ABSTRACT
Detecting the most relevant brain regions for explaining the distinction between cognitive conditions is one of the most sought after objectives in neuroimaging research. A popular approach for achieving this goal is the multivariate pattern analysis (MVPA) which is commonly conducted through the searchlight procedure as well as a number of other approaches. This is due to advantages of such methods which include being intuitive and flexible with regards to size of the search space. However, these approaches suffer from a number of limitations that lead to misidentification of truly informative voxels or clusters of voxels which in turn results in imprecise information maps. The limitations of such procedures mainly stem from several factors such as the fact that the information value of the search spheres are assigned to the voxel at the center of them (in case of searchlight), the requirement for manual tuning of parameters such as searchlight radius and shape and other optimization parameters, overlooking the structure and interactions within the regions, and the drawbacks of using regularization methods in analysis of datasets with characteristics of common fMRI data. In this paper, we propose a fully data-driven maximum relevance minimum redundancy search algorithm for detecting precise information value of voxel-level clusters within brain regions while alleviating the above mentioned limitations. In order to make the algorithm efficient, we propose an implementation based on principles of dynamic programming. We evaluate and compare the proposed algorithm with the searchlight procedure using both real and synthetic datasets.
