Noise Reduction in Functional MR Images by Common Factor Models

摘要

This study proposes the use of common factor models to reduce noise in functional magnetic resonance (MR) images. The models estimate the errors due to instrumental instability, voxel specific noises and other nontask-related contamination. After noise reduction, the functional images can be analyzed by t-test, correlation analysis, independent component analysis, and neural network algorithms. This study also suggests the regression method for estimating both intensity waveforms in the reduced space (i.e., factor or component scores) and corrected waveforms in the original data space. The common factor models for noise reduction were tested in an event-related functional MR experiment using the concentric checkerboard pattern stimulus. The performance of common factor models was better than principal component analysis for noise reduction.