Diagnostic yield of post-mortem magnetic resonance imaging for cardiac anomalies in fetal and perinatal deaths: a systematic review and meta-analysis

Systematic review/qualitative synthesis

The 16 included studies collectively provide a comprehensive overview of the diagnostic performance of pmMRI in detecting cardiac anomalies in fetal and perinatal loss. The included studies demonstrate significant geographic clustering, with the majority originating from European centres (Belgium: 4 studies, United Kingdom: 4 studies) and two studies from India. This distribution reflects the established expertise in postmortem imaging within European tertiary centres, though the growing contribution from Asian centres suggests expanding global adoption of pmMRI techniques. Sample sizes varied considerably, from as few as eight fetuses to as many as 516, introducing some heterogeneity in the precision of diagnostic accuracy estimates across studies. The gestational age of fetuses included in the studies ranged from 12 to 41 weeks, with a majority of cases falling within the second trimester. Where reported, fetal weights demonstrate considerable variation (16–4,060 g), with several studies noting improved diagnostic performance in larger fetuses.

The interval from delivery (TOP/miscarriages/stillbirths) to imaging varied widely, with some studies performing pmMRI within a few hours and others extending beyond two weeks, particularly for formalin-fixed ex vivo imaging. The time to autopsy also ranged from same-day examinations to several days post-delivery. While most studies attempted to minimize this interval, the heterogeneity in timing protocols may influence diagnostic accuracy through post-mortem tissue changes and decomposition artifacts. The imaging modalities utilized predominantly involved 1.5 Tesla (T) and 3 T MRI scanners, with select studies [13], 24] employing ultra-high field 9.4 T MRI, particularly for early gestational age fetuses and ex vivo imaging. The integration of complementary modalities, including CT-guided biopsy, demonstrates an evolving multimodal approach to minimally invasive autopsy, suggesting that pmMRI may be most effective when employed as part of a comprehensive imaging strategy rather than as a standalone diagnostic tool [12]. Radiologists involved in these studies typically had substantial experience in fetal or cardiac imaging, ranging from 2 to 15 years.

The sensitivity of pmMRI in detecting cardiac anomalies, ranged widely from 50 % (Shelmerdine et al.) to 93.2 % (Gupta et al.) [11], 17]. In contrast, specificity was consistently high across studies, typically ranging from 75 to 100 %. Several studies, including those by Kang et al. and Shruthi et al., reported perfect specificity, implying that pmMRI is highly reliable in correctly excluding cardiac anomalies in non-affected cases [18], 19], 21]. This variability suggests that factors such as imaging protocols, operator expertise, and lesion characteristics/case complexity may significantly influence diagnostic performance. The DOR, a combined measure of sensitivity and specificity, varied across studies, with the highest observed by Shelmerdine et al. and Gupta et al. [11], 15]. It is to be noted that, recent studies generally reported higher diagnostic accuracy, likely reflecting advances in imaging technology and increased radiological expertise over time. The PPV was notably high across most studies, often exceeding 70 %, and in some cases reaching 100 %, suggesting that cardiac anomalies identified on pmMRI are highly likely to be true findings. The NPV was similarly strong, typically above 85 %. However, studies like Ruegger et al. and Shruthi et al. reported significant false negatives, showing that pmMRI may still miss certain cardiac anomalies, particularly small or complex lesions [12], 19].

Meta-analysis for pooled diagnostic metrics

A total of 12 studies, comprising of data of 1810 fetuses in total, were included in the diagnostic accuracy meta-analysis wherein the pooled sensitivity of pmMRI for detecting fetal cardiac anomalies was 0.76 (95 % CI: 0.71–0.80) (Figure 2A). The pooled specificity was high at 0.96 (95 % CI: 0.95–0.97) (Figure 2B). However, significant heterogeneity was observed for both pooled sensitivity (Chi-square=37.83, p=0.0001; I²=70.9 %) and specificity (Chi-square=120.44, p<0.0001; I²=90.9 %). The SROC demonstrated a high area under the curve (AUC) of 0.89 (SE=0.045) and a Q* index of 0.82 (SE=0.047), indicating excellent overall diagnostic performance of pmMRI in detecting fetal cardiac anomalies (Figure 3). Analysis of the diagnostic threshold using the Spearman correlation coefficient revealed no significant threshold effect (r=0.063, p=0.845). Moses’ regression model yielded an intercept of 3.03 (p=0.0008) and a slope of −0.48 (p=0.062), suggesting some variation in diagnostic accuracy across studies but no clear threshold bias. The pooled DOR was 55.35 (95 % CI: 22.73–134.79) (Supplementary Figure 1). Notably, some studies such as Shruthi M et al. and Ulm et al. demonstrated extremely high DORs (>300), while others such as Staicu et al. had much lower DORs (∼5.0) [14], 16], 19]. The DOR results also showed considerable heterogeneity (Chi-square=37.25, p=0.0001; I²=70.5 %). The pooled positive likelihood ratio (PLR) was 14.06 (95 % CI: 6.61–29.89), indicating that fetuses with cardiac anomalies were approximately 14 times more likely to test positive on pmMRI than those without anomalies, while, the pooled negative likelihood ratio (NLR) was 0.32 (95 % CI: 0.24–0.43), suggesting a moderate ability to rule out cardiac anomalies when pmMRI is negative (Supplementary Figure 2). Individual NLRs ranged from 0.08 to 0.59 [11], 17]. Heterogeneity for NLR (Chi-square=28.55, p=0.0027; I²=61.5 %) was lower than that for PLR (Chi-square=66.93, p<0.0001; I²=83.6 %).

Figure 2:

Pooled sensitivity (A) and specificity (B) of pmMRI in detecting fetal cardiac anomalies as compared to conventional autopsy.

Figure 3:

SROC of the diagnostic accuracy of pmMRI in detecting fetal cardiac anomalies.

Subgroup analysis

In the subgroup analysis of studies using 1.5 T pmMRI, seven studies were included [11], 15], 17], [19], [20], [21], [22]. The pooled sensitivity for detecting fetal cardiac anomalies was 0.76 (95 % CI: 0.70–0.82), while the pooled specificity was notably high at 0.97 (95 % CI: 0.96–0.98). The pooled PLR, was 22.50 (95 % CI: 8.23–61.54), indicating a strong ability to confirm cardiac anomalies when the test is positive, while the pooled NLR was 0.31 (95 % CI: 0.21–0.46), suggesting a moderate-to-high capacity to rule out disease in negative cases. Furthermore, the pooled DOR was 76.47 (95 % CI: 37.03–157.88), reflecting high overall diagnostic accuracy. Threshold effect analysis revealed no significant evidence of a threshold phenomenon (Spearman correlation coefficient=0.357, p=0.432). The SROC (from Moses’ model) had a non-significant slope (b=−0.158, p=0.29), supporting the absence of substantial threshold variation. Overall, the findings indicate that 1.5 T MRI offers high specificity and good diagnostic odds, though some variability in sensitivity across studies suggests the potential influence of case mix, gestational age, or imaging protocols (Supplementary Figures 3–6).

Four studies [12], 14], 16], 18] utilized 3 T pmMRI, and the pooled sensitivity of this subgroup was 0.76 (95 % CI: 0.67–0.84), and the pooled specificity was 0.93 (95 % CI: 0.89–0.96). While the sensitivity was similar to the 1.5 T subgroup, specificity was slightly lower, albeit still high. The PLR was 11.25 (95 % CI: 1.50–84.39), and the NLR was 0.31 (95 % CI: 0.17–0.57), both indicating good diagnostic performance, though with wider confidence intervals reflecting variability and smaller sample size. The pooled DOR was 40.41 (95 % CI: 3.87–421.53), suggesting a strong overall diagnostic capability, albeit with a wider range compared to the 1.5 T group. The SROC yielded an AUC of 0.78 and a Q index of 0.72*, reflecting fair to good discriminative ability. No significant threshold effect was found (Spearman correlation coefficient=−0.400, p=0.600), and the Moses model slope was non-significant (p=0.259), suggesting no major threshold bias. Overall, 3 T pmMRI showed comparable sensitivity but slightly lower specificity than 1.5 T, with wider confidence intervals and more heterogeneity – likely due to fewer studies and differing imaging protocols or case mix. Nonetheless, the diagnostic odds suggest good diagnostic utility of 3 T scanners in detecting fetal cardiac anomalies (Supplementary Figures 7–10).

Additionally, one study by Votino et al., evaluated pmMRI at an ultra-high field strength of 9.4 T, which demonstrated superior visualization of fetal cardiac structures even in early gestation (≤20 weeks) [24]. The 9.4 T MRI was able to delineate the four-chamber view, outflow tracts, aortic arch, and systemic veins in nearly all cases, with a reported sensitivity of 75 % and specificity of 100 % for outflow tract anomalies. This underscores the potential for ultra-high-field MRI to improve diagnostic accuracy, particularly in early gestation, though limited availability currently restricts its routine use.

Publication bias and risk of bias assessment

The inverse funnel plot (Figure 4) demonstrated noticeable asymmetry, indicating a potential risk of publication bias within the included studies. Visual inspection reveals notable asymmetry, suggesting the possibility of publication bias or small-study effects, where smaller studies with non-significant or favourable outcomes may be underrepresented. The pseudo 95 % confidence limits, shown as an inverted funnel, indicate where studies would typically fall in the absence of bias and the distribution of points outside this funnel further supports the presence of potential bias. The quality assessment using the QUADAS-2 tool is presented in Figure 5. Although, few of the included studies were limited by relatively small sample sizes, non-consecutive sampling strategy, and retrospective study design, the included studies generally ranged from low to moderate risk of bias.

Figure 4:

Inverse funnel plot for assessment of publication bias.

Figure 5:

Risk of bias assessment of included studies using QUADAS-2 tool.

Key findings in the context of existing literature

Although earlier studies consistently showed poor diagnostic yield of pmMRI for detecting cardiovascular anomalies, this was largely attributable to the use of conventional 2-D MRI sequences, which had a very low sensitivity of approximately 12 % for identifying structural cardiac defects [29]. For instance, Alderliesten et al., utilizing 1 T MRI scanner and 2D sequences, found no cardiac defects on postmortem MRI in 26 fetuses, despite five anomalies being confirmed at autopsy, resulting in 0 % sensitivity [34]. It is now recognized that the use of high-resolution, 3-D postmortem MRI can significantly improve diagnostic accuracy [35]. Three-dimensional imaging allows detailed visualization of complex cardiac structures in multiple planes, while longer scan times and higher resolution in the postmortem setting reduce partial volume effects, enabling better detection of small cardiac anomalies. This systematic review and meta-analysis represent the first comprehensive evaluation of the diagnostic accuracy of pmMRI for detecting cardiac anomalies in fetal and perinatal deaths. Our findings demonstrate that pmMRI exhibits good overall diagnostic performance, with a pooled sensitivity of 76 % and excellent specificity of 96 % when compared to conventional autopsy as the reference standard. The high specificity reported across the included studies, with some demonstrating perfect specificity highlights its excellent capacity to correctly rule out cardiac anomalies when absent, reducing unnecessary anxiety for families already coping with perinatal loss. Furthermore, the high area under the SROC curve (0.89) and substantial diagnostic odds ratio (55.35) support pmMRI as a reliable, non-invasive adjunct to conventional autopsy, especially in settings where consent for autopsy is not granted. Conversely, the moderate sensitivity suggests that pmMRI should be interpreted cautiously when negative, particularly in cases where clinical suspicion for cardiac anomalies remains high based on other factors such as family history, associated anomalies, or antenatal imaging findings.

Technical limitations and challenges of cardiac pmMRI

While cardiac pmMRI shows good overall diagnostic metrics for CHD, several technical limitations persist. One key challenge is fetal size. In the MARIAS study by Taylor et al. 10 % of cardiac pmMRI scans were non-diagnostic, all involving smaller fetuses under 24 weeks of gestation [22]. This issue might be addressed by developing pmCTA or using high-field MRI, which holds promise for improving image clarity, particularly in small fetuses and complex cardiac cases. In this context, Jawad et al. concluded that the diagnostic yield of pmMRI is significantly influenced by fetal body weight, with weight emerging as the strongest predictor of diagnostic success, surpassing the impact of gestational age [36]. They established that pmMRI is highly likely to provide diagnostic-quality images across all major organ systems, including the cardiovascular system, when fetal body weight is at least 535 g, with a 95 % confidence that 90 % of cases would yield diagnostic images. The diagnostic adequacy of pmMRI varied across organ systems, with the musculoskeletal system requiring the lowest body weight for acceptable imaging, while the cardiovascular and thoracic systems needed significantly higher weights for reliable assessment [36].

Another limitation is the occurrence of false positives. In the study by Taylor et al. apart from fetuses ≤24 weeks, only one significant misdiagnosis (cor triatriatum) upon pmMRI was reported, with most false positives involving minor septal defects, coarctation, aortic stenosis, or partial anomalous pulmonary venous return [22]. These misinterpretations may result from suboptimal image quality, over-interpretation of findings, or potentially from cardiac anomalies being overlooked at conventional autopsy. Furthermore, the location and complexity of cardiac lesions may significantly influence diagnostic performance. Votino et al. [24] reported that pmMRI achieved 100 % sensitivity in detecting situs and aortic arch abnormalities, while sensitivity decreased to 75 % for outflow tract anomalies and further to 66.7 % for four-chamber defects, while Shelmerdine et al. [17] and Rüegger et al. [12] found outflow tract anomalies, valve abnormalities, and coronary artery anomalies challenging to assess due to the absence of perfusion and dynamic motion in the post-mortem heart. Small ventricular and atrial septal defects are frequently missed due to limited resolution and tissue collapse post-mortem (Table 2) [15], 17], 35], 36]. Subtle conditions such as coarctation of the aorta, cor triatriatum and mild forms of hypoplastic left heart syndrome may go undetected [29], 36] while conduction system abnormalities remain beyond the detection capabilities of MRI and require histological evaluation [37]. These limitations underscore the importance of combining pmMRI with histopathology or high-resolution imaging when evaluating suspected CHD.

In one of the included studies (Tang et al.), the investigators performed pmMRI of ex vivo organs fixed in 10 % formalin [13]. Tissue fixation effects can alter tissue contrast and signal characteristics compared to living tissue. Formalin fixation, commonly used for specimen preservation, changes T1 and T2 relaxation times, potentially affecting diagnostic accuracy and requiring adjusted imaging protocols [38].

Sources of heterogeneity

In the meta-analysis, significant heterogeneity was observed across studies for most diagnostic metrics, with I² values exceeding 70 % for sensitivity, specificity, and diagnostic odds ratio and subgroup analyses provided meaningful insights into potential sources of this variability. Notably, studies utilizing 1.5 T pmMRI demonstrated consistently high specificity and a higher DOR, with relatively lower heterogeneity compared to the overall analysis. This suggests that 1.5 T scanners may offer more reliable diagnostic performance, potentially due to standardized imaging protocols and broader clinical familiarity. Conversely, studies employing 3 T pmMRI displayed comparable sensitivity but slightly reduced specificity and greater heterogeneity, likely reflecting differences in image acquisition protocols, interpreter expertise, and patient populations. Although 3 T pmMRI remains a valuable diagnostic option, these findings suggest a need for protocol standardization and operator training to improve diagnostic consistency. The single study using ultra-high-field 9.4 T MRI included in meta-analysis, demonstrated superior anatomical visualization of cardiac structures, even in early gestation. While this highlights the potential for future improvements in post-mortem imaging, the limited availability and high cost of ultra-high-field MRI currently restrict its widespread clinical use.

Study’s strengths and limitations

This systematic review followed rigorous methodology following PRISMA guidelines with a pre-registered review protocol. Subgroup analyses exploring the impact of MRI field strength and imaging parameters were performed. However, several limitations warrant consideration. The potential publication bias, with possible underrepresentation of smaller studies with negative or non-significant results and significant heterogeneity, limits the generalizability of pooled estimates. This heterogeneity likely reflects several factors including variations in study populations, gestational age ranges, types of cardiac anomalies studied, imaging protocols, and interpreter experience levels. Gestational age at death and fetal weight likely affect imaging quality and anatomical visualization, with earlier gestations and low fetal weight potentially more challenging to interpret [22], [23], [24, 36]. The spectrum of cardiac anomalies studied also varies considerably between studies, with some focusing on major structural defects while others include minor anomalies that may be more difficult to detect on imaging.

The relatively small number of studies in subgroup analyses reduces the precision of field strength comparisons. Also, only two studies employed ultra-high-field MRI (9.4 T), thereby restricting robust conclusions about its diagnostic superiority, over 1.5 and 3 T equipment. Furthermore, there was a predominance of studies from high-resource settings, for example, we encountered 8 studies from Europe, which may limit applicability in low- and middle-income countries where pmMRI access is limited. Additionally, interpreter experience and availability of specialized pediatric or fetal cardiac imaging expertise likely influence diagnostic accuracy but were inconsistently reported across studies.

Clinical recommendations and future directions

The evidence generated from this meta-analysis supports integrating pmMRI into routine perinatal death investigation protocols, particularly when conventional autopsy is declined or delayed, a standard protocol, as deduced from included studies and other existing literature is presented in Table 3 [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26, 37], 39]. pmMRI can be considered a highly specific tool for confirming cardiac anomalies in fetal and perinatal deaths, thereby playing a crucial role in supporting genetic counselling, and assessing the likelihood of recurrence. However, pmMRI should be viewed as complementary to, rather than a replacement for, conventional autopsy when the latter is acceptable to families. The moderate sensitivity suggests that negative pmMRI results should not preclude conventional autopsy, along with genetic testing, when detailed cardiac evaluation is clinically indicated. Standardization of imaging protocols and interpreter training is essential to reduce variability and improve diagnostic reliability across centres. The use of either 1.5 T and 3 T pmMRI scanners remains acceptable with protocol optimization and further research is needed to explore the incremental value of higher field strengths, that is 9.4 T, advanced imaging techniques, and artificial intelligence-assisted interpretation. Studies investigating the optimal timing of imaging relative to death, the impact of body preparation techniques, and the development of specialized fetal cardiac MRI sequences may further enhance diagnostic performance. Additionally, cost-effectiveness analyses comparing pmMRI to the conventional autopsy, considering both direct costs and indirect benefits such as improved family acceptance and genetic counselling outcomes, would provide valuable information for healthcare policy decisions.