Study design
This pilot study was conducted by retrospective enrolment of patients between January 2016 and July 2017. The mothers gave written informed consent for the use of their clinical data for research purposes prior to the examination. The study was approved by the regional ethics committee in Zürich (decision number: 2017-00167).
Patients
Foetal MRI, including within-session repetition of an IVIM sequence, was performed in 15 pregnancies (maternal age 33.7 ± 5.2 years (mean ± standard deviation [SD], range 24.6–40.8 years). In two cases, a follow-up foetal MRI was performed at two different time points during gestation with 2 and 2.5 weeks between scans and each follow-up scan included with-session repeated IVIM scans. These two measurements of the same cases were treated as independent samples, resulting in 17 IVIM datasets in total for repeatability analysis. The gestational age of the foetuses in the repeatability analysis was 26.3 ± 3.7 weeks (range 21–36 weeks). Foetal MRI was clinically indicated in all cases to rule out or confirm suspected pathologies detected during prenatal screening by ultrasonography. The clinical indication for MRI was isolated mild cerebral ventriculomegaly (n = 5), myelomeningocele (n = 8), sacrococcygeal teratoma (n = 1) and congenital bronchial atresia (n = 1).
To illustrate the IVIM technique in clinically relevant pathologies, we included two cases without within-session or across-gestation repeated IVIM data. These two foetuses were diagnosed: one with congenital diaphragmatic hernia (gestational age = 28 weeks) and the other with congenital cystic adenomatous malformation of the lungs (gestational age = 33 weeks).
MRI protocol
Foetal MRI was performed on two different clinical systems as part of the routine clinical assessment: 12 datasets with a 1.5-T Discovery MR450 unit, six datasets with a 3.0-T Discovery MR750 unit (General Electric Healthcare, Milwaukee, WI, USA). The assignment of the cases to an individual scanner was not controlled in the current study, but based on the availability of free scanner time. IVIM data were collected from January 2016 until March 2017 at the University Children’s Hospital Zürich. Pregnant women were examined in the supine position, feet first, and no contrast agents or sedatives were administered. In order to obtain optimal signal from the foetal head and body within the same session, the coil was readjusted to the position of the foetal structures investigated.
For each foetus, the IVIM imaging sequence was repeated twice with identical settings. The sequence relied on a DWI sequence optimised for foetal imaging, modified to accommodate more b-factors within a clinically feasible imaging time. Slices were positioned in the axial plane relative to the foetal brainstem for brain imaging and in the coronal plane relative to the foetal body for other organs and the placenta.
A dual spin-echo echo-planar sequence was used with echo time/relaxation time of 2200/75 ms, acquisition matrix 80 × 100, voxel size 2 × 2 mm, slice thickness 3 or 4 mm, slice gap 0.5 mm, number of slices 8–14, and 1 excitation. The tetra (tetrahedral) DW orientation scheme was used, which utilises four different combinations of x, y, and z diffusion gradients [20]. b-factor values were increased in 16 steps and one b0 image was acquired (b-factors: 0, 10, 20, 30, 40, 60, 80, 100, 150, 200, 300, 400, 500, 600, 700, 800 and 900 s/mm2). This scheme resulted in 64 DW images and one b0 image for each IVIM image series. The actual imaging time depended on the number of slices, which was adjusted to the size of the foetus and to the focus of the investigation, or specifically whether the brain (8–12 slices) or the whole foetal body and placenta (10–15 slices) were the most important organs for clinical decision-making. Imaging time per IVIM acquisition ranged from 1 min 40 s to 3 min 20 s.
IVIM post-processing
Post-processing was carried out using an in-house developed script written in BASH language for Linux. It utilised image processing algorithms from the Functional Magnetic Resonance Imaging of the Brain (FMRIB) Software Library (FSL) [21], C3D [22], and NIFTIREG [23] software packages for image registration and re-sampling. The image analysis script is available as online supplement material to this manuscript.
First, the raw IVIM data were viewed using the fslview command of the FSL software and the image frames with the most excessive subject motion were marked and removed from the analysis. This step was followed by a non-linear, free-form deformation-based registration of image frames with the reg_f3d command in the NIFTIREG tool, the registration steps of which are illustrated in Fig. 1. This image registration step used a fine deformation grid with a grid spacing of 6 × 6 × 6 mm for the low b-factor image frames and 12 × 12 × 12 mm grid for the high b-factor frames.
Volume of interest definition
After processing IVIM data, averaged images for low b-factors (b < 250 s/mm2) and high b-factors (500–900 s/mm2) were generated. Using the manual segmentation tool in the medical imaging interaction toolkit (MITK) [24], volumes of interest (VOIs) were placed over the central part of the placenta, on the foetal liver, lung parenchyma excluding the hili, kidneys bilaterally, cerebellum and brainstem, frontal or frontoparietal cortical mantle, and white matter of the frontal and parietal lobes (Fig. 2). All VOIs were drawn manually by one observer with four years of experience in foetal MRI. Three-dimensional interpolation in the MITK software was then used to smooth the borders of the delineated organ labels. For the kidneys and the placenta, better visual discrimination from surrounding tissues was achieved by delineating the VOIs by viewing the diffusion images with higher b-factors, while for the other structures we used the diffusion images averaged over lower b-factors.
IVIM model fitting
The IVIM parameters f, d, and D* were estimated based on the VOI-averaged signal intensity values to achieve a better signal-to-noise ratio (SNR). The analysis of diffusion and perfusion parameters with the IVIM model assumed two compartments without interactions [5]. A bi-exponential model (Eq. 1) was fitted in two steps on the averaged signal intensity over the VOIs:
$$ \frac{S}{S_0}=f\ast {e}^{-b{D}^{\ast }}+\left(1-f\right)\ast {e}^{- bd} $$
(1)
where S is the measured signal intensity, S
0 is the signal intensity without diffusion-weighting, d is the diffusion coefficient, D* is the pseudo diffusion coefficient, f is the perfusion fraction, and b is the b-factor.
First, the measurements were fitted for b-values higher than 250 s/mm2 to estimate the parameter d using a mono-exponential term. Then, f and D* were estimated keeping d fixed at the previously fitted value. The IVIM model fitting was carried out with the MITK diffusion toolkit.
Repeatability analysis
Repeatability of f, d, and D* over the repeated scans was measured as the test–retest variability:
$$ VAR\%={100}^{\ast}\frac{1}{N}\sum \limits_{i=1}^N\frac{\left| TES{T}_i- RETES{T}_i\right|}{\left( TES{T}_i+ RETES{T}_i\right)/2} $$
(2)
where N is the number of individuals and TEST
i
and RETEST
i
are the duplicate measurements for subject i.
Next, we tested whether the variability of the IVIM parameters is affected by possible confounds. Multiple, univariate analysis of variance (ANOVA) was carried out with the ‘General linear model’ module in SPSS v22.0 for Windows (Mathworks inc., Nattick, MA, USA). In this analysis, the test–retest difference (that is: \( \frac{\left| TES{T}_i- RETES{T}_i\right|}{\left( TES{T}_i+ RETES{T}_i\right)/2} \)) of f, d, and D* of each organ served as the dependent variable. We evaluated the effect of gestational age, maternal age, scanner field strength, and number of removed image frames on the test–retest difference of each IVIM parameter of the investigated organs. To reveal interactions between the IVIM parameters and the assumed confounds, we report results of the ANOVA tests. Values of p < 0.050 were considered significant.