A toolbox for the visualization and meta-analysis of functional brain organization of the cortical surface using an anatomical database

Abstract

The recently introduced MatLab toolbox VAMCA (Visualization And Meta-analysis on Cortical Anatomy) [1] provides surface-based visualization of mean cortical functional activations that are published as stereotaxic 3D coordinates (www/ebire.org/hcnlab). VAMCA uses a database of cortices from 72 healthy subjects to locate activations on a standardized cortical surface by extending the technique of multi-fiducial mapping [2] in order to perform meta-analyses. We demonstrate that the multifiducial method should be able to reproduce cortical locations of functional activations. Non-parametric statistical tests are provided for determining (a) whether two groups of foci are in the same cortical location; (b) the extent of overlap of the two groups’ foci; and (c) whether two groups of foci are differentially concentrated in any of the anatomically defined regions of interest (ROI) as defined using FreeSurfer [3]. We explore extending VAMCA’s functionality in two ways. First, we perform a proof-of-principle on the inclusion of studies in meta-analyses that do not report results in stereotaxic coordinates. E.g., electronic journal images can be captured, normalized to 3D space in a semi-automated fashion, and then projected to the cortical surface and used in meta-analyses. Second, a new distance metric, the “least axon distance (LAD)”, is introduced. LAD reflects the minimum distance passing through brain matter between any pair of cortical surface points and provides a functionally relevant alternative to the 3D and 2D distance metrics currently used in VAMCA’s statistical tests. Multi-fiducial Database: VAMCA helps users visualize probable cortical locations associated with MNI/Talairach 3D coordinates by enhancing the technique of multi-fiducial mapping [2]. A cortical database is used to map 3D locations to multiple control subjects’ cortical surfaces in order to assist cortical localization. Subject database: Brain anatomical data from young (18-48 y.o.) healthy subjects: 60 right-handed (33 male) and 12 left-handed (6 male) subjects are included. Processing: Freesurfer [3] renders cortical surfaces and registers them to a mean cortical surface. SPM5 segments and normalizes the T1 images to MNI-152 space using affine-only coregistration. Multi-fiducial map database: 72 Normalized cortical surface space 3D MNI space maps. ROI identification: Freesurfer [3] also identifies 3 different sets of anatomical-based ROIs on each of the 72 cortical surfaces: Lobar ROIs (6 ROIs), Gryal/Sulcal-based ROIs (82 ROIs), and intermediate-sized ROIs (36 ROIs). Flatmaps generated from the mean registered cortical anatomy Multi-fiducial mapping of a 3D foci (circled yellow) to 60 individual cortical locations (red). Green cross is the 2D median location of the 60 points – the likely surface location that is used in subsequent processing. Stereotaxic coordinates in text files processed in MatLab. An occipital lobe coordinate is circled. Cortical surface identification and rendering – gyri (green) and sulci (red) shown Cortical inflation and surface registration to a common cortical coordinate system ROIs marked out on mean cortical anatomy associated with MNI space Mapping a stereotaxic location to the cortical surface Anatomical (sulci=black) shows that most (65%) individuals place the coordinate in Lingual/OccTemp Sulc rather than in Calcarine. Monte Carlo analysis verifies how well noisy, simulated cortical fMRI activations processed in MNI space might be identified: (a) Simulated fMRI activations are centered in various locations across the normalized cortical hemisphere and mapped to the cortex for randomly selected database subjects. (b) Location noises are added to activation locations: on the cortical surface to simulate functional/anatomical colocation variability, surface registration errors, etc. in 3D after converting to individual subjects’ MNI coordinates to simulate fMRI/anatomy coregistration error, MNI normalization error, and cluster maxima location variability, etc. (c) Computed MNI coordinates are processed using VAMCA to estimate cortical activation displacement upon reconstruction.