| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Technical Developments |
1 From the Departments of Radiology (R.A.K.H., S.W., L.C., B.M.), Obstetrics (A.R., R.C.), and Biostatistics (B.S.), University Hospital Zürich, Rämistrasse 100, CH-8091 Zürich, Switzerland. Received June 19, 2000; revision requested July 24; revision received August 16; accepted September 12. Supported by a grant from EMDO Stiftung, Zurich, Switzerland. Address correspondence to R.A.K.H. (e-mail: rahel.kubik@dmr.usz.ch).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMaterials and Methods Results DiscussionREFERENCES |
|---|
Index terms: Fetus, MR, 856.121416 • Magnetic resonance (MR), three-dimensional, 856.12141 • Pregnancy, MR, 856.121416
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods Results DiscussionREFERENCES |
|---|
The purpose of this study was to validate fetal volumetry on the basis of 3D reconstructions of MR data sets acquired in utero with T2-weighted single-shot fast spin-echo (SE) MR imaging.
|
Materials and Methods |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods Results DiscussionREFERENCES |
|---|
Data were acquired with a 1.5-T MR system (CV/i; GE Medical Systems, Milwaukee, Wis) with the same torso phased-array coil that was used in the in vivo study to optimize the signal-to-noise ratio. Localizing two-dimensional MR imaging was performed with a fast spoiled gradient-echo sequence (repetition time msec/echo time msec of 150/1.6, flip angle of 60°, bandwidth of 31.25 kHz, section thickness of 8 mm, intersection gap of 2 mm, field of view of 36 x 36 cm, matrix of 256 x 160, and one signal acquired). T2-weighted single-shot fast SE (8/90, bandwidth of 31.25 kHz, section thickness of 4 mm, intersection gap of 0 mm, field of view of 26 cm, matrix of 256 x 192, and 0.5 signal acquired) MR images were acquired sagittal to the fetus to encompass the entire volume of the acrylic plastic vessel. Each section was acquired separately. The acquisition time for each image was less than 1 second. The interimage interval was set at 6 seconds to avoid saturation effects. High-spatial-resolution, T1-weighted fast SE (300/15, echo train length of two, bandwidth of 32.00 kHz, field of view of 24 cm, section thickness of 1.5 mm, intersection gap of 0, matrix of 512 x 512, and eight signals acquired) MR images were then obtained transverse to the fetus, with an acquisition time of 4 hours 37 minutes.
All MR imaging data (DICOM-3.0 format) were transferred over a high-speed optical Ethernet connection to a high-performance Unix-language workstation dedicated to image analysis and 3D reconstruction (Silicon Graphics, Mountain View, Calif). A commercially available medical image postprocessing software package (PROVISION, version 3.0b; Algotec, Raanana, Israel) was used for segmentation and 3D modeling.
The segmentation environment was section oriented, which allowed many tools to be applied in both two-dimensional and 3D applications. We found in the reconstruction process that the semiautomatic thresholding function, with corrections on each section by means of manual definition of a closed contour on a corresponding object, provided reasonable segmentation of fetal structures and the surrounding formalin. Once the structures of interest in a 3D image volume were segmented, the postprocessing software created a corresponding 3D surface model and automatically calculated the volume of each 3D reconstruction. Two investigators (L.C., R.A.K.H.), blinded to the results of volume displacement measurements, performed the manual segmentation of the surrounding formalin two times to determine reproducibility and to demonstrate the operator independence of the resultant volume measurement.
In Vivo Study
Twenty consecutive pregnant women (mean age, 28.9 years ± 5.3 [SD]; age range, 18.2–38.1 years) were asked to participate in the study, either at routine prenatal US in the obstetric clinic or after clinically indicated MR pelvimetry in our department. All participants provided their written consent after they were informed that the study was experimental and that there would be no consequence relevant to their care as a result of the findings. The study was approved by the local ethics committee.
The study included 18 singleton fetuses and two sets of twins, for a total of 22 fetuses. The range of age of gestation at MR imaging was 16 weeks six days to 41 weeks 3 days (mean, 32.1 weeks ± 8.9). At 33 weeks of gestation in one fetus, MR imaging was performed and polyhydramnios was diagnosed at US; the newborn died of Pena-Shokeir syndrome type 1 (fetal akinesia sequence) in the immediate postpartum period (15). In one twin pregnancy seen in the 21st week of gestation, one twin demonstrated severe growth retardation due to impaired hemodynamics. Prenatal findings were normal in all other pregnancies.
Prenatal US and MR imaging were performed the same day in 18 fetuses. In the remaining four cases, they were performed within 1 week (mean, 0.3 day ± 2.3; range, -6 to 7 days), which allowed direct comparison of the US and MR estimates.
In all cases, prenatal US was performed by an experienced investigator with a 4-MHz sector probe (128X P/10; Acuson, Mountain View, Calif) or multifrequency probe (3.5–5.1 MHz; Siemens, Issaquah, Wash). Fetal weights were calculated with the Hadlock formula (16,17). Fronto-occipital and biparietal diameters were measured, and head circumference was automatically calculated.
MR imaging was performed with a 1.5-T system (Signa Horizon LX and CV/i; GE Medical Systems). A torso phased-array coil was used to optimize the signal-to-noise ratio whenever possible. In 14 cases, mainly in the third trimester, however, the body coil was used to ensure a field of view large enough to cover the entire uterus.
In all cases, two-dimensional fast spoiled gradient-echo (150/1.6, flip angle of 60°, bandwidth of 31.25 kHz, section thickness of 8 mm, intersection gap of 2 mm, field of view adapted to the size of the uterus, matrix of 256 x 160, and one signal acquired) MR imaging was performed. Then, T2-weighted single-shot fast SE (8/90; bandwidth of 31.25 kHz; section thickness of 4 mm, intersection gap of 0; field of view of 26–40 cm, depending on uterine size; matrix of 256 x 192; and 0.5 signal acquired) MR imaging was performed sagittal to the fetus to encompass the entire uterine volume. In one case, the plane was coronal to the fetus. As in the postmortem study, the sections were acquired separately with a pause of about 6 seconds between each, for a total imaging time of about 6 minutes. The single-shot fast SE images were subsequently transferred to the workstation. In each case, the fetus, amniotic fluid, and placenta were segmented manually and reconstructed separately (L.C., R.A.K.H.) with automatic volume calculation.
To measure fronto-occipital diameter (FOD) and biparietal diameter (BPD), sections transverse to the fetal skull were reformatted from the MR data by means of multiplanar reconstruction. Measurements were performed on these transverse reformatted images, which depicted the cavum septi pellucidi, falx cerebri, and both thalamic nuclei (18). The head circumference (HC) was calculated automatically from these data by means of the following formula:
Statistical Analysis
Statistical analysis was performed (STATVIEW, version 4.5; Abacus Concepts, Berkeley, Calif), and results were expressed as the mean plus or minus SD. Fetal weight was calculated from the fetal volume measurements based on MR data sets with two methods—by assuming a mean fetal tissue density of 1.0 g/cm3 and with the formula of Baker et al (19):
For six fetuses examined with MR imaging and US within 1 week of delivery, fetal weights calculated with US biometry and those calculated with the MR data by using the two techniques were compared with newborn birth weights by means of paired t tests.
|
Results |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods Results DiscussionREFERENCES |
|---|
|
The mean volume of formalin (3,150 cm3) from the four reconstructions, which was calculated two times by the two investigators, was 3,174 cm3 ± 59. Thus, the systematic difference between the two methods (bias) was 24 cm3 (0.8%), and the maximal deviation was 107 cm3 (3.4%). The mean interinvestigator difference was 80 cm3 (2.5% of the actual value).
In Vivo Study
Three-dimensional reconstruction of the fetus and uteroplacental unit on the basis of MR data sets acquired with the T2-weighted single-shot fast SE sequence was feasible in all cases (Figs 2 –6). However, segmentation of the fetus and uteroplacental unit had to be performed mostly manually and took up to 6 hours for each case.
|
|
|
|
|
|
|
|
|
Fetal volume determined on the basis of the 3D reconstruction from MR data ranged from 109 to 3,723 cm3 (mean, 1,757 cm3 ± 1,254). By assuming a mean fetal tissue density of 1.0 g/cm3, mean fetal weight ranged from 109 to 3,723 g. With the formula of Baker et al (19), fetal weight ranged from 232 to 3,958 g (mean, 1,932 g ± 1,293). Fetal weight determined on the basis of US biometry with the Hadlock formula (16) ranged from 166 to 4,634 g (mean, 1,939 g ± 1,430).
Linear regression of fetal weight based on the MR and US estimates showed good correlation (r2 = 0.97). Because the MR estimate of fetal weight depends directly on volume, the correlation coefficient is not influenced by the formula used (ie, by assuming a mean tissue density of 1.0 g/cm3 or with the formula of Baker et al) (19).
Fetal weights determined on the basis of MR volumetry by assuming a mean tissue density of 1.0 g/cm3, were, on average, 182 g lower than the values determined on the basis of US biometry; the underestimations were significant (P = .008). With the formula of Baker et al (19), no significant difference was seen between the MR and US estimates of fetal weight (P = .90; 95% CI: -128, 113 g) (Fig 7).
|
|
Biparietal diameter values ranged from 40 to 104 mm (mean, 77 mm ± 22) with US biometry versus from 37 to 94 mm (mean, 71 mm ± 18) with MR imaging (P = .005) (Fig 8). MR imaging also underestimated fronto-occipital diameter values versus US (mean of 92 mm ± 27 and range of 42–121 mm vs mean of 95 mm ± 28 and range of 46–128 mm, respectively; P = .001). As a result, MR estimates of head circumference were significantly lower than their US counterparts (mean of 258 mm ± 70 and range of 125–339 mm vs mean of 272 mm ± 74 and range of 134–366 mm, respectively; P < .001).
|
|
Discussion |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods Results DiscussionREFERENCES |
|---|
In the six fetuses examined within 1 week of delivery in our study, fetal weights calculated with MR imaging with the two techniques correlated with the actual birth weights. Fetal weights determined with MR imaging by assuming a mean fetal tissue density of 1.0 g/cm3 were lower than birth weights (P = .03). With the formula of Baker et al (19), with a tissue density coefficient of 1.03, however, no significant difference was observed. On the basis of this small sample, the latter technique seems better suited for estimating fetal weight, although a potential increase in weight between MR imaging and birth could partly account for the underestimation with the much easier transformation factor of 1.0 g/cm3.
The adjustment of 0.12 (kg) in the formula of Baker et al (19) has a much higher influence on weight calculated in early pregnancy, which caused a large difference in fetal weight determined on the basis of MR imaging with the two methods. In a recently published study, Uotila et al (21) performed T1-weighted MR imaging within 48 hours of delivery in 10 normal pregnancies. They found a tissue density coefficient of 1.07. Thus, it remains to be determined in a larger study population which technique is most likely to reflect actual fetal weight in early and late gestation. MR imaging findings also resulted in underestimation of biparietal and fronto-occipital diameters compared with US findings for reasons that are not yet understood.
Accurate assessment of fetal weight during and at the end of pregnancy is useful to help manage labor and the neonatal period, because both intrauterine growth retardation and fetal macrosomia increase the risks of perinatal morbidity and mortality. US can be used to measure fetal weight only indirectly on the basis of anatomic measurements. Substantial inaccuracy was found in many studies comparing US estimates with actual birth weights, with a trend to overestimate low birth weight and underestimate high birth weight (16,22,23). Whether MR imaging could outperform US (eg, in macrosomic fetuses at high risk for shoulder dystocia) remains to be determined.
Most institutions now use the formula of Hadlock et al (16,17), which combines US measurements of fetal head circumference, abdominal circumference, and femur length. Results will thus be influenced if one parameter (eg, femur length) deviates from normal owing to anthropologic differences or disease. US measurements are also biased by maternal obesity, oligohydramnios, and head engagement in late pregnancy. Potential sources of error in the MR-based measurements include fetal motion and volume averaging.
MR imaging can be used to determine not only fetal weight but also amniotic fluid and placental volumes. A limitation of our study was the absence of a reference standard to validate our measurements. Thus, we decided to perform postmortem MR imaging in a formalin-preserved fetus embedded in a known amount of fluid with the same sequence as was used in the in vivo study. Volume calculations showed good agreement, with a bias of 0.8%. Although allowances must be made for a smaller fluid volume (except in polyhydramnios) and for segmentation of the amniotic fluid in vivo being slightly more difficult owing to fetal motion, volumetric measurements are likely to represent the real values. Measurements of amniotic fluid volume might prove clinically useful in the follow-up of polyhydramnios (eg, to monitor the response to therapeutic amniotic drainage).
The results of this study demonstrate the feasibility of 3D reconstruction from in vivo fetal MR imaging. Although the current image quality of 3D reconstruction may be insufficient for detecting subtle fetal contour malformation (eg, facial or limb abnormalities), higher definition will soon be achieved with refinements in sequence software and faster imaging. This will resolve the current problem of fetal and uterine movement between image acquisitions that cause translation of fetal body parts into two adjacent sections and, thus, enlarge the steps in the fetal surface contour in 3D reconstruction. In the present study, 3D reconstruction of a 33-week fetus with a postpartum diagnosis of Pena-Shokeir syndrome type 1 (15) showed a smooth surface of especially good quality. The fetal akinesia characteristic of this syndrome minimized motion artifacts, whereas the polyhydramnios facilitated differentiation between fetal tissues and the uteroplacental unit.
In the present study, segmentation of the fetus, amniotic fluid, and placenta had to be performed mostly manually and took about 6 hours in each case. Especially in pregnancies in the last trimester, only a small rim of amniotic fluid could be seen between the fetus and placenta or uterine wall, which made automatic segmentation with a threshold value impossible because the signal intensities of fetal tissues and the uteroplacental unit are very similar. Furthermore, the fetal brain and fluid-filled fetal organs (eg, urinary bladder) have high signal intensity similar to that of amniotic fluid on T2-weighted images and would thus also be excluded with an automatic threshold for the exclusion of amniotic fluid. Although 3D reconstruction of second trimester pregnancies was facilitated by a greater amount of perifetal amniotic fluid, it tended to be degraded by more fetal movement. The main factor limiting the routine use of 3D models of important anatomic structures in current clinical practice is the amount of operator time involved. Segmentation in our study depended on the knowledge of the medical expert expressed with the interactive computer segmentation tools. In the near future, automated methods will substantially reduce the time involved and increase the applicability of the method.
A combination of statistical classification and anatomic information has been used to segment MR images of different organs in new prototype software. Anatomic knowledge-guided algorithms have recently been used with promising results for the automatic detection and segmentation of pathologic structures from a combination of T1-, T2-, and proton-density–weighted MR images (24). Implementation of such algorithms in commercially available systems will provide faster and more user-friendly segmentation, which will also result in more accurate 3D models. Such future image postprocessing methods combined with improved data acquisition with fast MR imaging in utero will permit 3D reconstruction of the most detailed fetal and uteroplacental structures.
We believe that 3D reconstruction from sectional MR data sets will play an increasing role in the prenatal diagnosis of fetal morphologic anomalies and growth retardation and in the preoperative simulation of fetal surgery. Volumetry of the fetus and uteroplacental unit on the basis of fast MR data sets is feasible. MR estimates of fetal weight correlate closely with results at fetal US biometry and with birth weight.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Abbreviations: 3D = three-dimensional, SE = spin-echo
Author contributions: Guarantors of integrity of entire study, R.A.K.H., S.W., L.C., B.M.; study concepts, R.A.K.H., L.C., R.C.; study design, R.A.K.H., L.C.; literature research, R.A.K.H., L.C.; clinical studies, L.C., A.R., R.A.K.H.; data acquisition and data analysis/interpretation, L.C., R.A.K.H.; statistical analysis, B.S., L.C., R.A.K.H.; manuscript preparation, R.A.K.H.; manuscript definition of intellectual content, R.A.K.H.; manuscript editing, S.W., B.M., R.C.; manuscript revision and final version approval, all authors.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods Results DiscussionREFERENCES |
|---|
Related Article
This article has been cited by other articles:
|
|
 |
|
  S. Hassibi, N. Farhataziz, M. Zaretsky, D. McIntire, and D. M. Twickler Optimization of Fetal Weight Estimates Using MRI: Comparison of Acquisitions Am. J. Roentgenol., August 1, 2004; 183(2): 487 - 492. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
T. F. Reichel, R. M. Ramus, J. T. Caire, L. S. Hynan, K. P. Magee, and D. M. Twickler Fetal Central Nervous System Biometry on MR Imaging Am. J. Roentgenol., April 1, 2003; 180(4): 1155 - 1158. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 T. M. Keller, A. Rake, S. C. A. Michel, B. Seifert, G. Efe, K. Treiber, R. Huch, B. Marincek, and R. A. Kubik-Huch Obstetric MR Pelvimetry: Reference Values and Evaluation of Inter- and Intraobserver Error and Intraindividual Variability Radiology, April 1, 2003; 227(1): 37 - 43. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. Levine Three-dimensional Fetal MR Imaging: Will It Fulfill Its Promise? Radiology, May 1, 2001; 219(2): 313 - 315. [Full Text]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
/90, matrix of 256 x 256, section thickness of 4 mm, intersection gap of 0, and total acquisition time of 1 minute 38 seconds) (blue = formalin) and (right) a high-spatial-resolution fast SE sequence (100/15, matrix of 512 x 512, section thickness of 1.5 mm, intersection gap of 0, and total acquisition time of 4 hours 37 minutes).
