Estimation of pseudo-diffusion coefficient D* using different settings of low b-values in liver IVIM imaging
Meng-Chieh Liao1, Cheng-Ping Chien1, Shih-Han Hung1, Feng-Mao Chiu2, and Hsiao-Wen Chung1

1Graduate Institute of Biomedical Electronics and Bioinformatics, National Taiwan University, Taipei, Taiwan, 2Philips Healthcare, Taipei, Taiwan


The pseudo-diffusion coefficient (D*) in the liver estimated using intravoxel incoherent motion (IVIM) MRI currently suffers from inconsistent values reported in the literature. This study investigated the effect of low b-value settings on the estimation of D*. Data from healthy subjects with sixteen b-values were analyzed, with b-values of 0, 5, 10, and 15s/mm2 selectively removed and D* computed using a bi-exponential model. Results show progressive increases in D* estimations, with difference in values by a factor of two, which strongly suggest that the IVIM signals in the low b-value range do not obey single exponential decaying behavior.


Intravoxel incoherent motion (IVIM) is a technique to quantitatively distinguish the parameters of diffusion (true diffusion coefficient, D) and perfusion effects (pseudo-diffusion coefficient, D*)[1]. Typically in the liver, D* is believed to be an order of magnitude greater than D. Therefore, signal attenuation in the region of low b-value (b<20s/mm2) is dominated by perfusion effects, meaning that the signal at low b-values has a great effect on calculating D* value. According to previous literatures, however, the reported D* value in liver IVIM MRI vary to a great extent[2]. Therefore in this study, we aim to investigate the effects of low b-value settings on the estimation of D* in human liver IVIM MR imaging.


In this IRB approved study, 4 subjects (M/F 3/1, mean age 22.79±0.78 y) underwent IVIM diffusion-weighted imaging on a Philip 3T scanner using sixteen b values (0, 5, 10, 15, 20, 30, 40, 50, 60, 70, 80, 90, 100, 200, 400, 800s/mm2) with navigator to reduce influences from respiratory motion. Scanning parameters were as follow: FOV=340x340mm, TE/TR=45/1092.1ms, thickness=7mm, matrix size=128x128, NEX=6. ROI (diameter 1cm) was drawn on the right hepatic lobe (Fig.1), avoiding large hepatic vessels, bile ducts, and cardiac motion artifacts. Signals as a function of b-value were used to derive three parameters in IVIM equation by nonlinear least squared (NLLS) fitting: $$$Sb= S0[(1-f) ∙ e^{-bD} + f ∙ e^{-bD^*}]$$$--[1], where Sb and S0 represent the signal magnitude at b-factor equals b and 0s/mm2 respectively. Perfusion fraction is denoted by f, D is the diffusion coefficient, and D* is the pseudo-diffusion coefficient caused by perfusion. All data processing was carried out by self-designed Matlab scripts.

Each subject underwent the same scanning protocol twice, and two ROIs were selected from each volunteer. Thus, four sets of data were obtained in each subject, making the total number of data sixteen. Effects of low b-value settings were evaluated by removing data points corresponding to specific b-values. Four schemes were compared. First, all sixteen b-values were included (contained four very low b-values which were b=0,5,10,15 s/mm2); Second, b=5 s/mm2 was eliminated (i.e., b=0,10,15 s/mm2 were included); Third, b=5 and 10 s/mm2 were eliminated (i.e., b=0,15 s/mm2 were included); Last, b=15 s/mm2 was further eliminated (only b=0 s/mm2 preserved).


Four fitting cures are shown in Figure.2 which stand for four settings with different number of low b-values. Figure.3 shows the four curves magnified for the range of b<20s/mm2, showing substantially different calculated D* values as 0.1093, 0.0742, 0.0627, 0.0520mm2/s when including four, three, two, and one low b-values, respectively. As more low b-values were included for curve fitting, a larger estimated D* value was obtained. Figure.4 shows the distribution of D* value for all the sixteen ROIs, demonstrating similar behavior of D* dependency on the number of low b values included. In comparison, estimations of the true diffusion coefficient D (0.0011-0.0013mm2/s) and perfusion fraction f (10-30%) were relatively stable irrespective of the number of low b values included for each subject (data not shown).


The results showed an increasing tendency of D* as one selects more number of low b-values for analysis. The numerical values for D* range from 0.037-0.083mm2/s when including only b=0, to 0.081-0.165mm2/s when including b=0,5,10,15 s/mm2. The factor-of-two difference in D* strongly suggests that the IVIM signals within the low b-value range does not obey simple single exponential decaying behavior for the liver. The conventional bi-exponential model for IVIM liver imaging therefore may need further modifications in order for IVIM MRI to be useful for liver perfusion estimation.


I would like to acknowledge my indebtedness to my advisor, Mr. Chung, who has given me his constant help, read the manuscript with great care and offered me invaluable advice and informative suggestions. My thanks go to all members in MD704, who are always kind and patient in helping me to search for useful materials relevant to my study. My thanks also go to all staff in Taipei Beitou Health Management Hospital, who have taught me about operating procedures during I conducted the experiment. Finally, my sincere thanks should go to my family members who have been pouring out their care, support and encouragement to me.


1. Le Bihan et al., Radiology, 161:401–407 (1986).

2. J.-P. Cercueil ·J.-M. Petit et al., Eur Radiol, 25:1541–1550 (2015).


Figure.1. Diffusion-weighted image with a ROI on the right hepatic lobe ( b=100s/mm2)

Figure.2 Plots of Biexponential decay curve using different number of very low b-value(red line contains four very low b values; blue line contains three very low blues, and so on)

Figure.3 Fitting curves in the region of very low b-value(b<20s/mm2)

Figure.4 Boxplot shows D* value in different number of very low b values. D* value increases as selecting more number of very low b values.

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)