|
|
|
| e-Pub |
Previous | 
Home
Section: New Results
Image Computing: Detection, Segmentation, Registration and Analysis
Quantitative analysis of T2/T2* relaxation time alteration
Participants : Benoit Combès, Anne Kerbrat, Olivier Commowick, Christian Barillot.
T2 and T2* relaxometric data (https://en.wikipedia.org/wiki/Relaxation_(NMR)) becomes a standard tool for the quantitative assessment of brain tissues and of their changes along time or after the infusion of a contrast agent. Being able to detect significant changes of T2/T2* relaxation time is an important issue. Generally, such a task is performed by comparing the variability level in the regions of interest to the variability in the normal appearance white matter. However, in the case of T2 and T2* relaxometry, this solution is highly problematic. Indeed the level of noise in the normal appearance white matter is significanlty smaller than the level of noise in more intense region (e.g. MS lesions). Our aim is to provide a Bayesian analysis of T2/T2* relaxometry estimation and alteration. More specifically, we build posterior distributions for the relaxation time and the relaxation offset by elucidating the dedicated Jeffreys priors. Then the resulting posterior distributions can be evaluated using a Monte Carlo Markov Chain algorithm. Such an analysis has three main advantages over the classical estimation procedure. First it allows in a simple way to compute many estimators of the posterior including the mode, the mean, the variance and confidence intervals. Then, it allows to include prior information. Finally, becauce one can extract confidence interval from the posterior, testing properly whether the true relaxometry time is included within a certain range of value given a confidence level is simple. This work was published as a conference paper in MICCAI 2016 [22].
Block-Matching Distortion Correction of Echo-Planar Images with Opposite Phase Encoding Directions
Participants : Renaud Hédouin, Olivier Commowick, Élise Bannier, Christian Barillot.
By shortening the acquisition time of MRI, Echo Planar Imaging (EPI) enables the acquisition of a large number of images in a short time, compatible with clinical constraints as required for diffusion or functional MRI. However such images are subject to large, local distortions disrupting their correspondence with the underlying anatomy. The correction of those distortions is an open problem, especially in regions where large deformations occur. We have proposed a new block-matching registration method to perform EPI distortion correction based on the acquisition of two EPI with opposite phase encoding directions (PED). It relies on new transformations between blocks adapted to the EPI distortion model, and on an adapted optimization scheme to ensure an opposite symmetric transformation. We have produced qualitative and quantitative results of the block-matching correction using different metrics on a phantom dataset and on in-vivo data. We have shown the ability of the block-matching to robustly correct EPI distortion even in strongly affected areas. This work has been accepted for publication in IEEE Transactions in Medical Imaging 2017.
An a contrario approach for the detection of patient-specific brain perfusion abnormalities with arterial spin labelling
Participants : Pierre Maurel, Jean-Christophe Ferré, Christian Barillot.
In this work, we introduce a new locally multivariate procedure to quantitatively extract voxel-wise patterns of abnormal perfusion in individual patients. This a contrario approach uses a multivariate metric from the computer vision community that is suitable to detect abnormalities even in the presence of closeby hypo- and hyper-perfusions. This method takes into account local information without applying Gaussian smoothing to the data. Furthermore, to improve on the standard a contrario approach, which assumes white noise, we introduce an updated a contrario approach that takes into account the spatial coherency of the noise in the probability estimation. Validation is undertaken on a dataset of 25 patients diagnosed with brain tumors and 61 healthy volunteers. We show how the a contrario approach outperforms the massively univariate General Linear Model usually employed for this type of analysis. This work as been published in Neuroimage [14].
Dictionary Learning for Pattern Classification in Medical Imaging: Why Does Size Matter?
Participants : Hrishikesh Deshpande, Pierre Maurel, Christian Barillot.
Sparse representation based dictionary learning (DL) technique has proved to be an effective tool for image classification. While standard DL methods are effective in data representation, several discriminative DL methods have been proposed for learning dictionaries better suited for classification. Majority of these methods, in pattern recognition applications, learn the dictionaries for each class and com- pare the error terms of sparse reconstruction for each dictionary. However this raises a question that is still an open problem in the sparsity community: What role does the size of each dictionary play in the classification process? In this work, we prove that this parameter is pivotal, especially in cases where there are variability differences between classes. We illustrate our assertion on standard and discriminative DL techniques in two applications: Lips detection in face images and the classification of multiple sclerosis lesions in multi-channel brain MR images.