Magnetic Resonance Prediction of Placenta Accreta Spectrum in Pregnancy: Validation of Radiomic Features

INTRODUCTION: Placenta accreta spectrum (PAS) or placental invasion into the uterine myometrium in women with previous cesarean delivery has become increasingly prevalent, affecting 1 in 500 pregnancies. Depending on the severity of placental invasion, patient management may involve a total hysterectomy at the time of cesarean delivery. We previously identified radiomic features from PAS-suspected MR images that were highly predictive of patient surgical outcome. After developing the algorithm, we needed to establish a validation cohort to test the radiomic model, a necessary step for reproducibility and generalization. PURPOSE: To validate the radiomic features previously found to correlate with the need for cesarean hysterectomy in patients with a high-risk for placenta accreta spectrum (PAS). METHOD: We performed an IRB approved retrospective review of 53 pregnancies from 2019 to 2023 of patients with clinically suspected PAS who had MR studies. Volumetric placental, uterus, and internal os regions of interest (ROIs) were manually segmented under the supervision of a board- certified radiologist with 30 years of OBGYN MR experience. Radiomic features were extracted following the image biomarker standardization initiative guideline using the pyRadiomics package. Placental Location within the Uterus (PLU) was described by a customized radiomic feature, determined as the angle between 2 vectors, with the tail being the epicenter of the uterus to the internal os and the epicenter of the placenta. RESULT: From our study, 25 patients (32.8±5.5 y/o) required cesarean hysterectomy with pathologic confirmation of PAS while 28 (33.0±5.6 y/o) underwent regular cesarean delivery. Estimated blood loss and gestational age at delivery were significantly different between groups (p<0.05). PLU and several other radiomic features were significantly different between those who required hysterectomy after cesarean delivery compared to those who did not in this new validation cohort of patients (p<0.05). DISCUSSION: We demonstrated that Placental Location within the Uterus (PLU) and other radiomic variables were predictive in detecting patients that needed hysterectomy in a validation cohort. This is an important step toward the development of a reliable, reproducible, and generalizable radiomic model for the prediction of PAS severe enough to result in cesarean hysterectomy.