more than three channels. We focussed the contribution onto the segmentation of tissues or structures of interest from multi-dimensional / multi-sequences brain MRI. This new multidimensional segmentation framework has been validated on simulated data and on clinical data (both pathological and healthy brains). We have exhibited the performances of this new method on various combinations of MRI sequences for the segmentation of normal and pathological tissues and showed how it is able to out perform competitive works. This work is under submission to an international journal.

Adaptive pixon represented segmentation for 3D MR brain images based on mean shift and Markov random fields

Participant : Christian Barillot.

Following Lei Lin and Daniel Garcia Lorenzo's PhDs, we proposed an adaptive pixon represented segmentation (APRS) algorithm for 3D magnetic resonance (MR) brain images. Different from traditional method, an adaptive mean shift algorithm was adopted to adaptively smooth the query image and create a pixon-based image representation. Then K-means algorithm was employed to provide an initial segmentation by classifying the pixons in image into a predefined number of tissue classes. By using this segmentation as initialization, expectation-maximization (EM) iterations composed of bias correction, a priori digital brain atlas information, and Markov random field (MRF) segmentation were processed. Pixons were assigned with final labels when the algorithm converges. The adoption of bias correction and brain atlas made the current method more suitable for brain image segmentation than the previous pixon based segmentation algorithm. The proposed method was validated on both simulated normal brain images from BrainWeb and real brain images from the IBSR public dataset. Compared with some other popular MRI segmentation methods, the proposed method exhibited a higher degree of accuracy in segmenting both simulated and real 3D MRI brain data. The experimental results were numerically assessed using Dice and Tanimoto coefficient [18] .

EM-ICP strategies for joint mean shape and correspondences estimation: applications to statistical analysis of shape and of asymmetry

Participant : Sylvain Prima.

In collaboration with B. Combès, we proposed a new approach to compute the mean shape of unstructured, unlabelled point sets with an arbitrary number of points. This approach can be seen as an extension of the EM-ICP algorithm, where the fuzzy correspondences between each point set and the mean shape, the optimal non-linear transformations superposing them, and the mean shape itself, are iteratively estimated. Once the mean shape is computed, one can study the variability around this mean shape (e.g. using PCA) or perform statistical analysis of local anatomical characteristics (e.g. cortical thickness, asymmetry, curvature). To illustrate our method, we performed statistical shape analysis on human osseous labyrinths and statistical analysis of global cortical asymmetry on control subjects and subjects with situs inversus [29] . This work was led within the ARC 3D-MORPHINE (http://3dmorphine.inria.fr ).

Surface-based method to evaluate global brain shape asymmetries in human and chimpanzee brains

Participant : Sylvain Prima.

Following Phd and PostDoc works from Benoit Combès and Marc Fournier, in this work we used humans and chimpanzees brain MRI databases to develop methods for evaluating global brain asymmetries. We performed brain segmentation and hemispheric surface extraction on both populations. The human brain segmentation pipeline was adapted to chimpanzees in order to obtain results of good quality. To alleviate the problems due to cortical variability we proposed a mesh processing algorithm to compute the brain global shape. Surface-based global brain asymmetries were computed on chimpanzee and human subjects using individual mid-sagittal plane evaluation and population-level mean shape estimation. Asymmetry results were presented in terms of axis-wise components in order to perform more specific evaluation and comparison between the two populations [35] . This work was led within the ARC 3D-MORPHINE (http://3dmorphine.inria.fr ).

Computational techniques for the analysis of endocranial cast and endocranial structures

Participant : Sylvain Prima.

Following Phd and post-doc worlks from Benoit Combès and Marc Fournier, a series of studies were led within the ARC 3D-MORPHINE (http://3dmorphine.inria.fr ) and were presented at the 1836th Journées de la Société d'Anthropologie de Paris (January 26-28) and at the 80th annual meeting of the American Association of Physical Anthropologists (April 12-16). These include: a method to assess 3D endocranial asymmetries in extant and fossil species: new insights into paleoneurology [48] ; a method to map the distance between the brain and the inner surface of the skull [51] , [34] ; a method to compare bony labyrinths in humans, chimpanzees and baboons [28] ; a method for the analysis of the endocranial shape and its relationship with ectocranial structures [41] ; a new reconstruction of the frontal lobe and temporal pole of the Taung (Australopithecus africanus) endocast [32] .

Evaluation of Registration Methods on Thoracic CT: The EMPIRE10 Challenge

Participant : Olivier Commowick.

We participated, as part of a collaboration with the Asclepios team, to the EMPIRE10 challenge on registration. EMPIRE10 (Evaluation of Methods for Pulmonary Image REgistration 2010) is a public platform for fair and meaningful comparison of registration algorithms which are applied to a database of intra-patient thoracic CT image pairs. Evaluation of non-rigid registration techniques is a non trivial task. This is compounded by the fact that researchers typically test only on their own data, which varies widely. For this reason, reliable assessment and comparison of different registration algorithms has been virtually impossible in the past. In this work we present the results of the launch phase of EMPIRE10, which comprised the comprehensive evaluation and comparison of 20 individual algorithms from leading academic and industrial research groups. All algorithms are applied to the same set of 30 thoracic CT pairs. Algorithm settings and parameters are chosen by researchers expert in the configuration of their own method and the evaluation is independent, using the same criteria for all participants. All results are published on the EMPIRE10 website (http://empire10.isi.uu.nl ). The challenge remains ongoing and open to new participants. Full results from 24 algorithms have been published at the time of writing. This article details the organisation of the challenge, the data and evaluation methods and the outcome of the initial launch with 20 algorithms. More details are available in [20] .