Intrauterine growth restriction (IUGR) is important to diagnose due to associations with increased perinatal mortality and morbidity.¹ However, current diagnostic tools such as indirect measures of umbilical artery (UA) Doppler are nonspecific to diagnose IUGR.¹ To determine if MRI can better detect IUGR, we recruited genetically identical monochorionic twins with selective IUGR (sIUGR). Given that Blood-oxygen-level-dependent (BOLD) MRI with maternal hyperoxia can detect changes in placental and fetal organ oxygenation in healthy human subjects^2,3 and BOLD MRI with maternal hyperoxia has distinguished rat IUGR models from controls,⁴ in this pilot study, we set out to determine if BOLD MRI with maternal hyperoxia could differentiate sIUGR and control twin fetuses.

Subjects: This IRB approved study enrolled 6 monochorionic diamniotic (MoDi) twins, gestational age from 28 to 34 weeks. Birth weight was collected after delivery, with appropriate for gestational age (AGA) defined as > 10th percentile, and SGA defined as < 10th percentile. Twins suspected of twin-twin transfusion were excluded.

Acquisition: Subjects were scanned on a 3T Skyra scanner (Siemens Healthcare, Erlangen, Germany) using an 18-channel body and 12-channel spine-receive arrays. The entire uterus was imaged using ss-GRE-EPI with in plane resolution 3 x 3 mm2, slice thickness 3mm, interleaved; TR = 5-7 s, TE = 30-38 ms, FA = 90°, BW = 2.3kHz/px. Total acquisition time 30 min. Maternal oxygen supply was alternated during the scan via non-rebreathing facial mask giving three 10 min episodes: 1. Normoxic (21% O2), 2. Hyperoxic (100% O2, 15L/min), 3. Normoxic (21% O2).

Processing: Bias field was estimated from the first normoxic episode using N4,⁵ and applied to all time points. Intra volume motion was corrected using non-rigid group-wise registration; pairwise registration was carried out with organ specific rigid and non-rigid body transformations in Elastix⁶ to correct for inter-volume motion. Outlier volumes were excluded based on transformation fields and temporal signal change in each voxel. Manual segmentation of placentae and fetal organs on reference frame was performed and propagated to all time frames. Time activity curves (TAC) were generated by taking mean signal intensity of ROI at each time point, filling outlier time points by linear interpolation. Finally, signal intensities were converted to ΔR2*, according to Eq(1), and resampled to give identical temporal resolution before statistical analysis across subjects. $$ S(t)=A*e - R2*t, ΔR2* = - log (S(TE)/S_baseline(TE))/TE, (1)$$ where S_baseline (TE) is the average of the first 10 min signal. Since R2* is inversely associated with blood oxygenation level SO2, we use decrease of R2* as a marker for increased blood oxygenation. Statistical Analysis: Because of long-range temporal correlations, we modeled the data using functional data analysis methods⁷ as implemented in the FDA toolbox in Matlab.⁸ We used a cubic b-spline basis with knots at two-minute intervals, then calculated mean and standard deviation as functions of time for AGA and SGA groups, and t-statistic and p-value functions comparing the group means. Average rate of signal change and between-group t-statistics at various time intervals was tabulated. Each individual subject was also examined for timing of significant oxygenation level increase at initiation of hyperoxia, at the end of hyperoxia, and at the end of second normoxic episode.

Notably, all placentae had significant oxygenation increase. In group comparison between AGA and SGA (Figure 2), the placenta exhibit slower rate of R2* change in SGA cases vs. control both at the beginning of hyperoxia and post hyperoxia (ΔR2* per minute: -0.78 s^-1min^-1 vs. -3.5 s^-1 min^-1, p <0.0001, and 0.61 s^-1 min^-1 vs. 2.3 s^-1 min^-1, p<0.01, measured at 11 min and 21 min respectively), indicating less efficient oxygen delivery in SGA placentae. In the fetal liver the difference in residual ΔR2* (SGA - control) = 6.5 s^-1 p<0.01, and ΔR2* (SGA - control) = 1.1 s^-1 p<0.05 in the fetal brain, which indicate lower oxygenation level for SGA cases in both fetal liver and brain. Further, it is observed that many SGA fetuses not only have lower oxygenation level compared to AGA fetuses, but they also exhibit lower oxygenation than the baseline (Table 1). Different trends between fetal brain and liver responses to hyperoxia are observed in some cases, which suggests variations in fetal hemodynamic autoregulation. BOLD MRI with maternal hyperoxia shows great potential in prenatally differentiating SGA fetus from controls. They have distinct response to maternal hyperoxia exposure. Their placentae show significant difference in rate of oxygen uptake as well. More clinical cases are needed to fully assess performance of BOLD MRI vs. clinical ultrasound.