Rapid Estimation of IVIM Pseudo-Diffusion Fraction with Correction of TE Dependence
Neil Peter Jerome1, Matthew R Orton1, Thorsten Feiweier2, Dow-Mu Koh3, Martin O Leach1, and David J Collins1

1CRUK Cancer Imaging Centre, Division of Radiotherapy & Imaging, Institute of Cancer Research, London, United Kingdom, 2Siemens Healthcare GmbH, Erlangen, Germany, 3Department of Radiology, Royal Marsden Hospital, London, United Kingdom


The biexponential IVIM model of diffusion does not account for distinct T2 values for the two components, commonly interpreted as blood and tissue, leading to a TE dependence of the pseudo-diffusion volume fraction parameter f. In this volunteer study, the addition of a small number of DWI scans at different TEs allows for fitting of an extended T2-IVIM model, returning TE-independent estimations of liver f (18.26±7.3 % compared to 27.88±6.0 % from conventional IVIM fitting), and T2s of 77.6 ± 30.2 and 42.1 ± 6.8 ms for pseudo- and true diffusion compartments, respectively.


The two-compartment IVIM diffusion model proposed by Le Bihan (1) is commonly used for DWI studies in the body. In this model, compartments are taken to represent pseudo-diffusion and true diffusion, which may in turn represent vascular and tissue fractions. Standard diffusion-weighted imaging (DWI) protocols are acquired at a single (usually minimum) TE and incorrectly assume a single (apparent) T2, thus causing the observed pseudo-diffusion fraction f to be dependent on the TE chosen (2). Distinct transverse relaxation constants for these components (T2p and T2t for pseudo- and true diffusion compartments, respectively) modify the standard IVIM model (Eq. 1) for echo time (TE) dependency (Eq. 2), where f is pseudo-diffusion fraction, D and D* are true and pseudodiffusion coefficients, T2p and T2t are the transverse relaxations constants for pseudo- and true diffusion compartments, and $$${S_{eff}} ={S_0}.\exp\left(\frac{-TE}{T_{2apparent}}\right)$$$:

$${S_{b}} ={S_{eff}}.\left[ f.\exp\left(-b.D^*\right) + \left(1-f\right).\exp\left(-b.D\right)\right]$$ Eq 1 (standard IVIM model)

$${S_{b,TE}} ={S_0}.\left[ f.\exp\left(\frac{-TE}{T_2p}\right).\exp\left(-b.D^*\right) + \left(1-f\right).\exp\left(\frac{-TE}{T_2t}\right).\exp\left(-b.D\right)\right]$$ Eq 2 (extended T2-IVIM model)

We present additional measurements at low b-values with increased TE as a method of deriving an estimate of f that is independent of TE, a parameter not commonly fixed in clinical MR studies. A b-value of 50mm-2s is sufficient to remove the pseudo-diffusion component in the liver (3), with the assumption that associated signal decay due to true diffusion is small (<5% for D of 1x10-3mm2s-1 in the liver). Full sampling of the b-value/TE space is challenging (4), however a clinical timeframe acquisition is able to give a TE-independent estimation of f that may be more accurate, and thus clinically useful and sensitive to modulation of pseudo-diffusion fraction, as well as providing native estimations of T2.


Volunteers (n=6) underwent coronal free-breathing DWI of the abdomen using a MAGNETOM Avanto 1.5T scanner (Siemens Healthcare, Erlangen, Germany), acquired twice (24hr separation) using a prototype sequence with the following parameters: diffusion delays δ 16.0ms and Δ 20.2ms, 3-scan trace monopolar diffusion scheme, TR 4000ms, FOV 380x380mm2, 16x5mm slices, matrix 128x128 (interpolated to 256x256), bandwidth 1628 Hz/pixel, SPAIR fat suppression, 7/8 partial Fourier, iPAT factor 2, and 12 averages. Seven b-values (0,10,50,100,200,400,800 mm-2s) were acquired at (minimum) TE 62ms (total 15 minutes), with three additional b-values (0,10,50 mm-2s) acquired at TE 80 and 100ms (additional 10 minutes). A region of interest (ROI) was drawn for a single slice over the liver for each volunteer (Figure 1a); fitting of the standard IVIM and extended T2-IVIM model was performed for mean ROI signal in each unregistered image (b-value and signal average), and voxel-wise in one volunteer, using custom MATLAB routines. The repeat measures coefficient of variation was calculated for each parameter across the cohort using log-transformed values.


Representative ROI and voxel-wise f values for the two models are shown in Figure 1; parameters derived from the IVIM and T2-IVIM models are given in Table 1; the TE-independent f is substantially smaller (mean 35±15% decrease, p=0.002, paired t-test; Figure 2), while both D and D* from the T2-IVIM model are comparable to conventional IVIM values. The T2t value returned for the liver tissue compartment is consistent with literature (5), although the T2p returned (77.6±30.2ms) was substantially lower than literature (5) (290ms). In this study, the CoV for D and D* was small in both models (see Table 2), which for D is consistent with previous work but lower for D* than previously observed in tumours (6). In the T2-IVIM model, the CoVs for f and T2p were large (>20%), which indicates the inherent difficulty in separating these two parameters.


DWI studies commonly use a minimum TE dictated by gradient hardware and diffusion scheme. The dependence of the IVIM pseudo-diffusion parameter f on TE may limit its clinical utility; when using the standard IVIM model, failure to account for distinct T2 values for the two volume fractions may bias estimates of f and lead to misinterpretations of observed changes. An extended T2-IVIM model that includes T2 for each component can be used to derive TE-independent (TE = 0ms) estimations of IVIM parameters (4). Full sampling of TE/b space is limited by minimum TEs and SNRat larger TE/b combinations, but T2-IVIM parameters derived from the addition of a small number of TE/b combinations (minimum 2 b-values at 1 extra TE, additional ~20% scan time) to a standard IVIM protocol may provide useful estimates for TE-independent f, D, D*, and complementary information from T2p, and T2t (7), within clinical DWI examination times.


CRUK and EPSRC support in association with MRC & Dept. of Health C1060/A10334, C1060/A16464 and NHS funding to the NIHR Biomedical Research Centre and the Clinical Research Facility in Imaging. Martin O Leach is a senior NIHR investigator. Neil P Jerome is funded by Imagine for Margo.


1. Le Bihan D, et al. Radiology; 1998;168:497 2. Lemke A, et al. Magn. Reson. Med 64:1580–1585 (2010) 3. Jerome NP, et al. Proc. Intl. Soc. Mag. Reson. Med. 21 (2013), #2201 4. Orton MR, et al. Submitted to ISMRM 2016 5. Stansiz GJ, et al. Magn. Reson. Med 54:507-512 (2005) 6. Miyazaki K, et al. Eur. Rad 25:2641-50 (2015) 7. Cheng L, et al. Proc. Intl. Soc. Mag. Reson. Med. 23 (2015), #2878


Figure 1: a) Example b100 image showing liver ROI, b) pseudo-diffusion fraction maps from (left) standard IVIM and (right) T2-IVIM model. c) Difference map of estimated f, and d) histogram of differences (voxel-wise fitting performed only for illustration; presented results are fitting of ROI mean values).

Figure 2: Ladder plot showing the estimated pseudo-diffusion fraction before and after removal of TE dependence using additional b/TE data (each point average from 2 scans). Grey lines are individual volunteers; black line is overall mean (n=6).

Table 1: Parameters for conventional IVIM and extended IVIM (with additional TEs) models using liver ROI in 6 healthy volunteers. The extended model estimates T2 values in pseudo- and true diffusion compartments (T2p and T2t, respectively)

Table 2: Coefficient of Variation (%) of parameters derived from IVIM and T2-IVIM modelling. In the T2-IVIM model, larger (>20%) CVs for the parameters T2p and f associated with the pseudo-diffusion compartment.

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