Skip to Main Content
Accurate detection of low contrast-to-noise ratio (CNR) blood oxygenation level dependent (BOLD) signals in functional magnetic resonance imaging (fMRI) data is important for presurgical planning and cognitive research. Robust detection is challenging in small regions of low CNR activation since the probability of detecting each individual voxel is low. We present a processing technique for improving the detection of localized low CNR BOLD signals in fMRI data. When applied to synthetic fMRI data, this blind estimation scheme significantly improves the probability of correctly detecting voxels in a small region of activation with a CNR between 0.5 and 1.0 compared to the standard general linear model approach. More activation is detected in expected (based on input stimulus) regions of experimental data after processing with the proposed technique.