Back to all papers

MRI-based radiomics and deep learning for the prenatal diagnosis of Placenta Accreta Spectrum: a systematic review and exploratory meta-analysis.

November 18, 2026pubmed logopapers

Authors

Sirico A,Pisani D,Sandullo L,Fatigati M,Mappa I,Maruotti GM,Rizzo G,Cobellis L

Affiliations (4)

  • Department of Woman, Child, and General and Specialized Surgery, University of Campania "Luigi Vanvitelli", Naples, Italy. Electronic address: [email protected].
  • Department of Woman, Child, and General and Specialized Surgery, University of Campania "Luigi Vanvitelli", Naples, Italy.
  • Department of Maternal and Child Health and Urological Sciences, Sapienza University of Rome, Rome, Italy.
  • Obstetrics and Gynecology Unit, Department of Public Health, University of Naples Federico II, Naples, Italy.

Abstract

accuracy of Placenta Accreta Spectrum (PAS) diagnosis using conventional magnetic resonance imaging (MRI) is still limited by subjective interpretation. Increasing evidence showed that radiomics and Deep Learning (DL) may improve diagnostic objectivity by extracting quantitative features from MRI data. This study aimed to synthesize the evidence on the diagnostic performance and clinical utility of these models. a systematic review and meta-analysis was conducted following PRISMA-DTA guidelines and PRISMA 2020 recommendations: PubMed, Scopus, and Web of Science databases from inception to December 2025. We searched for primary studies evaluating MRI-based radiomics or DL models for the binary diagnosis of PAS. Data were extracted to create 2 × 2 contingency tables and used a bivariate random-effects model to calculate pooled sensitivity, specificity, and to generate a summary ROC (SROC) curve. We assessed robustness via leave-one-out sensitivity analysis, and clinical utility using Fagan's nomogram and likelihood ratios. A secondary meta-analysis was performed for the prediction of postpartum hemorrhage (PPH) ≥1000 mL. twelve studies met the inclusion criteria (N = 1135 patients). The pooled sensitivity for PAS diagnosis was 0.83 (95% CI: 0.75-0.89) and pooled specificity was 0.82 (95% CI: 0.74-0.88). The summary AUC was 0.896, confirming high diagnostic accuracy. The positive likelihood ratio (LR+) was 4.61 and the negative likelihood ratio (LR-) was 0.21. Subgroup analyses based on AI model type (p > 0.05) and study design (p > 0.05) did not find significant differences between groups but revealed that single-center studies tended to report higher accuracy compared to multicenter studies. The secondary meta-analysis on massive PPH prediction (n = 4 studies, N = 353 patients: 165 PPH+ and 188 PPH-) yielded a pooled sensitivity of 0.88 (95% CI: 0.63-0.97), a pooled specificity of 0.78 (95% CI: 0.50-0.93), and an AUC of 0.901. MRI-based AI models demonstrate high exploratory pooled accuracy for the prenatal diagnosis of PAS in high-risk cohorts. These results suggest that these tools show promising exploratory accuracy and may provide decision support as an adjunct ("human-in-the-loop") in equivocal cases. However, their real-world clinical generalizability remains constrained by retrospective designs and index test threshold optimization.

Topics

Journal ArticleReview

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAISlice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.