Speaker
Description
Diffusion (D), perfusion fraction (PF), and pseudo-diffusion (D) can be obtained by IntraVoxel Incoherent Movement (IVIM) fitting. Segmented fitting is the standard procedure to fit this bi-exponential decay model [1]. It is based on 2 steps; 1st only intra- and extracellular diffusion is assumed for high b-values (larger than 180 mm2/s), thus the D and PF are obtained by linear regression of the logarithmic signal decay, 2nd the D is fitted from the exponential decay based on the previously fitted parameters. Other models based on ADC images can also provide the D and PF information based on IVIM approximation with only one linear regression [2]. Both linear regression-based models contain valuable residuals information for testing the robustness and to improve the fitted parameters. In addition, non-Gaussian diffusion anisotropy can be introduced to these models assuming kurtosis, although a quadratic dependence should be considered.
In silico Matlab simulations are performed to obtain the signal of a bi-exponential decay model. Rician noise is added to the signal to analyze the SNR dependence: 9, 15, 23, 48, 63 SNR. Prostate literature values of D, PF, and D* are then used to simulate the signal decay. After a study of the accuracy and precision of the method by residual weight and model classification, real patient images are used with the improved algorithm. Other techniques like Machine Learning, may be used for parameterization improvement.
[1] Koh et al. “Intravoxel Incoherent Motion in Body Diffusion-Weighted MRI: Reality and Challenges” AJR:196, June 2011
[2] Pang et al. “Intravoxel Incoherent Motion (IVIM) MR Imaging for Prostate Cancer: An Evaluation of Diffusion Coefficient and Perfusion Fraction Derived from Different b-Value Combinations” Magn Reson Med. 2013 Feb;69(2):553-62