A New Approach for Estimating Liver IVIM Fast Diffusion Coefficient in Two-Compartment Model and Fast Diffusion...

The intravoxel incoherent motion (IVIM) three-compartment model (3CM) potentially offers better classification for perfusion-rich pathologies. However, 3CM is often associated with less fitting stability than the two-compartment model (2CM). For liver intravoxel IVIM data, when the fit-starting b-value is adjusted from 2 s/mm² to b_i s/mm², PF_i, which indicates the perfusion fraction at b_i, gradually decreases as the initial fitting b-value b_i increases and eventually approaches zero. We define Y_i = ln [(1 - PF_i)/PF_i]. With a fitted line between Y_i and b_i, the intercept with the y-axis is the predicted value of Y₀ that follows the bi-exponential model. The PF₀ calculated from it eliminates the signals of other very rapid perfusion components and retains a more stable perfusion fraction measurement. The slope of the fitted line is (D_fast-D_slow). D_slow can be calculated by the standard IVIM model. Because the Y_i fitting assumes that the signal attenuation is bi-exponential, and the portion of the signal strength that exceeds the bi-exponential model at very low b-values is not used, the perfusion fraction detected is the fast decay perfusion component (F_fast) without the very fast decay perfusion component (F_vfast). In our study on healthy livers, Dataset-1 with 3.0 T data had 17 subjects scanned twice with a 10-12 days' interval, and paired scan-rescan had 13 subjects; Dataset-2 with 1.5 T data had 20 subjects scanned twice during the same scan session, and paired scan-rescan had 17 subjects. For the 2CM, IVIM parameters D_slow, D_fast, and PF_i were obtained with a nonlinear least square (NLLSQ) fit. Parameters obtained with segmented fitting were used for Y_i analysis. Results show that conventional fitting and Yi fitting derived very comparable IVIM parameters in mean value. D_fast obtained with Y_i fitting showed better scan-rescan stability than conventional segmented fitting. Perfusion fraction parameters showed more favorable scan-rescan stability with Y_i fitting.