Data acquisition for high resolution blood oxygen level dependent fMRI
Abstract
Blood oxygen level dependent functional magnetic resonance imaging (BOLD fMRI) is an important non-invasive technique for investigation of brain function. Using BOLD fMRI, brain activity can be localized with high spatial and moderate temporal resolution and the technique is therefore often used in neuroscientific research studies. Although BOLD fMRI is a well established technique, further development of acquisition, hardware and analysis is likely to improve both the accuracy and sensitivity of fMRI.
In this thesis, several different acquisition methods for BOLD fMRI, as well as some sources of errors, were investigated. Several variants of 3Decho- planar imaging (EPI) were investigated, as alternatives to 2D-EPI. 2D-EPI is the acquisition method commonly used today. Activation statistics based on datasets acquired using 2D-EPI and 3D-EPI, respectively, were compared. Noise from physiological processes such as cardiac motion, respiration and subject motion, had different impact on the datasets acquired by the two methods. By including the effect of the signal fluctuations caused by these noise sources most of the difference in the activation statistics based on the datasets from the two acquisition methods was removed. However, in studies of higher cognitive functions, fluctuations in physiological processes are likely to be partially correlated with the neuronal activity of interest. In such situations, 2D-EPI should be the preferred choice, as it has lower risk of bias in the fMRI data.
EPI acquisitions are prone to geometric distortion of the images, because of the way the MR-signal is sampled. These geometric distortions are largest near air filled cavities in the head, such as the sinuses and the ear channels. Several methods to correct these geometric distortions exist. Activation statistics were calculated based on both distortion corrected and uncorrected fMRI data which were acquired using an imaging protocol optimized for low geometric distortion. The correction only had a small effect in the hippocampal region. A larger effect was observed in the entorhinal cortex where a larger region was estimated to be activated in the corrected dataset.
A number of different ways to acquire undersampled 3D-EPI datasets, in combination with different reconstruction methods, were compared in the final study. The results indicated that compressed sensing techniques only have a limited potential to increase undersampling beyond that of traditional techniques, if only spatial sparsity is exploited.