Cohort
Ethical approval for the study was granted by the West London & GTAC Research Ethics Committee (Ethics No. 07/H0707/105), and written informed consent for participation was obtained from all pregnant participants. The participants for the normal control cohort comprised of healthy pregnant volunteers, women who had a previous child with a confirmed abnormality, or who had had a suspected fetal abnormality on ultrasound not present on MRI, or a mild non-CNS abnormality. Fetal brain MRI was performed at the Robert Steiner MRI Unit in Hammersmith Hospital between November 2007 and May 2013. Delivery summaries were obtained for all participants and reviewed for delivery complications and physical signs suggestive of genetic syndromes.
Participants with a singleton pregnancy with normal fetal brain appearance for gestational age, as reported on MRI, were included in the study. Subsequent exclusion criteria constituted delivery complications, congenital malformations or maternal infection, chromosomal abnormality, inadequate MR image quality, and abnormal developmental examination at either 1 or 2 years of age. Fetal gestational age (GA) was estimated from a first trimester dating ultrasound scan.
A total of 128 fetuses had a normal brain appearance on fetal MRI and were selected and evaluated for inclusion in the normal control cohort. 20 cases were excluded in accordance with our exclusion criteria: poor image quality due to fetal motion (1), positive infection screening (9), delivery complications (2), chromosome 22 duplication (1), language delay at 2 years of age (2), intrauterine growth restriction (1), seizures at 6 months of age (1), and no delivery summary and no parental contact after birth (3).
In the remaining 108 fetuses, MRI was performed at a median age of 29.43 week GA (range 21.29–38.86 weeks) (Fig. 1). 17 fetuses were scanned twice during gestation (range 27.86–38.71 weeks), while 1 fetus was scanned three times at different gestational ages. The normal control cohort consisted of 83 healthy volunteers, 15 women who had a previous child with a confirmed abnormality, 5 who had a suspected fetal abnormality on ultrasound excluded on MRI, and 5 who had a fetal mild non-CNS abnormality (Table 1). Clinical details regarding these cases are presented in detail in the Appendix. In summary, 127 MR scans of 108 fetuses were included in the control cohort (57 males/51 females).
Table 1 Cohort demographics
Imaging
Neuroimaging and reconstruction
Fetal MRI was performed using a 1.5 T MRI System (Philips Achieva; Philips Medical systems, Best, the Netherlands) with a 32-channel cardiac array coil placed around the mother’s abdomen. The mother was positioned in a left lateral tilt, no sedation was used, and the total duration of the MR examination did not exceed 60 min. Maternal temperature was measured using a tympanic thermometer prior to and after the scan. In instances, when the maternal temperature was ≥37.5 °C, the scan was rescheduled. A complete fetal brain clinical examination was performed in transverse, sagittal, and coronal planes. T2-weighted Single Shot Turbo Spin Echo (ssTSE) was acquired using the following scanning parameters: TR = 15,000 ms, TE = 160 ms, slice thickness of 2.5 mm, slice overlap of 1.5 mm, and flip angle = 90°. 3D reconstructed images were constructed using Snapshot MRI with Volume Reconstruction (SVR), as previously described (Jiang et al. 2007; Kuklisova-Murgasova et al. 2012). In summary, data sets from multiple ssTSE image stacks were acquired in three orthogonal planes using overlapping slices (four transverse, two coronal, and two sagittal acquisitions). The fetal brain was oversampled to ensure the acquisition of complete data sets even with significant motion. Post-acquisition processing and registration of raw images was performed on Windows and Linux workstations (total duration 40 min). All scans were reviewed for image quality, and the slices corrupted by motion artefacts and loss of anatomical detail were excluded from the proceeding analysis. Image registration is performed to align all images obtained based on the assumption of a rigid body, of constant shape and size, performing an unknown motion. Images were registered onto a self-consistent anatomical space of the fetal brain (volume with least motion), and using a scattered interpolation approach, all measured voxel intensities are used to reconstruct the 3D fetal brain with an accuracy of 0.3 mm. The reconstructed 3D volumetric data sets have high resolution, high signal-to-noise ratio, and full brain coverage essential for reliable volumetric analysis. Visual analysis of all acquired images was performed by an expert radiologist to exclude additional anomalies and confirm appropriate appearance for gestation. The 3D fetal volumetric brain data were orientated into standard axial, coronal, and sagittal projections, and the voxel size was interpolated from a reconstruction voxel size of 1.18 × 1.18 × 1.18 mm to 0.2 × 0.2 × 1 mm to aid visual display and assist placement of anatomical markers.
Quantification analysis (3D)
Volumetric measurements were produced on the 3D reconstructions from semi-automatic segmentations using ITK-SNAP (version 2.2.0, University of Pennsylvania, Philadelphia, PA, USA) (Yushkevich et al. 2006) in a two-step process. Automatic segmentation of the different intracranial regions is based on image contrast while utilising user-defined thresholds. Following completion of the automatic process, editing of each segmentation was performed manually using a digital drawing tablet (Intuos, Wacom, Germany) to remove incorrectly labelled areas. Supratentorial brain tissue volume was defined as the brain tissue above the tentorium, i.e., excluding the brainstem, cerebellum, and cerebrospinal fluid (CSF) spaces (Fig. 2). Total ventricular volume was defined as the volume of both left and right lateral ventricles including the choroid plexus but excluding the third and fourth ventricles and cavum septum pellucidum and vergae (CSP) (Fig. 2). Lateral ventricle volume refers to the volume of each lateral ventricle. Laterality was established by the position of the fetal heart, stomach, and liver on MR images in a coronal plane (no fetuses had situs inversus or dextrocardia as assessed on antenatal ultrasound). Cortical volume represents the total cerebral cortical gray matter and was manually segmented in 75 scans (37 males/38 females and GA range 21.29–38.86) (Fig. 2). Cortical segmentation was only performed in a sub-group of our cohort (chosen to span gestation) due to the laborious nature of the current segmentation process (~6 h per segmentation). The total cerebellar volume measurement included both the cerebellar and the vermis volumes and excluded the fourth ventricle (Fig. 2). Extra-cerebral CSF included all intracranial CSF spaces surrounding the supratentorial brain tissue and cerebellum and including the interhemispheric fissure space but not any ventricular structure or the CSP (Fig. 2). The time required for the manual editing of the different structures varied and were as following for a 28-week-old fetus: supratentorial brain tissue (1 h), total lateral ventricles (10 min), cortex (6 h), cerebellum (15 min), and extra-cerebral CSF (30 min).
Quantification analysis (2D)
Linear measurements were performed on the 3D reconstructions using ImageJ (version 1.40 g, National Institutes of Health, Bethesda, MD, USA) and included the brain biparietal diameter and fronto-occipital length, skull occipitofrontal diameter and biparietal diameter, head circumference, transverse cerebellar diameter, extra-cerebral CSF, atrial diameter, and vermis height, width, and area. The brain biparietal diameter was measured in a transverse plane as the maximum brain width (Fig. 3a). The fronto-occipital length of each hemisphere was measured in a sagittal plane as the distance between the extreme point of the frontal and occipital lobes (Fig. 3b). The skull occipitofrontal diameter was defined as the maximum distance between the frontal and occipital skull bones and was measured in a transverse plane by placing the cursors in the middle of the bone hypo-intense area (Fig. 3c1). The skull biparietal diameter was defined as the widest diameter of the fetal skull measured in a transverse plane using the “outer edge to inner edge” technique (Fig. 3c2) (Salomon et al. 2010). The head circumference was measured in two different ways to reflect the different measuring techniques used in ultrasonography. First using the equation: head circumference = 1.62 × [(skull biparietal diameter) + (skull occipitofrontal diameter)] and second using the eclipse tool, option in ImageJ, surrounding the fetal skull (Fig. 3d) (Salomon et al. 2010). The transverse cerebellar diameter was defined as the maximum lateral cerebellar distance in the transverse plane (Fig. 3e). The linear measurement of extra-cerebral CSF was calculated using the following formula: (skull biparietal diameter) − (brain biparietal diameter). The atrial diameter was measured according to the guidelines of the International Society of Ultrasound in Obstetrics and Gynaecology (ISUOG 2007) at the level of the atrium (Fig. 3f). More specifically on MRI, the atrial diameter was measured on a slice, where both the posterior aspects of the basal ganglia and the third ventricle were visible. The cursors are placed inside the low signal intensity of the inner edge of the ventricular wall and perpendicular to the long axis of the ventricle. The vermis height, width, and area were measured in the mid-sagittal plane (Fig. 3g–i). The vermis height corresponded to the maximum superior–inferior length and the width to the maximum distance between the fastigium and the posterior part of the vermis in the mid-sagittal plane. The vermis area was calculated using a free-hand drawing tool.
Developmental assessments
All children were invited for a developmental assessment at 1 and 2 years of age to ensure that normative data produced represent typically developing children. Assessments were performed by a clinical psychologist or paediatric neurologist. The Griffiths Mental Development Scale (GMDS) assessment was performed at year 1, and the Bayley-III Scales of Infant and Toddler Development (Bayley-III) assessment were performed at year 2 (Bayley 2006; Huntley 1995). The GMDS is composed of five separate scales: locomotor, personal–social, hearing and speech, eye-hand co-ordination, and performance. Developmental Quotient (DQ) scores in the range of 88-112 and Sub-Quotient (SQ) scores of 84–116 were considered to represent typical development. Developmental delay was defined as a DQ score below 88 (<1 standard deviation) (SD) or an SQ score below 84 (<1SD). The Bayley-III assessment comprises of three separate scales, cognitive, language (expressive and receptive), and motor (gross and fine). Scaled scores of 7–13 and composite scores of 85–115 were considered within typical development. Developmental delay was defined as a scaled score below 7 (<1SD) or a composite score below 85 (<1SD). Children scoring within the developmental delay range at any time point where excluded from the cohort. Parents that were unable to attend a developmental assessment were sent a parental filled questionnaire [Age and Stage Questionnaires–III (ASQ-III)] assessing communication, gross motor, fine motor, problem solving, and personal–social skills (age range 1–66 months). ASQ-III is a set of age-specific questionnaires (age range 1–66 months) that serve as a developmental screening modality. Developmental delay was defined as a score <2SD and children scoring within the developmental delay range at any timepoint where excluded from the cohort.
Statistical analysis
Statistical analysis was performed using the SPSS software package version 17 (SPSS Chicago, IL, USA). Normality of distribution was assessed using the Shapiro–Wilk goodness-of-fit test and the Q–Q plots for each variable. Correlation between variables was assessed with the Spearman’s rank correlation coefficient (r). A confidence level of 0.05 was considered significant. The Bonferroni adjustment was applied. Multiple raters were involved in the study. All raters received extensive training to obtain reliability and achieve a percentage difference between 2D and 3D measurements of less than 5%. Intra- and inter-rater reliability was performed for all measurements and included image spanning the GA range of the cohort. Intra-rater and inter-rater variability was assessed using the intra-class correlation coefficient and Bland–Altman plots on SPSS. The intra- and inter-class correlation coefficients for all 2D and 3D measurements were 0.99 (p < 0.0001). The relative growth rate represents the percent volume gain relative to the average volume for each structure. This was calculated assuming linear growth to make our results comparable to previous studies. The relative growth rate represents the percent volume gain relative to the average volume for each structure and was calculated using the formula: Relative Growth Rate = [(lnV
2-lnV
1)/(GA2 − GA1)] × 100, where ln is the natural logarithm, GA1 and GA2 are the gestational weeks at a given GA range, and V
1 and V
2 are the volumes of the intracranial structure at timepoints GA1 and GA2, respectively (Hoffmann and Poorter 2002). Slope comparison was performed in the StatsDirect statistical software (version 3.0) using linear regression.
The 5th, 50th, and 95th centiles for the 2D and 3D measures of each intracranial structure were constructed, as described by Royston and Wright (1998). This approach is based on a linear regression that models both the mean and SD across GA. Briefly, least-square regression analysis was used to estimate the mean curves of each measurement as polynomial functions of GA. A quadratic line showed the best fit for all 2D and 3D measurements, except for cortical volume, where an exponential fit was representative. The scaled residuals were calculated, and polynomial regression analysis was performed to estimate an appropriate curve representing the SD. A straight line was adequate for the SD of all 2D and 3D measurements, except for the cortical volume SD, where a quadratic line was more appropriate. The 5th, 50th and 95th centiles were calculated using the equation: centiles = mean + K × SD, where the mean and SD were substituted by the appropriate curve as estimated above and K is the corresponding centiles of the standard Gaussian distribution.
Agreement analysis of 2D measurements performed prior and after SVR reconstruction
The SVR methodology ("Neuroimaging and reconstruction") may not be readily available in clinical environments, and 2D measurements will often be performed on non-reconstructed images. To ensure reliable use of the calculator using 2D measurements performed prior and after SVR reconstruction, we have performed statistical agreement analysis between 2D measurements performed on T2-weighted ssTSE (non-reconstructed images typically acquired in a clinical scan) and their corresponding reconstructed images. All 2D measurements were performed by the same rater. The analysis was performed in a sub-cohort of ten fetuses (GA 21.71, 22.86, 25.29, 26.14, 28.29, 29.57, 31.71, 32.71, 34, and 36 weeks). Cases were selected to span the gestational period studied and only included symmetrical non-rotated images.