Deep learning-based regional radiomics analysis in placenta accreta spectrum evaluation

The University of Texas Southwestern Medical Center received a $467,542 project grant from the National Institute of Child Health and Human Development (NICHD) under the Child Health and Human Development Extramural Research program (CFDA 93.865), awarded August 15, 2025, with completion targeted for July 31, 2027. The award supports development of a fully automated, open-source computational pipeline utilizing deep learning algorithms and regional radiomics analysis to objectively assess hysterectomy risk in patients with placenta accreta spectrum (PAS)—a serious obstetric condition characterized by abnormal placental adherence to the uterine wall. The research addresses critical limitations in current imaging modalities (ultrasound and MRI) that rely on subjective interpretation and lack quantitative predictive measures.

The project delivers two primary technical products: (1) a novel deep learning-based segmentation algorithm to automatically isolate at-risk placental regions within the lower uterine segment on sagittal T2-weighted MRI images, leveraging anatomical landmarks such as the L5-S1 disc space; and (2) predictive models developed through extraction and analysis of regional radiomic features—including texture, shape, and location-based parameters—to improve diagnostic accuracy over existing methods. The University of Texas at Dallas serves as a sub-awardee, with Dr. Baowei Fei contributing as Co-Investigator to supervise development of quantitative image analysis and deep learning-based segmentation methodologies. These deliverables are designed to enhance diagnostic accuracy, facilitate improved surgical planning, and optimize delivery timing for high-risk PAS patients.

Abstract

Placenta accreta spectrum (PAS) is a serious obstetric condition characterized by abnormal adherence of the placenta to the uterine wall, leading to significant maternal morbidity and mortality during delivery. The increasing incidence of PAS correlates with the rise in cesarean deliveries. Current imaging modalities, such as ultrasound and magnetic resonance imaging (MRI), rely heavily on subjective interpretation, lack quantitative measures, and require extensive training. There is an urgent need for improved diagnostic and quantitative predictive tools, particularly in assessing PAS that is severe enough to require hysterectomy. This project aims to develop a fully automated, open-source pipeline utilizing deep learning (DL) algorithms and regional radiomics analysis to objectively assess the risk of hysterectomy in patients with high- risk PAS. By focusing on the lower uterine segment (LUS), the most common site of prior cesarean uterine scars and PAS pathology, the proposed research seeks to enhance diagnostic accuracy, facilitate better surgical planning, and optimize delivery timing. In aim 1, we will implement and test a novel DL-based segmentation algorithm to automatically isolate the 'at-risk' placental region within the LUS on sagittal T2W MRI images. Leveraging the anatomical marker of the pelvic inlet, the L5-S1 disc space, this will be the first approach to automatically segment presumed pathological areas of PAS. In aim 2, we will extract and analyze regional radiomic features from the segmented LUS placental region to develop predictive models for assessing the risk of hysterectomy. By incorporating advanced radiomic features, including texture, shape, and location-based parameters, we aim to improve the predictive accuracy over existing methods. The proposed research is innovative in its application of automated DL segmentation and regional radiomics to a critical area of obstetric care. The development of an open-source pipeline will provide a valuable resource for clinicians and researchers, promoting widespread adoption and collaboration. Successful completion of this project has the potential to significantly impact patient outcomes by enabling earlier and more accurate identification of PAS cases that require advanced surgical intervention. The multidisciplinary team, comprising experts in radiology, biomedical engineering, and medical imaging analysis, is well- positioned to achieve these goals. The project leverages extensive institutional resources, including a large, well-characterized dataset of placental MRI scans and state-of-the-art computational facilities. In summary, this project addresses a critical GAP in the management of PAS by introducing an objective, automated tool for risk assessment. The anticipated outcomes will advance the field of prenatal imaging and contribute to improved maternal and fetal health outcomes.