Comparison of tri-exponential decay vs. bi-exponential decay and full fitting vs. segmented fitting for modeling liver intravoxel incoherent motion diffusion MRI

Abstract

Purpose:To determine whether bi-or tri-exponential models, and full or segmented fittings, better fit IVIM imaging signal of healthy livers. Materials and methods:Diffusion-weighted images were acquired with a 3-T scanner using respiratory-triggered echo-planar sequence and 16 b-values (0~800 s/mm 2 ). Eighteen healthy volunteers had liver scanned twice in the same session, and then once again in another session.Region of interest (ROI)-based measurements were processed with bi-exponential model full fitting and segmented fitting (threshold b-value = 80 s/mm 2 ), as well as tri-exponential model full fitting and segmented fitting (threshold b-value = 200 s/mm 2 ). Results:With all scans' signal averaged, bi-exponential model full fitting showed D slow =1.14, D fast =193.6×10 -3 mm 2 /s, and PF=16.9%, and segmented fitting showed D slow =1.03, D fast =56.7×10 -3 mm 2 /s, and PF=21.3%. IVIM parameters derived from tri-exponential model were similar for full fitting and segmented fitting, with a slow (D' slow =0.98×10 -3 mm 2 /s; F' slow =76.4 or 76.6%), a fast (D' fast =15.1 or 15.4×10 -3 mm 2 /s; F' fast =11.8 or 11.7%) and a very fast (D' Vfast =445.0 or 448.8×10 -3 mm 2 /s; F' Vfast =11.8 or 11.7 %) diffusion compartments. Tri-exponential model provided an overall better fit than bi-exponential model. For bi-exponential model, full fitting provided better fit at very low and low b-values compared with segmented fitting with the later tended to underestimate D fast , however, segmented method demonstrated lower error in signal prediction for high b-values. Compared with full fitting, tri-exponential segmented fitting offered better scan-rescan reproducibility.Conclusion: For healthy liver, tri-exponential modelling is preferred than bi-exponential modelling. For bi-exponential model, segmented fitting underestimates D fast , but offers more accurate estimation of D slow . 4 the best fit for IVIM signal decay in the liver over the 0-800 s/mm 2 range (6). More recently, Wurnig et al. investigated an extensive DW-imaging protocol including 68 b-values and computed apparent diffusion coefficient 'spectra', and demonstrated the presence of a third component of diffusion in liver and kidney (7). However, till now tri-exponential decay model has not been studied at individual subject level with clinically applicable examination set-up. With 50 IVIM diffusion MR scans from 18 healthy volunteers, this study analyzes the fitting accuracy of biexponential vs. tri-exponential models, and full fitting vs. segmented fitting methods. Fitting residual error of predicted MR signal vs. measured MR signal with the four fitting approaches, as well as scan-rescan repeatability/reproducibility at individual subjects' level, are compared. An analysis of the error distribution in model signal prediction can offer insight into why triexponential model offers better fit than bi-exponential model for liver diffusion signal decay. The IVIM parameter physiological values associated with each fitting approach are also presented in this paper....

High performance of intravoxel incoherent motion diffusion MRI in detecting viral hepatitis-b induced liver fibrosis

“…Our this study, using a new set of healthy volunteers (n=26) and patients' data (n=16), provides a confirmation of our previous report (15). The value of IVIM parameters depend on the number and distribution of b-value series, as well as the threshold b-value when segmented fitting is applied (31)(32)(33). Moreover, it has been noted that the dependence of PF, Dslow, and Dfast on threshold b-value differs between healthy livers and fibrotic livers, with the healthy livers showing a higher degree of dependence (23).…”

“…The same as our last study (15), among the three IVIM parameters, PF also showed best diagnostic performance in this study. According to previous estimation, Dfast should be the most sensitive parameter for liver fibrosis evaluation (14,15,23); however, precise measurement of Dfast is difficult and this current study did not include sufficient very low b-values for precise Dfast quantification (33,34). More b-values and applying an optimized b-value distribution shall be able to reduce errors in IVIM parameter estimation.…”

“…One more limitation of the study is that our volunteers were on average younger than the patients. Certainly, the b-value distribution in this study is not ideal, at this stage we would think that adding very low b-values such as 3, 5, 8 10 s/mm 2 etc would be valuable for better quantification of both Dfast and PF (33,34). With IVIM imaging protocol of more very low b-values and better image post-procession, diagnosis of early stage fibrotic liver is likely to have increased reliability.…”

“…The combination of our these two studies may suggest the appropriateness to analyze IVIM data without b=0 s/mm 2 image. Whether to densely sample very low b-values (such as many b-values below 10 s/mm 2 ) will allow more precise fitting of very fast motion compartments require further studies (33).…”

“…With IVIM imaging protocol of more very low b-values and better image post-procession, diagnosis of early stage fibrotic liver is likely to have increased reliability. Other approaches for improved data post-processing may include motion correction, de-noising as well as better segmentation to statistically remove ill-fitted pixels prior to ROI analysis, and employ better fitting strategies (33). It is also likely that a multi-parametric approach will have even better accuracy for evaluating the spectrum of chronic liver disease (37)(38)(39).…”