Robust estimation of changes in transverse relaxation from magnitude MRI data: Application to vessel size imaging

Introduction: Transverse relaxation rates, R2 and R2* and the apparent diffusion coefficient (ADC) are all typically estimated from the exponential decay of MR signal magnitude as a function of either echo time (for R2 and R2*) or b-value (for ADC). Changes in R2 and R2* induced by ultrasmall paramagnetic iron oxide (USPIO) contrast agents, along with ADC, have recently been utilised to estimate vessel size index (VSI) in a range of contexts [1,2]. However, vessel size imaging tends to suffer from over-estimation of VSI when compared with gold-standard histological measurements [3]. In this study, a novel analysis method is described that offers a robust (accurate) approach to the estimation of ADC and of changes in R2 and R2* (ΔR2 and ΔR2*, respectively), based on the observation that noise in magnitude data is Rice-distributed [4]. This Bayesian maximum a posteriori (MAP) approach also offers benefits such as the provision of parameter uncertainties and estimates of the confidence that a change in R2 or R2* is significant, on a pixel-by-pixel basis. The approach is evaluated in vivo in PC3 orthotopic prostate tumours and compared with the conventional least-squares (LS) approach, which assumes noise to be normally-distributed.