Non-Cartesian Parallel Image Reconstruction for Functional MRI.

Abstract

Functional MRI (fMRI) is a widely preferred imaging modality to detect the regional brain activation associated with a task. However, the current fMRI techniques can suffer from performance degradation due to low temporal resolution from the hardware limitations, noise from systematic or measurement errors, and image distortions or artifacts from insufficient or inaccurate data. Fortunately, parallel imaging technique can address these problems by reducing readout time, thus, improving spatial or temporal resolution or suppressing susceptibility artifacts. However, due to the reduced number of samples, current parallel imaging technique experiences problems such as aliasing artifacts and low SNR. Therefore, our goal was to improve the parallel image reconstruction method to create truly effective solution for current limitation for fMRI for brain studies by alleviating its problems. Specifically, this dissertation describes investigation and analysis of methods for updating calibration data, such as the sensitivity map or GRAPPA coefficients, as well as by improving image reconstruction methods for non-Cartesian SENSE. For improving the image reconstruction algorithms, we focused on improving spiral SENSE with an iterative CG algorithm. This includes developing a joint estimation method of image and coil sensitivity map with a quadratic regularization on the estimation of sensitivity map, and developing an improved regularization technique that controls edge effects. Joint estimation of the image and sensitivity map resulted in much more robust performance compared to non-joint estimation approach when the initial sensitivity map is inaccurate. Selection of image support region method suppressed aliasing artifact more successfully when using a softening function with a proper mask size. For determination of the calibration data, we analyzed several self-calibrated sensitivity map estimation methods and investigated the effect of smoothing on time series fMRI data. We also compared several updating methods of calibration data on fMRI data. Proper selection of estimation and updating technique for calibration data improved image quality, time series SNR and brain activation. Our studies show the impact of improving and optimizing spiral SENSE and spiral GRAPPA, which may produce more effective solutions for parallel imaging in functional MRI.