Information distribution analysis in the fMRI measurements with degree of locality estimation

Abstract

In this investigation, we propose a new method for analyzing and representing the distribution of discriminative information for pattern analysis of functional Magnetic Resonance Imaging (fMRI) data. For this purpose, a spatially local mesh with varying size, around each voxel (called seed voxel) is formed. The relationship among each seed voxel and its neighbors are estimated using a linear regression model by minimizing the square error. Then, the optimal mesh size which represents the connections among each seed voxel and its surroundings is estimated by minimizing Akaike's Final Prediction Error (FPE) with respect to the mesh size. The degree of locality is represented by the optimum mesh size. If the estimated mesh size is small, then the seed voxel is assumed to be connected to few other voxels; if it is large, the voxel is assumed to be massively connected to other voxels. It is shown that, the local mesh size with highest discriminative power depend on the individual subjects. Surprisingly, the optimal mesh size remains the same for the recognition task of different categories. The proposed method was tested on a memory task, which requires retrieval of item and temporal order information from memory. For each participant, estimated arc weights of each local mesh with different mesh size are used to classify the two types of information retrieved from memory (i.e. item and temporal order). Classification accuracies for each subject are found using k-Nearest Neighbor (k-NN) method. The results indicate that the proposed local mesh model with optimal mesh size can successfully represent discriminative information.