Errors of a controlled Time-Concentration Curve fitting in CT perfusion uniformly affect perfusion values computed with two different methods

Purpose: Assessing to what extent the errors on the Time-Concentration Curves (TCCs) fitting in Computed Tomography perfusion (CTp) affect voxel-based perfusion parameters. Materials and Methods: 21 patients with colorectal cancer, underwent at diagnosis axial CTp examinations (60 samples, every 1s during the first 30s, every 3s after) of normal liver. The TCCs of a selected Region of Interest (ROI) were fitted using an in-house algorithm and a standard Gamma-Variate (GV) model, after TCC denoising. Blood Flow (BF) values were computed on the GV-fitted TCCs according to Maximum Slope (MS) and Deconvolution (DV) methods, assuming a dual-input one-compartment model. Fitting errors, measured in Hounsfield Unit (HU), were assessed for Gaussianity (t-test, p-value<0.05), and variances for the different BF values, to investigate possible relationships with BF values. Results: In all the 21 examinations, the fitted GV’s showed a Gaussian error distribution (p-value=10-5), with an averaged range of errors of -3÷+7HU. Moreover, errors resulted independent from BF values and, for each patient, uniformly distributed over the whole range of BF values, with same variances, for MS and DV. Finally, the voxel-wise error mean is even spatially uniform, meaning that all voxels underwent the same averaged fitting error. Conclusions: The fitting procedure setup allowed achieving errors 1) coming from a random process, 2) independent from BF values, and 3) with same uncertainties, 4) spatially uniform. These all are the features needed to achieve repeatable and reproducible perfusion parameters, and this fitting procedure finally promotes the theoretical reliability of the CTp parameters.