Overall Objectives
Research Program
Application Domains
Highlights of the Year
New Software and Platforms
New Results
Bilateral Contracts and Grants with Industry
Partnerships and Cooperations
XML PDF e-pub
PDF e-Pub

Section: New Results

Statistical models for Neuroscience

Advanced statistical analysis of functional Arterial Spin Labelling data

Participants : Florence Forbes, Aina Frau Pascual.

Joint work with: Philippe Ciuciu from Team parietal and Neurospin, CEA Saclay.

Arterial Spin Labelling (ASL) is a non-invasive perfusion MR imaging technique that can be also used to measure brain function (fASL for functional ASL). In contrast to BOLD fMRI, it gives a quantitative and absolute measure of cerebral blood flow (CBF), making this modality appealing for clinical neuroscience and patient's follow-up over longitudinal studies. However, its limited signal-to-noise ratio makes the analysis of fASL data challenging. In this work, we compared different approaches (GLM vs JDE) in the analysis of functional ASL data for the detection of evoked brain activity at the group level during visual and motor task performance. Our dataset has been collected at Neurospin on a 3T Tim Trio Siemens scanner (CEA Saclay, France), during the HEROES project (Inria Grant). It contains BOLD data (165 scans, TR=2.5s, TE=30ms, 3x3x3mm3) and functional pulsed ASL data (Q2TIPS PICORE scheme [Luh,00], 165 scans, TR=2.5s, TE=11ms, 3x3x7.5 mm3) of 13 right-handed subjects (7 men and 6 women) of age between 20 and 29. The experimental design consists of a mini-block paradigm of visual, motor and auditory tasks with 16 blocks of 15s each followed by 10s of rest. Data have been scaled, realigned, and normalized. For univariate analysis, the images have also been spatially smoothed with a Gaussian kernel of 5 mm full width half at maximum. Three data analysis approaches have been compared: (a) univariate General Linear Model (GLM) that considers canonical shapes for the perfusion and hemodynamic responses; (b) physiologically informed joint detection estimation (PI-JDE) [4] that jointly estimates effect maps and response functions in a multivariate manner in a Bayesian framework; (c) A restricted version of PI-JDE that considers fixed canonical shapes for the perfusion and hemodynamic responses (PRF and HRF, respectively), defining an intermediate approach between the first two. Since methods (b)-(c) embed adaptive spatial regularization, they do not require a preliminary smoothing of the data. Our results demonstrate that the PI-JDE multivariate approach is a competing alternative to GLM for the analysis of fASL: it recovers more localized and stronger effects. Our findings also replicate the state-of-the-art by showing more localized activation patterns in perfusion as compared to hemodynamics.

BOLD VEM multi session extension of the JDE approach

Participants : Florence Forbes, Aina Frau Pascual.

Joint work with: Philippe Ciuciu from Team parietal and Neurospin, CEA Saclay.

The fast solution of the JDE approach for BOLD fMRI presented in [5] uses a variational expectation maximization (VEM) algorithm and considers a single session of BOLD data. This paper shows the faster performance of this algorithm with respect to the Markov Chain Monte Carlo (MCMC) approach presented in earlier work, with similar results. In fMRI, usually several sessions are acquired for the same subject to be able to compare them or combine them. In [73], a multiple-session extension of the JDE approach has been proposed to analyze several sessions together. The solution proposed uses MCMC and considers that the response levels have a mean value per condition and a common variance between sessions. In the context of Aina Frau's PhD, a VEM solution of this extension has been implemented. Experimental results have shown that the solution of the multiple-session VEM is not very different from the average of the results computed with single session VEM. For this reason, we proposed a heteroscedastic version of the multiple-session VEM. It amounts to considering session-specific variances. The goal is to be able to weight the importance of the different sessions so as to diminish the contribution of any potential noisy session to the final parameter estimates.

Estimating biophysical parameters from multimodal fMRI data

Participants : Florence Forbes, Pablo Mesejo Santiago.

Joint work with: Jan Warnking from Grenoble Institute of Neuroscience.

Functional Magnetic Resonance Imaging (fMRI) indirectly studies brain function. With Jan M. Warnking (Grenoble Institute of Neurosciences) we worked on the estimation of biophysical parameters from fMRI signals. We first used only BOLD signals, using a stochastic population-based optimization method to estimate 15 parameters without neither providing initial estimates nor computing gradients. Initial results were published at MICCAI 2015 and in the IEEE JSTSP journal [81], [82]. Also a MATLAB toolbox was released (see software section). The current ongoing work is to study the impact of the combination of different fMRI modalities in the estimation of this biophysical parameters. We can use 3 fMRI modalities (BOLD, ASL and MION) and 13 rats. We ran our optimizer with all possible combinations of modalities. The initial hypothesis was that as long as we introduce more fMRI modalities we would like to see more consistent estimates but we need to assess possible limits due to potential lack of data: only 13 rats, 6 of them without MION, and potential outliers among the rats that would better be excluded from the analysis.

Multi-subject joint parcelation detection estimation in functional MRI

Participant : Florence Forbes.

Joint work with: Lotfi Chaari, Mohanad Albughdadi, Jean-Yves Tourneret from IRIT-ENSEEIHT in Toulouse and Philippe Ciuciu from Neurospin, CEA Saclay.

fMRI experiments are usually conducted over a population of interest for investigating brain activity across different regions, stimuli and subjects. Multi-subject analysis usually proceeds in two steps: an intra-subject analysis is performed sequentially on each individual and then a group-level analysis is carried out to report significant results at the population level. This work considers an existing Joint Parcellation Detection Estimation (JPDE) model which performs joint hemodynamic parcellation, brain dynamics estimation and evoked activity detection. The hierarchy of the JPDE model is extended for multi-subject analysis in order to perform group-level parcellation. Then, the corresponding underlying dynamics is estimated in each parcel while the detection and estimation steps are iterated over each individual. Validation on synthetic and real fMRI data shows its robustness in inferring group-level parcellation and the corresponding hemodynamic profiles. This work has been presented at ISBI 2016 [42].

Automatic segmentation and characterization of brain tumors using robust multivariate clustering of multiparametric MRI

Participants : Florence Forbes, Alexis Arnaud.

Joint work with: Emmanuel Barbier and Benjamin Lemasson from Grenoble Institute of Neuroscience.

Brain tumor segmentation is a difficult task in the field of multiparametric MRI analysis because of the number of maps that are available. Furthermore, the characterization of brain tumors can be time-consuming, even for medical experts, and the reference method is biopsy which is a local and invasive technique. Because of this, it is important to develop automatic and non-invasive approaches in order to help the medical expert with these issues. In this study we use a robust statistical model-based method to classify multiparametric MRI of rat brains. The voxels are gather into classes resulting from multivariate multi-scaled Student distributions, which can accommodate outliers. First we adjust a mixture model on a reference group of rats to learn the MRI characteristics of healthy tissues. Second we use this model to delineate the brain tumors as atypical voxels in the data set of unhealthy rats. Third we adjust a new mixture model only on the atypical voxels to learn the MRI characteristics of tumorous tissues. Finally, we extract a fingerprint for each tumor type to make a tumor dictionary.

Our data set is composed of healthy rats (n=8 rats) and 4 groups of rats bearing a brain tumor model (n=8 per group). For each rat, we acquired 5 quantitative MRI parameters along 5 slices. And the proposed tumor dictionary reaches a rate of 75% of accurate prediction with a leave-one-out procedure.

Monitoring brain tumor evolution using multiparametric MRI

Participants : Florence Forbes, Alexis Arnaud.

Joint work with: Emmanuel Barbier, Nora Collomb and Benjamin Lemasson from Grenoble Institute of Neuroscience.

Analyzing brain tumor tissue composition can improve the handling of tumor growth and resistance to therapies. We showed on a 6 time point dataset of 8 rats that multiparametric MRI could be exploited via statistical clustering to quantify intra-lesional heterogeneity in space and time. More specifically, MRI can be used to map structural, eg diffusion, as well as functional, eg volume (BVf), vessel size (VSI), oxygen saturation of the tissue (StO2), characteristics. In previous work, these parameters were analyzed to show the great potential of multiparametric MRI (mpMRI) to monitor combined radio- and chemo-therapies. However, to exploit all the information contained in mpMRI while preserving information about tumor heterogeneity, new methods need to be developed. We demonstrated the ability of clustering analysis applied to longitudinal mpMRI to summarize and quantify intra-lesional heterogeneity during tumor growth. This study showed the interest of a clustering analysis on mpMRI data to monitor the evolution of brain tumor heterogeneity. It highlighted the type of tissue that mostly contributes to tumor development and could be used to refine the evaluation of therapies and to improve tumor prognosis.

Assessment of tissue injury in severe brain trauma

Participant : Florence Forbes.

Joint work with: Michel Dojat and Christophe Maggia from Grenoble Institute of Neuroscience and Senan Doyle from Pixyl.

Traumatic brain injury (TBI) remains a leading cause of death and disability among young people worldwide and current methods to predict long-term outcome are not strong. TBI initiates a cascade of events that can lead to secondary brain damage or exacerbate the primary injury, and these develop hours to days after the initial accident. The concept of secondary brain damage is the focus of modern TBI management in Intensive Care Units. The imbalance between oxygen supply to the brain tissue and utilization, i.e. brain tissue hypoxia, is considered the major cause for the development of secondary brain damage, and hence poor neurological outcome. Monitoring brain tissue oxygenation after TBI using brain tissue O2 pressure (PbtO2) probes surgically inserted into the parenchyma, may help clinicians to initiate adequate actions when episodes of brain ischemia/hypoxia are identified. The aggressive treatment of low PbtO2 values (<15mmHg for more than 30 minutes) was associated with better outcome compared to standard therapy in some cohort studies of severe head-injury patients. However, another study was unable to find similar benefits to patient outcome. MRI is an excellent modality for estimating global and regional alterations in TBI and for following their longitudinal evolution. To assess the complexity of TBI, several morphological sequences are required for assessing volume loss. Moreover, diffusion tensor imaging (DTI) offers the most sensitive modality for the detection of changes in the acute phase of TBI and increases the accuracy of long-term outcome prediction compared to the available clinical/radiographic prognostic score. Mean Diffusivity (MD) or Apparent Diffusion Coefficient (ADC) have been widely used to determine the volume of ischemic tissue, and assess intra- and extracellular conditions. A reduction of MD is related to cytotoxic edema (intracellular) while an increase of MD indicates a vasogenic edema (extracellular). Changes of MD are expected with severe TBI. The volume of lesions on DTI shows a strong correlation with neurological outcome at patient discharge. We consider a clinically relevant criterion to be the volume of vulnerable brain lesions after TBI, as previously suggested. In consequence, we need an automatic segmentation method to assess the tissue damage in severe trauma, acute phase i.e. before 10 days after the event. Skull deformation, the presence of blood in the acute phase, the high variability of brain damage that excludes the use of anatomical a priori information, and the diffuse aspect of brain injury affecting potentially all brain structures, render TBI segmentation particularly demanding. The methods proposed in the literature are mainly concerned with volumetric changes following TBI and scarcely report lesion load. In this work, we report our methodological developments to assess lesion load in severe brain trauma in the entire brain. We use P-LOCUS to perform brain tissue segmentation and exclude voxels labeled as CSF, ventricle and hemorrhagic lesion. We propose a fusion of several atlases to parcel cortical, subcortical and WM structures into well identified regions where MD values can be expected to be homogenous. Abnormal voxels are detected in these regions by comparing MD values with normative values computed from healthy volunteers. The preliminary results, evaluated in a single center, are a first step in defining a robust methodology intended to be used in multi-center studies. This work has been published in [58].

Automatic multiple sclerosis lesion segmentation with P-Locus

Participant : Florence Forbes.

Joint work with: Michel Dojat from Grenoble Institute of Neuroscience and Senan Doyle from Pixyl.

P-LOCUS provides automatic quantitative neuroimaging biomarker extraction tools to aid diagnosis, prognosis and follow-up in multiple sclerosis studies. The software performs accurate and precise segmentation of multiple sclerosis lesions in a multi-stage process. In the first step, a weighted Gaussian tissue model is used to perform a robust segmentation. The algorithm avails of complementary information from multiple MR sequences, and includes additional estimated weight variables to account for the relative importance of each voxel. These estimated weights are used to define candidate lesion voxels that are not well described by a normal tissue model. In the second step, the candidate lesion regions are used to populate the weighted Gaussian model and guide convergence to an optimal solution. The segmentation is unsupervised, removing the need for a training dataset, and providing independence from specific scanner type and MRI scanner protocol. The procedure was applied to participate to the MSSEG Challenge at Miccai 2016 in Athen: Multiple Sclerosis Lesions Segmentation Challenge Using a Data Management and Processing Infrastructure [55].