|
|
|
| e-Pub |
Previous | 
Home
Section: New Results
Image Computing: Detection, Segmentation, Registration and Analysis
A Mathematical Framework for the Registration and Analysis of Multi-Fascicle Models for Population Studies of the Brain Microstructure
Participant : Olivier Commowick.
Diffusion tensor imaging (DTI) is unable to represent the diffusion signal arising from multiple crossing fascicles and freely diffusing water molecules. Generative models of the diffusion signal, such as multi-fascicle models, overcome this limitation by providing a parametric representation for the signal contribution of each population of water molecules. These models are of great interest in population studies to characterize and compare the brain microstructural properties. Central to population studies is the construction of an atlas and the registration of all subjects to it. However, the appropriate definition of registration and atlasing methods for multi-fascicle models have proven challenging. This paper proposes [32] a mathematical framework to register and analyze multi-fascicle models. Specifically, we define novel operators to achieve interpolation, smoothing and averaging of multi-fascicle models. We also define a novel similarity metric to spatially align multi-fascicle models. Our framework enables simultaneous comparisons of different microstructural properties that are confounded in conventional DTI. The framework is validated on multi-fascicle models from 24 healthy subjects and 38 patients with tuberous sclerosis complex, 10 of whom have autism. We demonstrate the use of the multi-fascicle models registration and analysis framework in a population study of autism spectrum disorder.
Multimodal rigid-body registration of 3D brain images using bilateral symmetry
Participants : Sylvain Prima, Olivier Commowick.
In this paper we show how to use the approximate bilateral symmetry of the brain with respect to its interhemispheric fissure for intra-subject (rigid-body) mono- and multimodal 3D image registration. We propose to define and compute an approximate symmetry plane in the two images to register and to use these two planes as constraints in the registration problem. This 6-parameter problem is thus turned into three successive 3-parameter problems. Our hope is that the lower dimension of the parameter space makes these three subproblems easier and faster to solve than the initial one. We implement two algorithms to solve these three subproblems in the exact same way, within a common intensity-based framework using mutual information as the similarity measure. We compare this symmetry-based strategy with the standard approach (i.e. direct estimation of a 6-parameter rigid-body transformation), also implemented within the same framework, using synthetic and real datasets. We show in [44] our symmetry-based method to achieve subvoxel accuracy with better robustness and larger capture range than the standard approach, while being slightly less accurate and slower. Our method also succeeds in registering clinical MR and PET images with a much better accuracy than the standard approach. Finally, we propose a third strategy to decrease the run time of the symmetry-based approach and we give some ideas, to be tested in future works, on how to improve its accuracy.
Distortion Correction in EPI Diffusion Weighted Images
Participants : Renaud Hedouin, Olivier Commowick.
We have compared and developed several methods which correct distortion of EPI images. The most popular method field map do not give optimal results. We have implemented and improved a method based on reversed phase encoding gradient which give good results. To correct diffusion weighted images this method only need one reversed phase encoding gradient image which not need substantial additional acquisition time.
Using bilateral symmetry to improve non-local means denoising of MR brain images
Participants : Sylvain Prima, Olivier Commowick.
The popular NL-means denoising algorithm proposes to modify the intensity of each voxel of an image by a weighted sum of the intensities of similar voxels. The success of the NL-means rests on the fact that there are typically enough such similar voxels in natural, and even medical images; in other words, that there is some self-similarity/redundancy in such images. However, similarity between voxels (or rather, between patches around them) is usually only assessed in a spatial neighbourhood of the voxel under study. As the human brain exhibits approximate bilateral symmetry, one could wonder whether a voxel in a brain image could be more accurately denoised using information from both ipsi- and contralateral hemispheres. This is the idea we have investigated in this paper [45] . We define and compute a mid-sagittal plane which best superposes the brain with itself when mirrored about the plane. Then we use this plane to double the size of the neighbourhoods and hopefully find additional interesting voxels to be included in the weighted sum. We evaluate this strategy using an extensive set of experiments on both simulated and real datasets.
Detection of Multiple Sclerosis Lesions using Dictionary Learning
Participants : Hrishikesh Deshpande, Pierre Maurel, Christian Barillot.
Multiple sclerosis (MS) is a chronic, autoimmune, inflammatory disease of the central nervous system, in which certain areas of brain develop MS lesions, which are characterized by demyelination. Over the last years, various models combined with supervised and unsupervised classification methods have been proposed for detection of MS lesions using magnetic resonance images. Recently, signal modeling using sparse representations (SR) has gained tremendous attention and is an area of active research. SR allows coding data as sparse linear combinations of the elements of over-complete dictionary and has led to interesting image recognition results. The dictionary used for sparse coding plays a key role in the classification process. In this work, we have proposed to learn class specific dictionaries and develop new classification scheme, to automatically detect MS lesions in 3-D multi-channel magnetic resonance images.
Multiple Sclerosis Lesion Detection in Clinically Isolated Syndromes
Participants : Yogesh Karpate, Olivier Commowick, Christian Barillot.
Quantitative assessment of Multiple Sclerosis Lesions (MSL) in Clinically Isolated Syndromes (CIS) is important, as they are a precursor to subsequent stages of the disease. We address the problem of lesion patch detection with respect to Normally Appearing Brain Tissues (NABT). Our approach consists in learning rotationally invariant MSL and NABT multimodal intensity signatures based on 3D spherical gabor descriptors. This learning step, done once and for all, is followed by a testing step for the patient patches with an exemplar SVM. First, we develop a framework for selecting focused region of interest (fROI) using linear SVM for scoring. This allows an excellent trade-off between speed and accuracy. Second, building rotational invariant and scale independent features for accurate representation of image signatures. The extracted features are sensitive to the orientation of the analyzed image. This is a drawback in classification and retrieval applications. We handle this problem by using shperical Gabor descriptors. And last, we apply max pooling for down sampling of feature vectors. For the classification purpose we use a standard linear Support Vector Machine(SVM). The main contribution of the work is to build binary classifier to discriminate NABTs and MSLs based upon robust image representation. We have validated our approach on synthetic and real patient data. The synthetic lesion data is generated with noise, without noise and with bias field. Further, validation is carried out in three different scenarios. First, we evaluate our classifier using K-fold started with cross validation using NABT from healthy volunteers and MSL from CIS patients, then the detection of NABT and MSL from CIS patients on known patches is performed. The last evaluation concerned the full search algorithm.
Intensity Normalization in Longitudinal MS Patients
Participants : Yogesh Karpate, Olivier Commowick, Christian Barillot.
This work proposes a longitudinal intensity normalization algorithm for multi-channel MRI of brain of MS patient in the presence of lesions, aiming towards stable and consistent longitudinal segmentation. This approach is parametric and developed using two different forms of Robust Expectation Maximization (EM). The first is Spatio-Temporal Robust Expected Maximization (STREM) and other being EM with beta divergence. We validated our method on real longitudinal multiple sclerosis subjects.