The reliability of parameters obtained by fitting of ¹H NMRD profiles and ¹⁷O NMR data of potential Gd³⁺-based MRI contrast agents

Synthetic variable temperature ¹H NMRD profiles and ¹⁷O NMR relaxation and shift data were generated with a model based on the Solomon-Bloembergen-Morgan and Freed theories and then fitted simultaneously or individually. The effects of the fitting procedure and of experimental uncertainties on the resulting best-fit parameters were investigated. The most reliable best-fit parameters were obtained when all data were included in a simultaneous fitting procedure. Fitting of only NMRD and/or ¹⁷O NMR data provided considerably less accurate best-fit parameters. Very large deviations from the values of the parameters used for the construction of the datasets were obtained due to the combined effects of uncertainties resulting from the fitting and from the data. For these fittings, the accuracy of the best-fit parameters appeared to be strongly dependent on the magnitude of synthetic parameters applied. For example, the accuracy of τ_M ^C was low around τ_M ^S = 10^-8 s. The parameters τ_V and Δ² are strongly correlated in fittings of only ¹⁷O NMR data. Consequently, only the ratio of these parameters can be evaluated in this way. The observations underline the need to reduce the number of adjustable parameters by constraining as many as possible of them at values obtained by independent techniques. The inaccuracies observed in these simulations come in addition to those caused by the inadequacy of the Solomon-Bloembergen-Morgan theory, particularly at low magnetic field strengths.