Toward open sharing of task-based fMRI data: the OpenfMRI project

Abstract

The large-scale sharing of task-based functional neuroimaging data has the potential to allow novel insights into the organization of mental function in the brain, but the field of neuroimaging has lagged behind other areas of bioscience in the development of data sharing resources. This paper describes the OpenFMRI project (accessible online at http://www.openfmri.org), which aims to provide the neuroimaging community with a resource to support open sharing of task-based fMRI studies. We describe the motivation behind the project, focusing particularly on how this project addresses some of the well-known challenges to sharing of task-based fMRI data. Results from a preliminary analysis of the current database are presented, which demonstrate the ability to classify between task contrasts with high generalization accuracy across subjects, and the ability to identify individual subjects from their activation maps with moderately high accuracy. Clustering analyses show that the similarity relations between statistical maps have a somewhat orderly relation to the mental functions engaged by the relevant tasks. These results highlight the potential of the project to support large-scale multivariate analyses of the relation between mental processes and brain function.

Keywords: classification; data sharing; informatics; metadata; multivariate.

Figure 1

An overview of the draft data organization scheme for the OpenFMRI project. A schematic of the directory tree for a dataset is presented, with each subdirectory shown on a separate branch. This structure allows specification of an arbitrary set of tasks, runs, and statistical models. The key files included in the base directory for the dataset specify features that are consistent across all of the data (such as demographics, task naming, and scan ordering), while key files in subdirectories specify details that may change across models or runs.

Figure 2

Rendered maps of the voxels with significant loadings on the 20 ICA components identified statistical images for the datasets listed in Table 1. Each column displays the loading for a single component; voxels shown in red had positive loading for that component.

Figure 3

Classification accuracy using reduced-dimension data, as a function of the number of ICA components in the dimensionality reduction step. Dimensionality reduction was first performed on an independent set of data (from run 2 for each subject), and the data from run 1 were then projected onto those components. Reported accuracy (thick lines) reflects average accuracy of task classification across all data points, from a total of 25 possible labels. The thin lines at the bottom reflect the empirically derived 95% cutoff for the null hypothesis of chance accuracy, obtained by performing the same classification 100 times with randomized labels, and taking the 95th largest value. RBF, radial basis function; SVM, support vector machine.

Figure 4

Hierarchical clustering of statistical maps across tasks, after projection into the 20-dimensional ICA space depicted in Figure 2.