Understanding the visual cortex by using classification techniques

Abstract : In this thesis, we present different approaches for statistical learning that can be used for studying the neural code of cognitive functions, based on brain functional Magnetic Resonance Imaging (fMRI) data. In particular, we study the spatial organization of the neural code, i.e. the spatial localization and the respective weights of the different entities implied in the neural coding. In this thesis, we focus on the visual cortex. In the first part of this thesis, we introduce the notions of functional architecture, neural coding and functional imaging. Then, we study the limits of the classical approach for the characterization of the neural code from fMRI images, and the advantages of a recent method of analysis, namely inverse inference. Finally, we detail the statistical learning approaches used for inverse inference, and we evaluate them on real data. This study highlights the limitations of these approaches, that will be addressed during this thesis. In particular, we focus on the use of spatial information within statistical learning methods. In a second part, we describe the three main contributions of this thesis. First, we introduce a Bayesian framework for sparse regularization, that generalizes two reference approaches. Then, we propose a supervised clustering method, that takes into account the spatial structure of the images. The resulting weighted maps are easily interpretable, and this approach seems particularly interesting in the case of inter-subjects inference. The last contribution of this thesis aims at including the spatial information into the regularization framework. This regularization is thus used in both regression and classification settings, and extracts clusters of predictive voxels. This approach is well suited for the decoding problem addressed in this thesis.
Complete list of metadatas

Cited literature [249 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00550047
Contributor : Vincent Michel <>
Submitted on : Thursday, December 23, 2010 - 11:31:31 AM
Last modification on : Thursday, March 7, 2019 - 3:34:14 PM
Long-term archiving on : Monday, November 5, 2012 - 2:51:58 PM

Identifiers

  • HAL Id : tel-00550047, version 1

Collections

Citation

Vincent Michel. Understanding the visual cortex by using classification techniques. Human-Computer Interaction [cs.HC]. Université Paris Sud - Paris XI, 2010. English. ⟨tel-00550047⟩

Share

Metrics

Record views

599

Files downloads

1121