Introduction

Autopsy has an undisputed role in confirming or refuting antemortem diagnosis and in identifying a cause of death where this could not be determined before death.¹ Cardiac abnormalities are found in ≤35% of fetal autopsies,^2,3 of which only ≈50% may be detected antenatally.^4,5 Furthermore, cardiac defects are seen at autopsy in ≈10% of sudden deaths in infants^5,6 and could be the cause of death in ≤84% of these cases.⁷ Although the majority of such abnormalities are structural,² only 40% are detected before death.⁶ Thus, fetal and pediatric cardiac autopsies have a crucial role in counseling parents with regard to both the cause of death of their child and the implications of such findings for future pregnancies,^8,9 as well as for quality assurance of antenatal screening programs and antemortem diagnostic procedures.

Editorial see p 1909

Clinical Perspective on p 1944

Despite this, there has been a global reduction in fetal and pediatric autopsies in the last decade. Current fetal autopsy rates are <50%, and neonatal autopsy rates are <20% in the United Kingdom,¹⁰ which is well below the recommended minimum by the Royal Colleges of Pathologists and Obstetrics and Gynaecologists (United Kingdom).¹¹ Several researchers have used postmortem magnetic resonance (MR) imaging as an alternative for conventional autopsy in the last decade, particularly for the brain.¹² However, the published data on postmortem cardiovascular MR (CMR) imaging has been limited to conventional 2-dimensional sequences and shows extremely poor sensitivity (varying between 0% and 25%) for the detection of structural cardiac abnormalities.¹³

The aim of this study was to compare, in a blinded fashion, the diagnostic accuracy of postmortem, 3-dimensional (3D) CMR imaging with conventional autopsy and histopathology assessment in fetuses and children.

Methods

Study Participants

We performed postmortem CMR imaging in an unselected population of fetuses and children referred for autopsy to the Great Ormond Street Hospital for Children or University College London Hospitals over a 3-year period between March 2007 and September 2011.¹⁴ We obtained informed parental consent either as part of the autopsy consenting process or by separate telephone consenting, as described previously.^15–17 The study had institutional approval (04/Q0508/41).

All of the fetuses and children (≤16 years of age) undergoing conventional autopsy were eligible for recruitment. CMR imaging was performed as soon as practically possible. Cases were excluded if consent was not available or if CMR imaging could not be performed before autopsy. All of the subjects were stored in a mortuary at 4°C before CMR imaging.

Imaging Technique

We performed postmortem CMR imaging using a 1.5-T MR scanner (Avanto, Siemens Medical Systems, Erlangen, Germany) and the following MR sequences: (1) 3D T₂-weighted turbo spin echo, (2) 3D T₁-weighted volumetric interpolated breath-hold examination, and (3) 3D constructive interference in the steady state (Table 1). CMR imaging and transfer of fetuses, neonates, and children between the mortuary and the MR scanner were performed by 2 MR technicians with 5 years of CMR experience and an MR research fellow with 1 year of CMR experience (R.J. and W.N. in the Magnetic Resonance Imaging Autopsy Study Collaborative Group).

Reporting of CMR Images

A specialist in pediatric CMR (12 years of experience; A.M.T.) reported all of the CMR images using the OsiriX platform (OsiriX Foundation, Geneva, Switzerland). The imager was given the age of the case but was otherwise blinded to all of the clinical details and the subsequent autopsy findings. All of the abnormal cardiac lesions, irrespective of cause of death, were reported. If it was not possible to define whether the heart was normal or abnormal related to image quality (poor resolution), the image was deemed to be nondiagnostic.

Reporting of Autopsy Data

Conventional autopsies were performed by 1 of the 4 perinatal or pediatric pathologists (10–20 years experience, including N.J.S. and R.J.S.) or a specialist pediatric cardiac pathologist (12 years experience; M.A.). All of the clinical information, except the CMR data, was available to the pathologist.

Data and Statistical Analysis

Data were collected into a Microsoft Access database (version 2003, Microsoft Inc, Redmond, WA), which was developed based on the Royal College of Pathologists (United Kingdom) postmortem guidelines.¹¹ The imager and pathologists could only access their respective sections of the database until 3 months after the 400th case was reported.

Primary outcomes were sensitivity, specificity, positive predictive valve, and negative predictive value where postmortem CMR (index test) defined the cardiac diagnosis compared with conventional autopsy (gold standard). Subgroup analysis was performed on the basis of age (fetuses ≤24 weeks’ gestation, fetuses >24 weeks’ gestation, and children ≤16 years of age) and type of cardiac lesion (all heart diseases [infective, ischemic, structural], structural heart defects [major and minor] or major structural heart defects only).

To determine diagnostic accuracy (ie, sensitivity and specificity) within ±5% with 95% confidence required 400 cases if diagnostic accuracy was only 50%. If diagnostic accuracy was 90%, then primary outcome was estimable to within ±3% for 400 cases. Exact methods were used to calculate confidence intervals (CIs) and χ² test to examine statistical significance. SPSS (version 19 for Macintosh, SPSS Inc, IBM, Armonk, NY) was used for data analysis.

Results

Demographic Data

Postmortem CMR was performed on 423 cases. The first 20 cases were used for MR sequence optimization and for imager training, and in 3 cases autopsy data could not be retrieved. Of the remaining 400 cases, 185 were fetal cases at ≤24 weeks’ gestation; 92 were fetuses from >24 weeks’ gestation up to term; and 123 were newborns and children (≤16 years of age). Hospital-requested autopsies (fetal medicine unit or inpatients) accounted for 292 cases, with Her Majesty’s Coroner (medical examiner) –requested autopsies accounting for the remainder, 6 fetuses (2.2%) and 102 children (82.9%). Mean time (SD) between death/delivery and postmortem MR imaging was 4.5 days (2.5 days).

Conventional autopsy studies revealed a total of 47 cardiac abnormalities (11.8%; Table 2), 20 (10.8%) in fetuses ≤24 weeks’ gestation; 5 (5.4%) in fetuses >24 weeks’ gestation, and 22 (17.9%) in newborns and children >1 week of age to ≤16 years of age.

Nondiagnostic Cases

Three cases were excluded from analysis; in 2 cases no conventional autopsy was performed and in 1 insufficient CMR sequences were performed. Thirty-eight CMR data sets (10%) were nondiagnostic (37 in fetuses ≤24 weeks’ gestation; 1 in a fetus >24 weeks’ gestation). There were 3 cardiac abnormalities in this group. Subsequent analysis of the data has been performed on the remaining 359 cases, which had 44 cardiac abnormalities identified by conventional autopsy.

Comparison of Cardiovascular MR With Conventional Autopsy

The overall sensitivity, specificity, positive predictive value, and negative predictive value with 95% CIs are given in Table 3. Subgroup analysis of the 147 fetuses ≤24 weeks’ gestation with diagnostic images, the 90 fetuses >24 weeks’ gestation, and the 122 neonates and children ≤16 years of age are shown in Table 3. Figures 1–3 show representative examples of CMR images

• Download figure
• Download PowerPoint

• Download figure
• Download PowerPoint

• Download figure
• Download PowerPoint

For fetuses ≤24 weeks’ gestation, 3 diagnoses of congenital heart disease were missed, 2 tetralogy of Fallot (1 case misdiagnosed as ventricular septal defect [VSD] and other as normal) and 1 VSD. The remaining 14 cardiac abnormalities were detected. There were 5 overcalls by postmortem CMR imaging, 2 VSDs, 1 nonspecific cardiac abnormality, 1 aortic valvular stenosis in a fetus with biventricular hypertrophy, and 1 total anomalous pulmonary venous drainage.

For fetuses >24 weeks’ gestation, there were no missed major structural cardiac diagnoses; 1 case of papillary necrosis of unknown cause in a stillborn was missed and wrongly diagnosed as coarctation. Postmortem CMR detected the remaining 5 abnormalities defined at conventional autopsy. There were 6 overcalls, 1 complex congenital heart disease (cor triatriatum; Figure 4), 3 VSDs, 1 possible coarctation, and 1 anomalous right upper lobe pulmonary venous drainage.

• Download figure
• Download PowerPoint

For infants and children, no structural abnormalities were missed (n=13); however, 8 cases of myocarditis (2%) were not diagnosed at postmortem cardiovascular imaging. There were 2 overcalls of minor cardiovascular abnormalities, both atrial septal defects (Figure 4).

When all of the nondiagnostic cases were considered as false positives, CMR had a sensitivity of 82.4% (59.0–93.8%) and specificity of 74.9% (67.8–80.8%) in fetuses ≤24 weeks’ gestation (Table I in the online-only Data Supplement). When the nondiagnostic cases were considered as false negatives, the sensitivity dropped to 25.9% (16.1–38.9%) and specificity increased to 96.2% (91.3–98.4%) in this subgroup (Table II in the online-only Data Supplement). There was no change in diagnostic accuracy in other age groups on sensitivity analysis.

Discussion

In this large study of postmortem CMR imaging, contrary to previous reports, we have demonstrated good diagnostic use of postmortem CMR imaging in fetuses, newborns, and children. In the series of 400 cases, CMR failed to detect only 2 cases of significant structural heart disease abnormalities (both tetralogy of Fallot, false negatives), and both of these were in small fetuses ≤24 weeks’ gestation. A minor cardiac abnormality was missed (VSD). There were 13 overcalls (false positives) of structural heart disease, of which only 1 was for complex congenital heart disease (cor triatriatum), the remainder being for simple congenital heart defects (VSD, atrial septal defect, coarctation, partial anomalous venous drainage, and aortic stenosis). All but 2 overcalls occurred in fetuses, with 1 in a term stillbirth in whom cor triatriatum was diagnosed at CMR (of note, no cause for this stillbirth [all body systems] was defined at conventional autopsy in this case). Importantly, a normal scan in fetuses >24 weeks’ gestation, newborns, and children meant that there was no structural cardiac abnormality. However, whereas CMR yields excellent results for structural heart disease, in newborns and children, conventional autopsy identified myocarditis as the cause of mortality in 2% of cases, but this was not detected at postmortem cardiac imaging.

Previous Postmortem CMR Imaging Studies

Several authors have previously reported poor accuracy of postmortem CMR imaging. Alderliesten et al¹⁸ compared postmortem CMR imaging with conventional autopsy in 26 fetuses. Cardiac abnormalities were seen in 5 cases at autopsy, but none of these cases were detected by postmortem CMR imaging (sensitivity, 0%). Breeze et al¹⁹ have reported a comparison of postmortem whole body imaging and autopsy in 36 fetuses. Of the 8 fetuses who had cardiac lesions, only 2 were detected by postmortem CMR imaging (25% sensitivity). Cohen et al²⁰ have reported 2 cases (no cardiac lesions) of postmortem CMR in sudden infant death; however, both the images were nondiagnostic.

All of these studies used 2-dimensional CMR imaging with low resolution (section thickness, 2–5 mm), and general radiologists performed the interpretation of cardiac images. Our study differs from the previously published data described above in several ways. First, we used 3D instead of 2-dimensional imaging, which enabled accurate identification of complex structures in any imaging plane. Second, we used much higher resolution and longer scan times, so that partial volume effects were minimized, thereby enhancing the potential to accurately identify small structures. Third, a specialist pediatric CMR imager reported images. The higher accuracy of postmortem CMR imaging in our study could be related to these factors.

In our study, a high-resolution 3D constructive interference in the steady-state sequence was most useful and T₁-weighted (volumetric interpolated breath-hold examination) images least useful. It is important for imagers reporting postmortem CMR to be familiar with normal postmortem artifacts so that these are not reported as pathologic findings. Small pericardial and pleural effusions, intracardiac air, and blood clots are considered as postmortem artifacts and are better visualized on postmortem CMR than at conventional autopsy. Similarly, ventricles may have a thickened appearance after death, and this should not be mistaken for ventricular hypertrophy. Although valve leaflets can be seen clearly when closed, they are more difficult to visualize if fully open.

The other strength of this study was the double-blinded study design and prospective data collection into a large predesigned structured database (prepared in consultation with a large team of pathologists, imagers, a fetal medicine specialist, a geneticist, a neonatologist, pediatricians, and statisticians), which allowed for categorical reporting of both conventional autopsy and postmortem CMR data in a similar way. Double-blinded reporting (imager and pathologist blinded to each other’s data) ensured that the index test (postmortem CMR) was completely independent of the reference test (ie, conventional autopsy as the gold standard). Although we collected more than 100 variables describing each structure in heart, analysis was based on the key pathologic diagnosis, as per standards for the reporting of diagnostic accuracy study guidelines; therefore, none of our 400 cases in the study was counted more than once.

Potential Limitations/Solutions

There were several issues related to smaller fetuses at ≤24 weeks’ gestation. First, all but one of the nondiagnostic CMR scans (10% of the total) were seen in this group of cases. Furthermore, the 3 structural heart disease misses and 5 of the 13 overcalls were seen in this group. The reason for all of these misses was related to the small size and hence often poor image resolution seen in these cases; despite these limitations, the majority of cases had diagnostic images that were useful for defining the underlying diagnosis. The underlying problem of image resolution in this group could be addressed by carrying out imaging at high-field strength, and 3 recent reports have shown the value of 9.4-T for whole body imaging²¹ and the heart²² and 3-T imaging of fetal heart.²³ With improved resolution, we would expect the diagnostic accuracy for fetuses ≤24 weeks’ gestation to be as good as the older fetuses and young neonates.

Outside the fetuses ≤24 weeks’ gestation, only 1 major structural abnormality was overcalled (cor triatriatum), with the remaining overcalls being septal defects (atrial septal defects or VSDs), a possible aortic coarctation, aortic stenosis, or 1 case of partial anomalous pulmonary venous drainage. All of these false-positive CMR diagnoses could be because of inadequate image resolution, imager misdiagnosis, or no detection of the cardiac lesions at conventional autopsy. All of the lesions identified, including cor triatriatum (a rare diagnosis with a flimsy membrane in the left atrium that could be missed or destroyed during the conventional autopsy) could have been easily missed at conventional autopsy. In particular, the diagnosis of coarctation is difficult and subjective at both CMR and conventional fetal autopsy. A single experienced pediatric cardiac imager reported all of the images, and the results may not extend to less experienced radiologists.

Clinical Use

Even without the potential use of 9.4-T imaging for small fetuses, for all fetuses and neonates, CMR imaging could be used as the first-line assessment for structural heart disease. If all of the nondiagnostic and positive CMR scans were referred for conventional autopsy, only 1 diagnosis would have been missed (although there were 2 missed diagnoses in the cohort, 1 of these was diagnosed as having a VSD [actual diagnosis tetralogy of Fallot] and so would have had a positive diagnosis and hence been referred for conventional autopsy). A total of 80 conventional autopsies would be required instead of the 320 performed, a reduction of 75%. However, if we assume that 9.4-T imaging would reduce the number of nondiagnostic MR scans and provide diagnostic accuracy similar to that of the older fetuses and neonates and that some of the overcalls are conventional autopsy misses, then CMR imaging could be seen as a replacement for conventional autopsy.

However, for newborns, infants, and children, because of the potential to miss the diagnosis of myocarditis (reported in ≤2% of cases),²⁴ at present cardiac tissue is necessary in all older patients, should a diagnosis of structural heart disease not be made (myocarditis was not seen in any older case with structural heart disease). This may be possible using a less-invasive laparoscopic approach,²⁵ or, in selected cases, percutaneous biopsy done under image guidance may be a satisfactory alternative to open dissection and biopsy,²⁶ but this requires further evaluation. Developments in tissue characterization of myocarditis and myocardial damage in the living using T1 and T2 mapping²⁷ and magnetization transfer MR imaging methods may mean that identification of myocarditis by CMR becomes a reality in the future, but again, this will require further study.

Ultimately, a stepwise process, starting with CMR imaging, which potentially proceeds to less invasive sampling of the heart and, if required, to a full conventional autopsy that is carried out with close parental involvement is more likely to get agreement from parents for undertaking limited cardiac autopsy in such cases.²⁸ Furthermore, it is possible that the use of postmortem CMR imaging as a routine adjuvant to cardiac autopsy may improve the accuracy of cardiac autopsy by guiding the pathologists to specific pathologic lesions. Moreover, 3D CMR imaging has unique advantages of reconstruction in any plane and as permanent and reproducible data storage.

Cardiac anomalies are often present in fetuses with extra cardiac malformations, and this may be missed or wrongly interpreted in antenatal scans^3,29; thus postmortem CMR examination is crucial in such cases and would be far more acceptable to parents.²⁸ Additional findings from postmortem cardiac scans may change the antemortem diagnosis and, therefore, may even alter the recurrence risks for future pregnancies.³

Conventional fetal and pediatric autopsies are performed only in specialized centers and may often require transfer of bodies over long distances and a consequent delay in funeral. Unlike antemortem cardiac MR imaging, postmortem 3D CMR imaging can be performed in local hospitals with minimal training of radiographers, with images transferred to the specialist pediatric cardiac MR radiologists for interpretation. Total time for CMR imaging may be over an hour; nevertheless, it is likely that postmortem CMR imaging is performed out of hours, and little monitoring is needed during the scan; therefore, scan time may be of little significance.

In conclusion, we have shown that 3D postmortem CMR imaging can provide equivalent structural information to that of conventional autopsy in the majority of larger fetuses, newborns, and children. This technique may have a major role in developing less invasive autopsy methods. Moreover, routine use of postmortem CMR imaging as an adjuvant to conventional autopsy may increase the yield from conventional autopsy.

Acknowledgments

We thank all of the parents who participated in this study, at one of the most difficult times in their lives.

Footnotes