Intravoxel incoherent motion diffusion-weighted MRI of the abdomen: The effect of fitting algorithms on the accuracy and reliability of the parameters.

Abstract

To evaluate the influence of fitting methods on the accuracy and reliability of intravoxel incoherent motion (IVIM) parameters, with a particular emphasis on the constraint function. Diffusion-weighted (DW) imaging data were analyzed using IVIM-based full-fitting (simultaneous fit of all parameters) and segmented-fitting (step-by-step fit of each parameter), each with and without the constraint function, to estimate the molecular diffusion coefficient (Dslow ), perfusion fraction (f), and flow-related diffusion coefficient (Dfast ). Computational simulations were performed at variable signal-to-noise ratios to evaluate the relative error (RE) and coefficient of variation (CV) of the estimated IVIM parameters. DW imaging of the abdomen was performed twice at 1.5 Tesla using nine b-values (0-900 s/mm2 ) in 12 health volunteers (6 men and 6 women; mean age: 30 years). The measurement repeatability of IVIM parameters in the liver and the pancreas was evaluated using the within-subject coefficient of variation (w CV). In simulations, full-fitting without the constraint function yielded the largest RE (P < 0.001 for Dslow and f; P ≤ 0.044 for Dfast ) and CV (P ≤ 0.033 for Dslow and f; P ≤ 0.473 for Dfast ) for IVIM parameters among all four algorithms. In volunteer imaging, full-fitting without the constraint function also resulted in the poorest repeatability for Dslow (w CV, 17.12%-65.45%) and f (w CV, 19.35%-42.84%) in the liver and pancreas, while the other algorithms had similar repeatability values (w CV, 4.05%-11.99% for Dslow and 9.65%-18.66% for f). Measurement repeatability of Dfast (w CV, 29.52%-85.01%) was the poorest among the IVIM parameters. For accurate and reliable measurement of IVIM parameters, segmented fitting or full-fitting with the constraint function should be used for IVIM-based analysis of DW imaging. 3 Technical Efficacy: Stage 2 J. MAGN. RESON. IMAGING 2017;45:1637-1647.