Evaluating and Enhancing the Precision and Accuracy of Quantitative MRI

Abstract

In contrast to traditional magnetic resonance imaging (MRI) that produces weighted images contaminated by experiment- and/or scanner-specific parameters (e.g., receive coil sensitivity, receiver gain, and image scaling), quantitative MRI (qMRI) has the potential to provide maps whose intensities represent one of the biophysical parameters (e.g., T1, T2, T2* relaxation times, magnetization transfer, proton density, mean/radial/axial diffusivity, and cerebral blood flow). To obtain a meaningful quantitative map, measurements need to be both accurate and precise. However, numerous factors could compromise the accuracy and/or precision of the measurement. This includes gradient nonlinearities, eddy currents, gradient heating, radiofrequency transmit field (B1+) inhomogeneity, thermal noise, subject motion, physiologically-induced field fluctuations, to name a few. In light of these challenges in qMRI, this thesis aimed to evaluate and improve the accuracy and/or precision of qMRI measurements. From a wide range of available qMRI measures, diffusion MRI and T1 relaxation time mapping were the focus of this thesis.

In diffusion MRI, the actual b-value played out on the scanner may deviate from the nominal value due to magnetic field imperfections (e.g., gradient nonlinearities, imaging gradient interactions, concomitant fields, eddy currents, and gradient miscalibration). In Chapter 2 of this thesis, an image-based approach for correcting voxel-wise b-value errors in diffusion MRI is proposed. This method acquires diffusion MRI data from a water phantom while monitoring the temperature of the phantom to estimate the effective b-value map for a specific sequence. Subsequently, this effective b-value map is used in the diffusion analysis to obtain more reliable diffusion-related parameters. The proposed method was tested in both phantom and in vivo experiments. The apparent diffusion coefficient maps of the homogeneous water phantom estimated using the effective b-value map showed the expected spatial uniformity as well as a marked improvement in consistency across diffusion directions. The b-value correction for the brain data resulted in 5.8 % and 5.5 % decrease in mean diffusivity and angular differences of the primary diffusion direction of 2.71°and 0.73° inside gray and white matter, respectively. Moreover, the mean diffusivity map became more homogeneous across the brain with the b-value correction. A higher impact of the b-value correction is expected for the gradient systems that are designed to generate an extremely high amplitude and/or slew rate, at the expense of gradient linearity.

Diffusion MRI is an inherently signal-to-noise ratio (SNR)-limited technique. Thus, increasing SNR is particularly important to achieve accurate and precise measurements of diffusion-related parameters. Chapter 3 of this thesis systematically compares SNR of a diffusion data set acquired with a spiral readout against those acquired with several variants of the conventional diffusion sequence with echo-planar-imaging (EPI) readouts. The SNR gains of spiral over full-Fourier EPI readouts in white matter were 48-96 % depending on the diffusion encoding scheme. Moreover, spiral readout provided higher SNR than partial-Fourier EPI, having 42-43 % SNR gain in white matter. Such SNR gains can be exploited in various ways. While maintaining the scan time, a diffusion data set with a better SNR can be obtained with spiral readouts. Another way is to reduce the scan time of spiral acquisition (by reducing the number of averages) while maintaining the same SNR level as EPI. For example, for the observed SNR gain of spiral over full-Fourier EPI with split gradient single spin-echo sequence (48 %), scan time can be shortened by a factor of 2.2 without any SNR loss. Finally, the expected SNR gain can be used to increase the resolution while maintaining the scan time and SNR. Again, for the SNR gain of 48 %, the isotropic resolution can be increased from 1.0 mm to 0.88 mm when spiral readouts are used.

Variable flip angle (VFA) method is a widely used T1 mapping method, capable of acquiring high-resolution whole-brain three-dimensional T1 map in a clinically feasible scan duration. As the VFA method relies on the relation between spoiled gradient-echo signal and flip angle, the exact knowledge of B1+ field is crucial. Thus, for the high field MRI (≥ 3T), the inclusion of the B1+ map has become a standard procedure in the VFA method. In chapter 4 of this thesis, the precision of T1 estimates as measured using the VFA method is established theoretically and experimentally with emphasis on the noise propagated from B1+ measurements. It was demonstrated, in the in vivo experiments, that improving the precision of the B1+ measurements significantly reduces the variance in the estimated T1 map.

In chapter 5 of this thesis, the intra- and inter-vendor reproducibility levels of the VFA T1 measurement are established at 3T. Although intra-scanner variability of voxel-wise T1 values was acceptably small, significant systematic bias was observed in T1 values between the two vendors. If not taken into account, this bias would reduce the statistical power in a group comparison study that involves scanners from different vendors, requiring a larger number of subjects to detect the same effect size in comparison to studies conducted with a single scanner. Therefore, a pilot study is necessary in the design phase of the study to ensure that the benefit of a multi-center study (that it can recruit a larger number of subjects more easily) outweighs the reduced statistical power caused by the higher variability.