Optimal Sampling Schedule For Model-based Quantification Of Chemical Exchange Saturation Transfer MRI

Chemical exchange saturation transform, is a new type of contrast imaging technology that can be used to detect small amount of specific molecules dissolved in water. Unlike traditional MR agents, CEST imaging base on chemical exchange between labile protons in different chemical environments with different chemical shift. CEST effects can be described by Bloch-Mc Connell equations modified with exchange terms.Generally, the sampling schedule for CEST imaging is to sample the whole investigated frequency offsets with even distribution, which is called evenly distributed sampling schedule(EDS). This sampling schedule had been proved to be inefficient because some of data collected are minimally informative. An optimal sampling schedule(OSS) was proposed by Y. K. Tee et al. Results showed that OSS had improved the accuracy of parameter estimation using model based least square curve fit.In this study, some research have been performed as follow: we proposed an improved algorithm for finding optimal sampling frequencies. Average sensitivity curves based on prior distributions of model parameters were calculated to describe the expect variance of z-spectrum while tiny changes occurred in model parameters. OSS for two-pool PARACEST and three-pool DIACEST were calculated by maximizing the determinant of Hessian matrix. Simulations were conducted to evaluate the performance of OSS include: the relationship between the performance of OSS and the distribution of true values of model parameters, the performance of OSS in different noisy scenarios and the performance of OSS when the number of samples decreased.This study have several contributions: first, optimal sampling schedules were proposed for two-pool PARACEST model and three-pool DIACEST model based on average sensitivity curves, the proposed OSS placed more samples around the chemical shift of CEST agent. Second, there is a strong connection between the performance of OSS and the distribution of true values of model parameters. Hessian matrix can describe this relationship qualitatively. Third, in the circumstances of this simulation, OSS show higher precision in parameter estimation than EDS in different noisy scenarios. Finally, the relative errors acquired by OSS are still far less than EDS method even in the very low sample number range. OSS is able to provide more accurate parameters comparing to EDS scheme in the same experimental time. Simulation shows that the number of samples should be kept above 60 for DIACEST and 50 for PARACEST to assure adequate precision.