Subjects were recruited prospectively into a longitudinal case-control study in which volumetric MRI and 1H-MRS were performed at a single prenatal time-point, in the neonatal period prior to any cardiac surgery, and at 1 year of age. This report includes data from the prenatal evaluation.
The inception cohort included pregnant women undergoing fetal echocardiography at Children's Hospital Boston between September 2007 and April 2008. Mothers of fetuses with confirmed CHD were recruited as cases. Normal controls were recruited from healthy pregnant volunteers and from pregnancies with a normal fetal echocardiogram performed for a family history of CHD or for suspected CHD in the current pregnancy. The decision to approach a mother about participation was left to the discretion of the attending fetal echocardiographer. Exclusion criteria for both cases and controls included gestational age greater than 36 completed weeks; multiple gestation pregnancy; congenital infection; gestational diabetes; maternal contraindication to MRI; fetal ultrasound findings of dysmorphic features, dysgenetic brain lesions, or anomalies of other organ systems; prenatally documented chromosomal abnormalities; or inadequate MRI data quality due to excessive fetal motion. If structural brain anomalies were identified on fetal or neonatal MRI, the anomalies were reported, but these fetuses were not included in the analysis.
In cases, the MRI was typically scheduled on the same day as a follow-up fetal echocardiogram, usually several weeks after enrollment. In most control fetuses, the MRI was also performed several weeks after enrollment, but the echocardiogram was not repeated. Thus, the duration between the study echocardiogram and MRI was variable, and tended to be longer in controls than cases.
Written informed consent from mothers was obtained according to a protocol approved by the Children's Hospital Committee for Clinical Investigation.
Complete anthropometric ultrasound and anatomic and Doppler echocardiographic studies were performed in all fetuses, and reviewed by two attending fetal echocardiographers (WT, DWB). Aortic and pulmonary arterial flows were calculated using pulsed-wave Doppler as: flow = valve area (computed from valve diameter) × velocity-time integral (area under the curve of the Doppler signal) × heart rate. Combined ventricular output was computed as the sum of aortic and pulmonary flow, and indexed to estimated fetal weight. Middle cerebral and umbilical arterial flow velocities (systolic, diastolic, and mean) were measured using pulsed-wave Doppler, and pulsatility and resistance indices were calculated. Patients were categorized by diagnosis as having hypoplastic left heart syndrome (HLHS) or variants thereof, d-transposition of the great arteries (TGA), pulmonary atresia, other cardiac diagnosis, or normal.
Estimated GA was based on maternal dates or first-trimester ultrasound measurement, if available. Fetal weight was estimated by Hadlock's formula [22
], and fetal weight percentile was calculated from equations reported by Doubilet et al. [23
Fetal Brain MRI
MRI scans were performed using a 1.5 Tesla scanner (Achiva, Philips Medical System, Netherlands) and a 5-channel phased array cardiac coil. Multiplanar single shot turbo spin echo (SSTSE) imaging was performed (TEeff
=120 msec, TR=12500 msec, 0.625 signal averages, 330 mm field of view, 2 mm slice thickness, no interslice gap, 256 × 204 acquisition matrix, acquisition time 30-60 seconds). 1
H-MRS was performed using a single voxel technique and the spectra were acquired using a point resolved echo spectroscopic (PRESS) sequence (TE=144 msec, TR=1500 msec, 64 signal averages, scan time approximately 3 minutes). Maternal sedation was not administered. The 4.5 cm3
) volume of interest was placed within the cerebral hemisphere at the level of the centrum ovale, as described by Girard et al. [24
], to obtain metabolic information about the normal developing white matter ().
Fetal MR image reconstruction. Raw fetal MRI images (column A), fetal intracranial cavity extraction (column B), and fetal MRI images following intensity correction and slice alignment (column C).
Fetal MRI studies were analyzed by an attending neuroradiologist (RLR) who was blinded to all clinical data. Abnormal patterns of brain development and maturation, and the presence of encephaloclastic lesions, were documented.
Quantitative volumetric MRI analysis
Post-acquisition processing was undertaken on a Linux workstation. MR images of the fetal intracranial cavity were manually masked to extract the fetal brain from the intra-uterine tissue. Each image was corrected for non-uniformity intensities induced by radio frequency and shading artifacts using the Non-parametric Non-uniform intensity Normalization (N3) tool [25
]. In order to minimize the effects of image degradation secondary to fetal motion, an iterative slice-by-slice registration was used, which normalized the intensity of each slice [26
]. The middle brain slice, defined as the slice at the mid-point of the brain, was used as a reference, and the global median intensity of each subsequent slice was normalized to the reference using a single scale intensity normalization with a non-linear registration approach. A Gauss-Seidel iterative schema was then applied to register the slices to a weighted average of the considered slice and the two neighboring slices. The rigid-body registration of the slices reached a steady point when it was identical to the previous registration ().
Manual segmentation of Fetal MR coronal images. 3-D manual segmentation of intracranial cavity volume (red), total brain volume (blue) and cerebral spinal fluid (yellow) using coronal images.
After correction for fetal motion, coronal slices were manually segmented using MINC software (www.bic.mni.mcgill.ca/software/minc
) to measure intracranial cavity volume (ICV), total brain volume (TBV), and cerebrospinal fluid volume (CSF). ICV included CSF and cerebral, cerebellar, and brainstem parenchymal volumes. TBV included cerebral, cerebellar and brainstem parenchymal volumes but excluded intraventricular and extraventricular CSF (). CSF was derived by subtracting TBV from ICV. Volumes were determined by counting the number of voxels in the segmentation, and multiplying by the volume of each voxel; i.e., the volume of each slice equaled the product of slice thickness and the measured coronal area. Total ICV, TBV, and CSF were computed as the sum of volumes measured on each coronal slice.
In fetal MRS, the voxel of interest was placed in the cerebral hemisphere at the level of the centrum ovale.
All measurements were made by a single operator (CL) who was blinded to clinical data. Each coronal area was traced twice, and the average of the two measurements was used for volume calculations. Intra-observer correlation coefficients were determined for measurements of ICV (n=481), TBV (n=458), and CSF (n=481) on all individual slices from 10 MRI studies, and were ≥0.997 in all cases. Intra-observer correlation coefficients were also determined for the summed volumes ICV, TBV, and CSF from 15 MRI scans, and were 0.98 for ICV, 0.96 for TBV, and 0.95 for CSF.
MR spectroscopy analysis
Four metabolites were assessed by 1
H-MRS: N-acetyl aspartate (NAA), choline (Cho), creatine (Cr), and lactate. Specifically, NAA, Cho and Cr peaks were identified and measured, and the presence or absence of lactate (identified by an inverted doublet peak around −1.33) was recorded. Cho:Cr and NAA:Cho ratios were calculated when MR spectra satisfied a reliable fit with a standard deviation <20% for each metabolite [27
The primary hypothesis was that GA- and fetal weight-adjusted TBV in late gestation would be smaller in CHD fetuses than in controls. A secondary hypothesis was that GA- and weight-adjusted measures of cerebral metabolism would be abnormal in CHD fetuses relative to controls. Comparison of demographic and baseline characteristics between cases and controls was performed with chi-square analysis or Wilcoxon rank sum test. To assess for differences in volumetric and 1H-MRS measures between groups, linear regression was performed and models were fit with the appropriate interaction terms and type III sum of squares testing. The equality of regressions was tested between groups using a single F test; comparison of Y-intercepts (25 weeks GA for GA-adjusted analyses) was also performed. Multivariable models were built after consecutive analysis of individual independent variables adjusted for GA and weight percentile. For analysis of each of the volumetric outcomes, independent variables included presence of CHD, CHD diagnostic category (HLHS, TGA or pulmonary atresia, other CHD or control), presence or absence of antegrade flow in the aortic arch, percentage of combined ventricular output through the aortic valve, and cerebral artery flow indices. For analysis of quantitative metabolic outcomes, the presence of lactate was also included as an independent variable and models were adjusted for GA along with TBV rather than weight percentile. All statistically significant associations are noted. Outcome variables were presented graphically with linear or quadratic curve fitting. In regression models, coefficients of determination (R2 values) reported in the text reflect the goodness of fit for each independent variable included in the model, or for the complete model if no independent variables were included other than GA and weight percentile (or TBV for metabolic outcomes). Scatterplots are presented with linear or quadratic curve fitting for the groups depicted; R2 values shown in the figures reflect the goodness of fit for the fitted curve, and not for the adjusted linear regression model. Because MRI studies in control fetuses were typically obtained later in gestation than echocardiograms, echocardiographic data were not included as predictor variables in analyses that included control fetuses, although adjustment was performed using weight percentiles derived from ultrasound estimates of fetal weight.
The authors had full access to the data and take responsibility for its integrity. All authors have read and agree to the manuscript as written.