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

Placenta Accreta Spectrum (PAS) comprises a range of conditions defined by abnormal trophoblastic invasion into the myometrium, including placenta accreta, increta, and percreta. This disorder is a leading cause of severe maternal morbidity, frequently resulting in massive postpartum hemorrhage (PPH), peripartum hysterectomy, and associated complications. Despite its severity, PAS remains undiagnosed before delivery in a significant proportion of cases, ranging from half to two-thirds in some series, leading to unexpected massive hemorrhage and increased maternal morbidity. Currently, prenatal diagnosis relies on a multi-modal approach. Although ultrasound (US) remains the first-line screening tool, the conventional assessment of MRI signs for PAS frequently relies on qualitative features. While recent literature has begun to explore MRI radiomics in pregnancies with suspected PAS, the overall efficacy of this technique and its limits are still unclear. This study aims to bridge this gap by synthesizing the diagnostic and prognostic performance of MRI-based AI models for the prediction of PAS and the associated PPH.