Metric optimized gating
The MOG technique achieves time-resolved cine PC MR flow measurements without direct cardiac gating. It uses an image metric to identify a heart rate, which, when used as a hypothetical trigger to synchronise the CMR data, gives the most coherent flow information. In this way it identifies what the heart rate was during the acquisition and synchronises the information from the scan accordingly.
To allow for complete reconstruction in time resolved MR, each segment is sampled at every cardiac phase. Typically, an ECG signal or a peripheral pulse monitor is used for cardiac gating, however neither of these are readily available for fetal imaging.
In the absence of a cardiac gating signal, each segment is acquired continuously for a period of time longer than the expected fetal heart rate. This oversampled data is retrospectively reconstructed using hypothetical cardiac triggers across the normal range of fetal heart rates. Using these hypothetical cardiac triggers, the data are grouped by cardiac phase and reconstructed to produce a time-series of PC MR images. An image metric, in this case entropy, is evaluated on the reconstructed images to determine the level of mis-gating artefact. To improve its specificity to fetal pulsations, metric values are calculated over a region surrounding the target fetal vessel and its accompanying artefact. This process is repeated and the heart rate iteratively adjusted. The accepted reconstruction is that which optimizes the image metric.
To allow for heart rate variation during the acquisition, a constant heart rate is assigned to each half of the total acquisition. These heart rates are then independently adjusted until the image metric is optimized. Increasing the number of segments improves the potential agreement between the true and hypothetical trigger positions at the expense of processing time.
This prospective study was approved by the institutional ethics review board. Written consent was obtained from all adult volunteers and pregnant mothers.
Five adult volunteers underwent imaging on a commercial 1.5T CMR system (Avanto, Siemens Medical Solutions, Erlangen, Germany). The volunteers had localisers performed to target both common carotid and internal jugular veins simultaneously, and were scanned using a head and neck coil. The volunteers were exercised using an CMR-compatible bicycle until they reached a steady heart rate within the normal fetal heart rate range (110–160 beats per minute). They continued to cycle to maintain a steady average heart rate while cine PC measurements of the carotid and jugular vessels were made using conventional pulse gating with a finger pulse monitor. These pulse-gated images were acquired at high temporal and spatial resolution to provide accurate reference flow values. Immediately after the pulse-gated scan, the measurements were repeated using MOG. The scan parameters for the MOG PC measurements reflect the need for shorter scan times when imaging fetal vessels, where movement artefacts degrade image quality. The parameters used were felt to represent a reasonable compromise, with shorter scan times achieved whilst maintaining adequate temporal resolution to capture flow curve characteristics and spatial resolution to resolve flow in the vessels being interrogated (target vessel diameters: 4–10 mm). The following scan parameters were used for the MOG PC acquisitions: VENC 150 cm/s, slice thickness 5mm, field of view 240 mm, phase field of view 100% + 33% phase oversampling, matrix size 192×192, voxel size 1.25×1.25×5 mm, echo time 2.92 ms, repetition time 6.55 ms, flip angle 20°, 1 average and 4 views per segment. A typical R-R interval of 520 ms resulted in a temporal resolution of 52.4 ms giving approximately 10 true cardiac phases, which were interpolated to 15 calculated phases. A typical scan time for each vessel was 34 seconds.
Twelve pregnant mothers underwent MR imaging with the same CMR system used in the adult volunteer study. No sedation was used. All fetuses were singletons and had normal echocardiograms and obstetric findings. The median gestational age was 37 weeks with an age range of 30–39 weeks. Prior to moving the patient into the CMR scan room, the fetal heart rate was measured for 5 minutes using a CTG device (GE Corometric, Fairfield, Connecticut, USA). CMR data was obtained ensuring oversampling of the lowest fetal heart rate detected by CTG, for later reconstruction using MOG. The mother was made comfortable in a supine or lateral decubitus position in the magnet and a cardiac surface coil was placed over the maternal abdomen. The cardiac and spine coils were used in combination for imaging. An imaging protocol was then followed, starting with localisers, followed by a steady state free precession (SSFP) breath hold 3-dimensional acquisition of the whole fetus, and 3-plane static SSFP anatomical images through the fetal thorax as proposed by previous fetal CMR publications[22, 23].
Cine MOG PC acquisitions were then performed with the imaging plane prescribed perpendicular to the main pulmonary artery (MPA), ascending aorta (AAo), superior vena cava (SVC), ductus arteriosus (DA), descending aorta at the diaphragm (DAo), and umbilical vein (UV) using the anatomical images to plan the prescriptions, as in post natal PC CMR. The intra-abdominal portion of the UV, proximal to the portal branches, was targeted to avoid complex flow behaviour. Measurements were attempted in the right and left pulmonary arteries (RPA & LPA) to assess pulmonary flow. Pulmonary blood flow (PBF) was also calculated indirectly from MPA minus DA flow. The correlation between these two measurements of pulmonary blood was used as an indicator of the accuracy of the MOG technique. The combined ventricular output (CVO) was calculated as the sum of the MPA and AAo plus 3% of this sum estimated for coronary flow based on previous lamb data. Foramen ovale (FO) flow cannot be measured directly, however, since left ventricular output is made up of AAo and coronary flow and left ventricular venous filling is comprised of FO flow and PBF, it follows that FO flow can be calculated as the sum of AAo and coronary flows minus PBF. The velocity encoding range was tailored for the individual vessels with: 150 cm/s for the MPA, AAo, DA and DAo; 100 cm/s for the RPA, LPA and SVC; and 50 cm/s for the UV. In five subjects, each measurement was repeated to assess for reproducibility. The fetal scans were completed in 45 minutes per subject.
Using analysis software created in our laboratory (MATLAB, MathWorks, USA), the individual PC measurements were reconstructed using metric optimization according to our published technique. Processing was performed in a semi-automated manner, with the user selecting a region of interest centred on a pulsatile vessel on the magnitude image from the PC scan. The software then performed the MOG analysis for the region of interest selected and produced a map showing the metric value for every combination of average heart rates during the first and second halves of the acquisition within a range of heart rates from 110–180 beats per minute. The software identified the combination of heart rates with the lowest metric value and the user confirmed its appropriateness visually from its position shown on the map. Having identified the proper heart rates, the reconstructed PC CMR images were exported to standard commercial cardiovascular post processing software (Q-flow 5.2, Medis Medical Imaging Systems, Leiden, Netherlands) for flow quantification with regions of interest drawn around the vessels of interest. Two radiologists independently contoured the fetal PCs to assess for inter-observer variation. The flows obtained by the more experienced cardiovascular radiologist were used for further analysis.
In order to calculate indexed flows, we adopted a modified version of an established CMR segmentation technique to estimate the fetal weight using commercially available software (Mimics, Materialise Group, Leuven, Belgium)[23, 25, 26]. A semi-automated tracing tool was used to define the interface between the high signal amniotic fluid and lower signal uterus on the 3 dimensional SSFP acquisition of the whole uterus. The fetus was then separated from the amniotic fluid with a signal intensity threshold tool. Finally, a cutting and filling tool which interpolated between sample slices finalised the dataset and allowed the segmentation software to calculate the fetal volume. The fetal weight was then estimated from the fetal volume using a conversion factor developed by Baker et al. based on fetal density: 0.12 + 1.031 ×fetal volume (ml) = MR weight (g)[25, 26]. The total post processing time for a case was approximately 2 hours, including 30 minutes for the segmentation.
With the exception of the box and whisker plot, which uses medians and quartiles, all values in the text are expressed as means ± standard deviations. The adult and fetal flows in the reproducibility and inter-observer validation plots are the total values in milliliters per minute (ml/min) while the fetal flows have been indexed to the fetal weight and are given in milliliters per minute per kilogram (ml/min/kg). Fetal flows are also expressed as percentages of the CVO. For each vessel, flows measured between the subjects were normally distributed. Linear regression and Pearson’s correlation coefficient (R) were used to compare measurements of the same flows made with different techniques, the relationship between FO shunt and PBF, and the inter-observer agreement using MATLAB. Bland-Altman plots were also used to compare sequential measurements made in fetal vessels, the inter-observer variation for the same fetal flows, and the comparison of MOG and conventional gating in the adult validation experiment.