Signal Processing Methods for the Inter-subject Registration of Neuroimaging Data.

摘要

This thesis addresses the problem of inter-subject registration of neuroimaging data using both structural MRI (sMRI) and functional MRI (fMRI) data. This is an important step for improving the statistical power of fMRI group analyses. The standard approach to this problem matches anatomical features of the brain, such as major anatomical landmarks, or curvature patterns of the cerebral cortex. It is well known, however, that an accurate inter-subject functional correspondence cannot be derived using only anatomical features, since the size, shape and anatomical location of functional loci vary across subjects.;We propose and discuss a number of inter-subject registration algorithms that match functionally-defined features of the brain in order to derive an inter-subject functional correspondence. In the process, this thesis introduces a new form of registration objective that matches intra-dataset patterns of similarity. This objective has roots in the graph matching literature, and is motivated for the inter-subject registration problem based on recent neuroscience experiments that study functional connectivity, or the intra-subject temporal synchrony of functional response between remote regions of the brain. Here, the brain is modeled as a graph, with edges between pairs of vertices weighted by the similarity of functional response. The proposed algorithms are extensively validated on real fMRI experimental data, along with a comparative study against state-of-the-art anatomically-based registration algorithms.;Throughout this thesis, we address a number of computational challenges brought on by the high-dimensional nature of the fMRI data and the form of registration objective. A particular emphasis is placed on developing computationally efficient algorithms in the face of such large datasets.