This research plan outlines crucial software enhancements to a program called 3dsvm, which is a command line program and graphical user interface (gui) plugin for AFNI (Cox, 1996). 3dsvm performs support vector machine (SVM) analysis on fMRI data, which constitutes one important approach to performing multivariate supervised learning of neuroimaging data. 3dsvm originally provided the ability to analyze fMRI data as described in (LaConte et al., 2005). Since its first distribution as a part of AFNI, it has been steadily extended to provide new functionality including regression and non-linear kernels, as well as multiclass classification capabilities. In addition to its integration into AFNI, features that make 3dsvm particularly well suited for fMRI analysis are that it is easy to spatially mask voxels (to include/exclude them in the SVM analysis) as well as to flexibly select subsets of a dataset to use as training or testing samples. It has been used to generate results for our own work and for collaborative efforts and has been cited as a resource by others (Mur et al. 2009;Hanke et al. 2009). Despite many positive aspects of 3dsvm, the priorities of PAR-07-417 address a genuine need that this software project has - the ability to focus on improvements that will increase its dissemination and interoperability. A major motivation for PAR-07-417 is to facilitate the improved interface, characterization, and documentation to enhance the extent of sharing and to provide the groundwork for future extensions.
Our aims are well aligned with this program announcement. Further, there is a growing need in the neuroimaging community for tools such as 3dsvm. Since 3dsvm is not a new project, is tightly integrated into the software environment of AFNI, and can be further integrated to enable better functionality to support needs as diverse as NIfTI format capabilities to rtFMRI, this proposed project will help to further the NIH Blueprint for Neuroscience Research by supporting its need for wide-spread adoption of high-quality neuroimaging tools.
This proposal focuses on improving, characterizing, and documenting an existing neuroinformatics software tool. The project described will help to further the NIH Blueprint for Neuroscience Research by supporting its need for wide-spread adoption of high-quality neuroimaging tools.
