Following perinatal death, organ weights at autopsy may provide evidence of growth restriction and pulmonary hypoplasia. Whilst postmortem magnetic resonance imaging (MRI) may provide comparable information to autopsy about structural abnormalities, its ability to provide reproducible data about organ size has yet to be determined. We examined the feasibility of using postmortem MRI to provide estimates of organ size and weight.
Twenty-five fetuses of gestational age from 16 to 40 weeks underwent postmortem MRI prior to autopsy. Fetal lung, brain and liver volume estimations were performed by two observers using the stereology technique on postmortem MRI slices. Fetal lung, brain and liver weights were recorded at autopsy. Organ volume estimates and autopsy organ weights were compared using regression analysis, and estimates of fetal organ densities made. Interobserver variability was assessed using a Bland–Altman plot. Receiver–operating characteristics curve (ROC) analysis compared MRI brain : liver volume ratios to autopsy brain : liver weight ratios.
A linear relationship between organ volume estimates and organ weight was observed. Estimated densities for the fetal brain, liver and lung were 1.08 g/cm3, 1.15 g/cm3 and 1.15 g/cm3, respectively. Interobserver 5th and 95th percentile limits of agreement for fetal brain, liver and lung were − 5.4% to + 7.9%, − 11.8% to + 8.3% and − 14.3% to + 8.7%, respectively. For MRI organ volumes to detect a brain weight : liver weight ratio ≥ 4, ROC analysis demonstrated an area under the curve of 0.61, with an optimal cut-off of 4.1.
Following fetal or neonatal death, and termination for abnormality, an autopsy performed by a specialist pathologist may identify a cause of death1, and provide information to guide future parental and obstetric decision-making2. However, consent rates for perinatal autopsy are often low and are declining in a number of countries around the world2, 3.
A less invasive alternative to conventional autopsy is postmortem magnetic resonance imaging (MRI)4–8. One limitation of this technique is that it cannot on its own provide fetal organ tissue for microscopic examination. Furthermore, an important part of the conventional autopsy is the assessment of fetal organ growth by measuring organ weights9. This may provide evidence of abnormal fetal growth, including growth restriction and overgrowth syndromes, or other abnormalities such as pulmonary hypoplasia10–12.
Fetal growth restriction is implicated in many stillbirths1, but often goes undetected antenatally13. Autopsy evidence of growth restriction may include histological changes such as thymic cortical depletion14 and an abnormally-elevated brain weight : liver weight ratio10, 15. The latter is typically over 4 : 1 in cases of fetal growth restriction, although some investigators have used a ratio of 3 : 110, 15, 16. This pattern of brain and liver growth reflects the ‘brain-sparing’ fetal response to hypoxia and placental insufficiency17. If parents decline invasive autopsy or limit the examination to exclude the brain, such signs of fetal growth restriction may go undetected.
Pulmonary hypoplasia is a leading cause of neonatal death, and may be a consequence of prolonged oligohydramnios, compressive thoracic lesions, or skeletal dysplasia11. The diagnosis of pulmonary hypoplasia is usually based on the lung weight : body weight ratio (quotient) at autopsy, either alone or in combination with the radial alveolar count11. If parents decline conventional autopsy, such objective evidence of pulmonary hypoplasia will go undetected.
Weighing of major fetal organs is not possible directly using postmortem MRI. We therefore sought to determine whether postmortem MRI could provide comparable information to conventional autopsy by estimating fetal organ volumes and from this deriving organ weights in a non-invasive manner. Although postmortem imaging has been used to assess organ weights in the context of forensic medicine in children and adults18, we are unaware of any reports describing its use in the context of perinatal autopsy.
As part of a prospective study comparing postmortem MRI with conventional autopsy, 25 fetuses following termination for fetal abnormality, miscarriage, or stillbirth underwent ‘whole-body’ postmortem MRI prior to conventional autopsy, following parental consent and with research ethics committee approval (Cambridge REC 02/004). Fetal body weights ranged from 117 to 3270 g, with corresponding gestational ages of 16–40 weeks. The fetuses were imaged at 1.5T (EX, Excite GE Healthcare, Milwaukee, WI, USA) using head, knee or wrist receiver coils according to fetal size, as described previously19, using FSE-XL T2-weighted sequences. Typical parameters are shown in Table 1. Sequences were selected for depiction of anatomy, and not specifically for organ volume estimation. Conventional autopsies were performed by specialist perinatal pathologists and included weighing major fetal organs9. Prior to imaging and autopsy, the fetuses were stored in refrigerated compartments within either the maternity hospital or the hospital mortuary, at 4 °C. Postmortem MRI studies were performed as soon as possible following delivery within the service constraints of the MRI unit and the availability of the specialist MRI staff.
Table 1. Typical magnetic resonance imaging parameters for FSE-XL T2-weighted sequences in the three anatomical orthogonal planes*
Coronal and sagittal sequences include both fetal head and body; separate axial sequences were performed for fetal head and body.
Effective echo time (ms)
Repetition time (ms)
Echo train length
Field of view (cm)
Slice thickness (mm)
2–3, no gap
2, no gap
512 × 256, 384 × 256 for axial sequences
256 × 256
Number of excitations
Minimum of 4
Phase field of view
0.5 used to reduce time for axial and sagittal sequences
0.75 for axial sequences
5–8 min each plane
7 min 30 s
Organ volumes were estimated independently by two investigators: a fetal medicine research fellow (A.C.G.B.) and a radiology research fellow (F.A.G.). Volumes were measured using the stereology application on the ANALYZE® v.5 software package (BIR, Mayo Foundation, MN, USA) on a GE Advantage Workstation. This employed a grid superimposed over the axial sections (or coronal where these were not available) through fetal brain, liver and thorax. The number of grid points overlying the organ of interest was recorded thereby giving an estimate of the volume. This (the ‘Cavalieri’ method) has been demonstrated to be a time-efficient and accurate way of assessing brain volume in adults20, and organ volumes in fetuses21. Brain volume estimations included the cerebral hemispheres, cerebral ventricles and cerebellum. Fetal liver volume estimations included the gallbladder and intrahepatic vessels. Fetal lung volumes included the intrapulmonary vasculature. A grid size was selected to ensure a minimum of 400 test points per organ, reduced to 300 test points per lung pair owing to the smaller relative size of these organs. This minimum number of test points corresponds to the sum of the number of points overlying the organ of interest on consecutive slices through the fetal lungs and liver, and alternate slices for the fetal brain. The use of alternate slices for the brain reduced the time taken for estimations to about 10–15 min per organ, while maintaining a coefficient of error of < 5%. Passing–Bablok regression22 was used to compare the mean of the two observers' volume estimations with the actual organ weight recorded at autopsy. This technique does not assume that the reference method is free of imprecision and that the distribution of differences between the methods is constant for all observations23, and was therefore considered more appropriate than the least squares regression method. Interobserver agreement was assessed using a proportional Bland–Altman plot24 and as the data were not normally distributed, 5th and 95th centile limits of agreement were calculated for each organ25. Statistical analysis was performed using MedCalc software package (version 9.0, www.medcalc.be) and Microsoft Excel 2002 (Microsoft Corporation, Washington, USA).
Brain volume : liver volume ratios were calculated by dividing the mean estimate of brain volume by that of the liver volume. Receiver–operating characteristics (ROC) curve analysis determined the sensitivity and specificity of different ratios of brain volume : liver volume estimated from MRI in identifying an abnormal brain weight : liver weight ratio. Separate ROC analyses were performed for a brain weight : liver weight ratio of ≥ 3.5 and ≥ 4.0.
The median interval from delivery to autopsy was 4 (range, 2–11) days, and the corresponding interval from MRI to autopsy was 2.5 (range, 1–8) days. Comparison of fetal lung volume estimates and fetal lung weights from autopsy were made in 25 fetuses. Calculation of brain weight: liver weight ratio was possible in only 20 fetuses owing to a combination of either organ autolysis or limited parental consent. A linear relationship between the mean organ volume estimates and organ weight at autopsy was observed (Figure 1, Table 2).
Table 2. Regression analysis for the magnetic resonance imaging estimated mean organ volume compared to organ weight at autopsy
95% CI for intercept
95% CI for slope
Cusum test for linearity
y = − 1.36 + 0.93x
− 6.40 to 6.21
0.83 to 0.99
No significant deviation from linearity (P > 0.10)
y = 2.93 + 0.87x
2.00 to 5.10
0.78 to 0.94
No significant deviation from linearity (P > 0.10)
y = − 0.27 + 0.87x
− 0.67 to 0.35
0.85 to 0.90
No significant deviation from linearity (P > 0.10)
The reciprocal of the gradients gave an estimated average tissue density for the three organs, which was compared with published pediatric standards where available26. The calculated density of the fetal liver was 1.15 g/cm3 (published figure for a 1-year-old child: 1.05 g/cm3) and the calculated density of the brain was 1.08 g/cm3 (published figure for a 1-year-old child: 1.03 g/cm3). The corresponding calculated density of the fetal lung was 1.15 g/cm3 (published figures for non-aerated adult lung: 1.04–1.09 g/cm3)27.
A non-parametric Bland–Altman analysis demonstrated good interobserver agreement (Figure 2). The 5th and 95th percentiles for limits of agreement were − 5.4% to + 7.9% for the fetal brain, − 11.8% and + 8.3% for the fetal liver and − 14.3% to + 8.7% for the fetal lungs. Interobserver agreement improved with increasing organ size. Corresponding median biases for the organs were − 2.3% for the fetal brain, + 0.5% for the fetal liver and − 1.2% for the fetal lungs.
ROC curve analysis for the mean of two observers' estimates demonstrated an area under the curve of 0.61 (95% CI 0.37–0.81). An optimal cut-off of > 4.1 achieved 45% sensitivity and 100% specificity for an autopsy brain weight : liver weight ratio of ≥ 4. The corresponding area under the curve for the mean of two observers' estimates to detect a brain: liver weight ratio of ≥ 3.5 was 0.75 (95% CI 0.51–0.91), with an abnormal cut-off of 3.33. This cut-off achieved a sensitivity of 75% with 75% specificity. The full ROC analyses are illustrated in Figure 3 and are available as a supplementary table online (Table S1).
Previous reports on postmortem MRI have concentrated on the ability to detect structural fetal abnormalities4–7, 19, 28. However, the majority of fetal deaths occur in fetuses with no anatomical abnormalities29, 30. In such cases, MRI autopsy may ostensibly offer little diagnostic information—the fetus may simply appear normally formed. This study demonstrates that ex-utero MRI measurements of both the fetal liver and brain volume are highly correlated to organ weights recorded at autopsy. Estimation of organ volumes can allow individual organ weights to be derived if the density of the fetal organ in question is known. This would enable the MRI autopsy to assess clinically important questions that otherwise might have gone unanswered, for example as to whether fetal growth restriction contributed to fetal death, or whether pulmonary hypoplasia is present.
We report calculated densities for fetal brain and liver that closely correspond to published reference standards of pediatric organs26 but are greater than those for one-year-old children. As MRI studies were performed prior to autopsy, it is possible that fluid losses contributed to the discrepancy, as suggested by a study in fetal sheep31. However, fluid loss would be expected to reduce organ weight at autopsy with a corresponding reduction in tissue density18. Furthermore, autolysis following fetal death or termination of pregnancy may contribute to changes in tissue density.
The calculated density for the fetal brain more closely corresponds to that of the infant brain than for the liver. This may reflect the more accurate estimation of brain volume owing to its larger size relative to the liver and its more clearly defined boundaries on T2-weighted MR images. Indeed, the 95% limits of agreement were narrowest and median bias lowest for the brain compared to the other organs measured in this study. Difficulties in measuring the fetal liver in utero have been described using MRI echo planar imaging32 and three-dimensional (3D) ultrasound33, although these are often due to motion artifacts that are not applicable to postmortem MRI. There may also be metabolic differences in the fetal and pediatric livers that contribute to differences in density, for example glycogen concentration. Such differences may be more evident in growth-restricted fetuses34 or may change with advancing fetal maturity and gestational age. Indeed many fetuses in this study had brain weight : liver weight ratio (quotients) at autopsy in excess of 4, suggesting a degree of growth restriction.
We investigated the possibility of estimated brain and liver volumes predicting the brain weight: liver weight ratio recorded at autopsy, a ratio that is in widespread clinical and research use. Prenatal analogs of this ratio have been developed, including the head circumference : abdominal circumference ratio using conventional ultrasound biometry35 and brain volume : liver volume ratios using 3D ultrasound36 and MRI32. Brain volume : liver volume ratios deduced from MRI studies provide support for a brain-sparing effect in utero detectable using MRI volumetry32.
The brain : liver weight ratio has been described as a useful but unsatisfactory indicator of fetal growth restriction in itself15. In this study, the postmortem brain volume: liver volume ratio appeared to be only a fair predictor of an abnormal brain weight : liver weight ratio (taken as ≥ 4) as measured at autopsy. The area under the ROC curve was approximately 0.6, and a volume ratio cut-off of approximately 4 had 45% sensitivity, but 100% specificity for a brain weight : liver weight ratio of ≥ 4. Volume ratios performed better for a brain: liver weight ratio ≥ 3.5 (area under ROC curve 0.75; sensitivity and specificity 75% for cut-off of 3.33), but ratios between 3.5 and 4 may be less clinically relevant.
While the development of real-time 3D and four-dimensional ultrasound techniques has coincided with renewed interest in the estimation of fetal lung volumes37, their ability to predict lethal pulmonary hypoplasia remains uncertain. The lung-to-head ratio appears to predict survival in diaphragmatic hernia38 and its association with lung volumes on 3D ultrasonography has recently been described39. 3D ultrasound lung volumes using the 30° Virtual Organ Computer-aided AnaLysis (GE Healthcare) technique appear lower than corresponding estimates using fetal MRI40, but good agreement has been demonstrated between 3D ultrasound and lung volumes measured at autopsy41. Furthermore, sonographic fetal lung volume to body weight ratio closely agrees with lung weight : body weight ratio at autopsy42. Our data suggest that, while estimation of fetal lung volumes is possible using postmortem MRI, the interobserver variability was greater than for the brain and liver.
The lung weight : body weight ratio is commonly used by pathologists in the postmortem assessment of pulmonary hypoplasia. We did not directly assess lung volume at autopsy; instead the lungs were weighed to allow calculation of the lung weight: body weight ratio. Reproducible estimates of fetal lung weights should be possible using postmortem MRI, with good interobserver variability of fetal lung volumes (with 5th and 95th centile limits of agreement of − 14.3 to + 8.7%). We are unaware of any study comparing in-utero estimations of fetal lung volume by 3D ultrasound or MRI with those at postmortem MRI and autopsy.
We would caution against the extrapolation of these calculated tissue densities to the in-utero fetus, particularly as a degree of desiccation may occur after death or following delivery. Furthermore, our estimates of fetal tissue densities are based on a small sample of fetal deaths, predominantly from the second trimester, and may not be appropriate for other populations, such as neonatal deaths. Nevertheless, the 95% confidence intervals suggest that they closely approximate to those published for pediatric tissues26.
In summary, this study compared postmortem estimations of fetal organ volumes to the corresponding organ weights measured at autopsy; from this fetal organ densities were calculated. Postmortem estimation of fetal organ volume from MRI has good interobserver agreement, which was best for the brain and liver and worst for the lungs. For all organs studied, agreement appears to improve with increasing organ size. The ratio of brain to liver volume is a fair indicator of an abnormal brain weight : liver weight at autopsy. Postmortem estimations of fetal organ volumes may therefore provide useful information about fetal growth and organ development when conventional autopsy is declined by parents, but in this assessment and that of detection of structural anomalies, conventional autopsy remains the gold standard43.
Members of the Cambridge Post-mortem MRI Study Group include: Dr Justin Cross, Dr Patricia Set, and Ilse Joubert of the Department of Radiology, Addenbrooke's Hospital; Dr Anita Whitehead and Dr Flora Jessop of the Department of Histopathology, Addenbrooke's Hospital, and Dr Gerald Hackett, Division of Maternal–Fetal Medicine, Addenbrooke's Hospital. We acknowledge the assistance of the pediatric pathology service staff of Addenbrooke's Hospital: N. Wood, G. Kenyon, S. Brown and M. Macer. This study was supported by a grant from the Trustees of the Addenbrooke's Charities and the Fund for Addenbrooke's. ACGB's salary was in part funded by Cambridge Fetal Care. Images were produced with the help of Richard Black, Medical Physics, Addenbrooke's Hospital, and statistical assistance was provided by Sarah Vowler of the Centre for Applied Medical Statistics, Cambridge.