CONSENSUS PATTERNS IN FUNCTIONAL NEUROIMAGING: Overview

A wide range of techniques and software tools have become available with which to process fMRI datasets. To date this has not been accompanied by development of a similarly wide range of performance metrics or benchmark datasets with which to evaluate and compare the tools. One of the primary goals of this proposal is to provide a mechanism for calculating performance metrics within an extensible software framework for evaluation of fMRI processing tools, using benchmark datasets and their test results. In addition, we propose to develop the use of combinations of activation maps in order to generate consensus maps with improved performance metrics, and to avoid having to choose a single approach to post-acquisition processing and data analysis of fMRI datasets.

Similarity Pattern Maps (SPMs) after histogram equalization for four data analysis models applied to a patch of real fMRI noise with a simulated fMRI signal composed of a spatially varying hemodynamic response function convolved with a simple on-off signal response. The consensus image appears closer to the ground-truth binary image than to any of the individual images. [FIR, finite impulse response; NNS, Neural network saliency map; PFT, Parametric Fourier transform; CVA, Canonical variate analysis] Figure from: Hansen LK, Nielsen JA, Strother SC, Lange N. Consensus Inference In Neuroimaging. Neuroimage 13:1212-1218, 2001.