VARIOGRAM MODELING FOR SPATIAL CORRELATION IN STRUCTURAL MRI IMAGES

Abstract

In recent years neuroimaging techniques growth help us to understand the working of the human brain by using structural magnetic resonance imaging (sMRI) and functional magnetic resonance imaging (fMRI). Structural MRIs are used to show the main structure of the brain organism such as gray matter, white matter, and cerebrospinal fluid (CSF). The functional MRI is used to study the brain activity when performing an assigned task like eyes moving and talking. It uses the bloodoxygen- level dependent (BOLD) contrast, when there is an activity in a region of the brain the blood flow to that area will increase. In this research, geostatistical techniques such as the variogram and kriging approaches are utilized to uncover the spatial correlation in structural magnetic resonance imaging (sMRI) data and to predict the effect of a brain tumor on brain regions. We propose different variogram models approach for three brain slices containing a brain tumor and we find that the best models for slices 10 and 11 is the exponential model and for slice 12 is the Gaussian model, the best model is selected by using the cross - validation method in kriging. A bootstrap resampling method is used to estimate the empirical variogram values and the parameters of the selected models.