Model-based deconvolution for DSC-MRI: A comparison of accuracy, precision, and computational complexity of parametric transit time distributions

Abstract

AbstractObjectDynamic susceptibility contrast MRI (DSC-MRI) is the current standard for cerebral perfusion estimation. Model-dependent approaches for DSC-MRI analysis involve assuming a parametric transit time distribution (TTD) to characterize the passage of contrast agent through tissue microvasculature. Here we compare the utility of four TTD models: namely, skewed-Gaussian, gamma, gamma-variate, and Weibull, to identify the optimal TTD for quantifying brain perfusion.Materials and MethodsDSC-MRI data were acquired in nine subjects at 1.5T, and normal-appearing white- and gray-matter signals were assessed. TTDs were compared in terms of: goodness-of-fit, evaluated using RMSE; noise sensitivity, assessed via Monte-Carlo-simulated noisy conditions; and fit stability, quantified as the proportion of total fits converging to the global minimum. Computation times for model-fitting were also calculated.ResultsThe gamma TTD showed higher fit stability, shorter computation times (p<0.008), and higher robustness against experimental noise as compared to other models. All functions showed similar RMSEs and the parameter estimates (p>0.008) were congruent with literature values.DiscussionThe gamma distribution represents the most suitable TTD for perfusion analysis. Moreover, due to its robustness against noise, the gamma TTD is expected to yield more reproducible estimates than the other models for establishing a standard, multi-center analysis pipeline.