Statistical Models for the analysis of ASL and BOLD Magnetic Resonance modalities to study brain function and disease

Abstract : Functional and perfusion imaging modalities are closely related since they both measure, directly or indirectly, blood flow in the brain. Functional Magnetic Resonance Imaging (fMRI) using the blood oxygen level dependent (BOLD) contrast exploits the magnetic properties of blood (oxy- and deoxyhemoglobin) to measure local changes in blood oxygen concentration in the brain. The neurovascular coupling allows us to infer brain function from fMRI images. Perfusion MRI images the cerebral vascular system by directly measuring blood flow. In particular, Arterial Spin Labeling (ASL) does not need contrast agents; it uses spins of endogenous water protons as a tracer instead. Usually ASL is used to probe the basal perfusion at rest. However, in the recent years, it has also been used as a functional imaging modality (as fASL) by tracking task-related perfusion changes. In contrast to the standard BOLD fMRI, results are quantitative, making this type of data attractive for use in clinical research.This thesis focuses on the investigation of the fASL modality and the development of new methods to analyze it. As previously done for BOLD data, a Bayesian framework is proposed for the analysis of fASL data. It provides a way of modeling activation values and both hemodynamic and perfusion response functions as probabilistic variables in the so-called joint detection estimation (JDE) framework. Bayesian models use a priori knowledge in the estimation of unknown parameters through the specification of probability distributions. In this work, we exploit this feature to incorporate physiological information to make the estimation more robust. In particular, we use physiological models based on the balloon model to derive a link between hemodynamic and perfusion responses and we turn this link into a prior distribution to regularize the estimation of the responses. A Markov Chain Monte Carlo solution with prior physiological knowledge has been first proposed for the estimation of the quantities contained in the fMRI signal. Since the computational cost of this algorithm is very high, we then reformulate the problem to use a variational expectation maximization approach that provides a much faster algorithm with similar results. The use of priors and constraints in this setting is also more straightforward.These methods have been evaluated on two different datasets using event-related and block designs with very simple experimental tasks. We show the performance of the methods investigated in comparison to standard methods at the subject and group levels. Experimental results show the utility of using physiological priors for improving the recovery of a perfusion response function. They also demonstrate that BOLD fMRI achieves better sensitivity to detect evoked brain activity as compared to fASL although fASL gives a more localized activation, which is in line with the existing literature. From the results, we discuss the impact of the modelling of spatial correlation, as well as the impact of the estimation of temporal responses.This work proposes new methodological contributions in the study of a relatively new fMRI modality that is functional ASL, and puts it into perspective with the existing techniques. Thus, we provide new tools for the neuroscientific community to study and understand brain function. These tools have been implemented in python in the PyHRF package.
Document type :
Theses
Liste complète des métadonnées

https://hal.inria.fr/tel-01440495
Contributor : Aina Frau-Pascual <>
Submitted on : Thursday, January 19, 2017 - 12:07:51 PM
Last modification on : Monday, April 9, 2018 - 12:22:16 PM
Document(s) archivé(s) le : Thursday, April 20, 2017 - 1:10:50 PM

Identifiers

  • HAL Id : tel-01440495, version 1

Citation

Aina Frau Pascual. Statistical Models for the analysis of ASL and BOLD Magnetic Resonance modalities to study brain function and disease. Medical Imaging. Université Grenoble Alpes, 2016. English. ⟨NNT : 2016GREAM086⟩. ⟨tel-01440495v1⟩

Share

Metrics

Record views

399

Files downloads

284