Preeclampsia (PE) is a serious multisystem disorder of pregnancy associated with placental insufficiency, leading to adverse maternal and foetal outcomes, including foetal growth restriction (FGR). Affecting 2–8% of pregnancies worldwide, PE remains a major cause of maternal and neonatal morbidity and mortality. Early and accurate detection is essential for improving clinical outcomes, yet current screening methods, such as Doppler ultrasound and maternal biomarker assessments, provide only limited predictive accuracy. Traditional imaging techniques struggle to capture the complex heterogeneity of the placenta, particularly in distinguishing PE from normotensive pregnancies and identifying those cases that may progress to FGR.
Recent advancements in deep learning radiomics (DLR) offer a novel, automated approach to placental assessment using magnetic resonance imaging (MRI). By combining radiomics, which extracts high-dimensional quantitative features from images, with deep learning, DLR enables more precise characterisation of placental abnormalities. A multicentre study explored the potential of DLR-based placental MRI in identifying PE pregnancies, differentiating between PE with and without FGR, and providing more accurate prognostic insights than conventional radiological assessment. The findings demonstrated the effectiveness of DLR in distinguishing pathological placental characteristics, suggesting its potential as a valuable clinical tool in maternal-foetal medicine.
Deep Learning Radiomics: A New Frontier in Placental Imaging
Radiomics has emerged as a powerful method for extracting hidden patterns from medical images and translating complex imaging data into clinically relevant insights. When integrated with deep learning, this approach enhances diagnostic accuracy by identifying subtle placental abnormalities that may not be evident through traditional imaging. The study employed a semi-supervised segmentation algorithm to analyse placental MRI scans from 420 pregnant women, comprising 140 PE cases and 280 normotensive pregnancies. The model extracted radiomic features, including texture, shape and wavelet characteristics, to build predictive models capable of distinguishing PE pregnancies from normotensive cases.
The findings revealed that the DLR signature demonstrated superior discrimination ability, achieving high accuracy in identifying PE pregnancies. The model outperformed conventional radiomic and deep learning methods, as well as human radiologists, in differentiating between PE and normotensive pregnancies. This highlights the potential of AI-driven imaging techniques to enhance early detection of PE, providing a more objective and reproducible assessment of placental abnormalities. Additionally, the automatic segmentation of placental MRI achieved a high Dice coefficient of 0.917, indicating strong agreement between automated and manual segmentation.
Predicting Foetal Growth Restriction in Preeclampsia
FGR is a major complication of PE, significantly increasing the risk of neonatal morbidity and long-term developmental impairments. Differentiating between PE cases that will progress to FGR and those that will not is a critical aspect of clinical management. The study found that DLR models effectively identified placental heterogeneity in PE pregnancies with FGR, demonstrating higher accuracy than conventional imaging techniques. The DLR approach achieved an area under the curve (AUC) of 0.918 when distinguishing PE with FGR from normotensive pregnancies, and an AUC of 0.742 when differentiating PE with FGR from PE without FGR.
Placental MRI findings indicated that the progression to FGR in PE pregnancies was associated with significant textural and geometric changes, captured effectively through radiomic and deep learning features. By providing a quantitative measure of placental dysfunction, DLR offers a more reliable and objective method for predicting which PE pregnancies are likely to develop FGR. This could enable earlier intervention, helping clinicians tailor antenatal care and monitor high-risk pregnancies more effectively. The ability of DLR to outperform human radiologists in distinguishing these subgroups further supports its potential role in improving prenatal screening and surveillance.
Clinical Implications and Future Directions
Integrating DLR-based placental MRI into routine prenatal care could significantly enhance the early detection and monitoring of PE pregnancies. By offering an automated and reproducible assessment of placental health, this approach has the potential to complement existing screening methods and improve risk stratification. The study’s findings suggest that DLR signatures could serve as an imaging-based biomarker for PE, reducing reliance on subjective radiological interpretation and improving diagnostic consistency across different clinical settings.
However, several challenges remain before DLR can be widely implemented in obstetric practice. Standardisation of imaging protocols is essential to ensure consistency across different MRI systems and healthcare institutions. Further multicentre validation studies are needed to confirm the generalisability of the findings and refine the DLR algorithms for clinical application. Additionally, integration of DLR with existing maternal-foetal assessment tools, such as Doppler ultrasound and biochemical markers, may enhance its predictive power and provide a more comprehensive risk assessment framework.
Future research should focus on refining AI models, incorporating larger datasets and exploring the potential of multimodal imaging approaches that combine MRI with other diagnostic techniques. Addressing these challenges will be crucial in translating DLR from research settings into clinical practice, ensuring that its benefits can be realised in routine prenatal care.
Deep learning radiomics represents a significant advancement in prenatal imaging, providing an automated, objective and highly accurate method for detecting PE and its complications. By leveraging placental MRI analysis, DLR offers a powerful tool for identifying high-risk pregnancies and predicting the likelihood of FGR development in PE cases. The study’s findings highlight the potential of AI-driven imaging to improve maternal-foetal outcomes by enabling earlier diagnosis and personalised clinical management. While further validation and refinement are required, the integration of DLR into obstetric care could mark a transformative step forward in maternal-foetal medicine, enhancing both diagnostic precision and patient care.
Source: European Journal of Radiology
Image Credit: iStock
