However, this kind of sampling schedule will contain samples that are minimally informative to the parameters of interest. For example, Clabile changes mainly affect saturation frequencies near the chemical shift of the exchangeable protons (around 1.9 ppm in this

study). Recently, an optimal sampling schedule (OSS) [40] was introduced to maximize the information for the parameters of interest from the measured data. OSS selects the saturation frequencies based on the parameter sensitivity functions which describe how sensitive the data are to changes in Pexidartinib the parameter values at a particular saturation frequency. When an OSS was optimized for ωw, Mlabile0 and Clabile, the algorithm proposed a schedule that sampled repeatedly around the water

center frequency and the chemical shift of the exchangeable protons with minimal or no samples at the other frequency offsets. By doing so, better signal to noise ratio Target Selective Inhibitor Library data are achievable, resulting in an improvement in the accuracy of the important parameters estimated from the model fitting. The results of this study, namely those in Figs. 1 and 4a, indicate that the predominant differences between the pulsed and continuous z-spectra occur around the two resonances which coincide with the frequency offsets most sampled by the OSS. This might imply that quantitative analysis of data acquired using pulsed-CEST with an OSS strategy may not be feasible with the continuous approximation and in this case the discretization method has to be used. In practical data analysis scenarios, the results in Fig. 2 indicate that the number of discretized segments required by the discretization Florfenicol method varies according to the pulsed parameters used and could be reduced from the benchmark (1024 segments) to minimize the computational cost. Previously, analysis has been performed by discretizing each pulse into 64 [30] or 512 [25] segments. The computation time required to calculate a spectrum using 512 segments per pulse was roughly 16 times (9.8 min/0.629 min) longer than 32 segments per pulse used in this study and 4 (9.8 min/2.483 min)

times longer than the largest discretization needed for the range of pulsed parameters simulated. The computational time reduction above was recorded from an Intel Xeon CPU E5520 @ 2.27 GHz with 8G of RAM. When discretized model fitting, which requires iterative calculation of the magnetization, is applied, using a smaller number of discretized segments is especially important as it will result in a substantial reduction in computational cost. Despite the reductions in computational costs afforded by the reductions in the number of discretization required in practice, analysis of pulsed-CEST data using a discretized pulse train is still high compared to the continuous equivalent (a few seconds to calculate a spectrum per iteration).