Mahboobeh Maghami,1Sayed Mohsen Hosseini,2,*
1. Department of Biostatistics and Epidemiology, School of Public Health, Isfahan University of Medical Sciences, Isfahan, Iran 2. Department of Biostatistics and Epidemiology, School of Public Health, Isfahan University of Medical Sciences, Isfahan, Iran
Introduction: Determining the location and intensity of brain activity in response to a stimulus or task is one of the goals of functional magnetic resonance imaging (fMRI). This study aims to describe classical and Bayesian methods and compare the power of classical and Bayesian models to detect activated regions.
Methods: Before analyzing FMRI data, pre-processing steps are performed on the data. These steps are to reduce all kinds of noise. These steps are motion correction, slice-timing, normalization and smoothing. After preprocessing the fMRI data, three techniques for fMRI analysis have been presented, with the statistical parametric map (SPM) being the most popular. We have also used the Bayesian method for analysis with two different algorithms.
Results: SPM has limitations in preserving data edges and can lead to inaccurate results. Bayesian inference is an alternative approach that uses the posterior probability distribution of activation given the data. Therefore, in this paper, we present data preprocessing, describe classical and Bayesian methods, and compare the power of classical and Bayesian models to detect activated regions. Active voxels are displayed in SPM maps for classical inference and PPM maps for Bayesian inference.
Conclusion: SPM identifies a smaller number of activated voxels than the PPM and SPM becomes more conservative and classical inference is relatively insensitive. The improved VB algorithm method for Bayesian inference detects more activated voxels than classical inference. The results of the Bayesian method with the original VB algorithm are very similar to those of the Bayesian method with the improved VB algorithm. However, the number of activated voxels is greater with the improved VB algorithm method.