A computerized diagnostic model for automatically evaluating placenta accrete spectrum disorders based on the combined MR radiomics-clinical signatures

Hao Zhu, Xuan Yin, Haijie Wang, Yida Wang, Xuefen Liu, Chenglong Wang, Xiaotian Li, Yuanyuan Lu, Guang Yang, He ZhangScientific 2022 · DOI

Abstract

We aimed to establish a computerized diagnostic model to predict placenta accrete spectrum (PAS) disorders based on T2-weighted MR imaging. We recruited pregnant women with clinically suspected PAS disorders between January 2015 and December 2018 in our institution. All preoperative T2-weighted imaging (T2WI) MR images were manually outlined on the picture archive communication system terminal server. A nnU-Net network for automatic segmentation and the corresponding radiomics features extracted from the segmented region were applied to build a radiomics-clinical model for PAS disorders identification. Taking the surgical or pathological findings as the reference standard, we compared this computerized model's diagnostic performance in detecting PAS disorders. In the training cohort, our model combining both radiomics and clinical characteristics yielded an accuracy of 0.771, a sensitivity of 0.854, and a specificity of 0.750 in identifying PAS disorders. In the testing cohort, this model achieved a segmentation mean Dice coefficient of 0.890 and yielded an accuracy of 0.825, a sensitivity of 0.830 and a specificity of 0.822. In the external validation cohort, this computer-aided diagnostic model yielded an accuracy of 0.690, a sensitivity of 0.929 and a specificity of 0.467 in identifying placenta increta. In the present study, a machine learning model based on preoperative T2WI-based imaging had high accuracy in identifying PAS disorders in respect of surgical and histological findings.

Introduction

Placenta Accreta Spectrum (PAS) is a high-risk obstetric condition where the placenta attaches too deeply or invades the uterine wall. It is a leading cause of severe postpartum hemorrhage and maternal mortality. While Magnetic Resonance Imaging (MRI) is a powerful diagnostic tool, interpreting these scans is notoriously difficult and highly dependent on a radiologist’s specific expertise. In regions lacking specialized radiologists, the risk of misdiagnosis is significant. This study presents an automated "second opinion" system designed to reduce human subjectivity and provide consistent risk assessment by integrating imaging data with clinical history.

Methods

The researchers developed a multi-stage "cascaded" modeling process to identify PAS disorders:

  1. Automated Segmentation: The nnU-Net deep learning framework was used to automatically delineate placental boundaries on sagittal T2-weighted MR images.
  2. Radiomics Feature Extraction: From the segmented regions, 107 radiomics features were extracted, capturing mathematical descriptions of the placenta’s shape, intensity, and texture (e.g., gray-level heterogeneity) that are often imperceptible to the naked eye.
  3. Clinical Integration: The most predictive radiomics features were combined with three key clinical variables: maternal age, gravida (number of pregnancies), and placental position.
  4. Model Construction: These features were integrated into a Logistic Regression (LR) model to create a "radiomics-clinical nomogram," providing a probability score for PAS.
Overview of the study workflow, including patient recruitment flowchart and the automated diagnostic pipeline.
The model integrates clinical features with three categories of radiomics features: first-order, shape, and texture.

Datasets

The study utilized a retrospective dataset of 512 pregnant women from a single institution in China, confirmed via surgical or pathological findings:

  • Training Cohort: 358 cases (110 PAS, 248 normal) from 2015–2017.
  • Internal Testing Cohort: 154 cases (47 PAS, 107 normal) from 2018.
  • External Validation Cohort: 28 cases (15 PAS, 13 normal) from a separate hospital using different MRI hardware (1.5T and 3.0T) to test generalizability.

Results

The nnU-Net achieved a mean Dice coefficient of 0.890, indicating high agreement with expert manual segmentation. In the internal testing cohort, the combined model achieved an accuracy of 0.825, a sensitivity of 0.830, and a specificity of 0.822. Notably, the model's Area Under the Curve (AUC) of 0.849 outperformed the 0.744 achieved by human radiologists. In external validation, the model maintained a high sensitivity of 0.929, ensuring few cases were missed, though specificity decreased to 0.467.

Discussion

A key finding was the predictive power of texture features (such as GLCM and GLDM), which likely capture intra-placental heterogeneity—physical changes like abnormal vascularity that indicate invasive placentation. The synergy between radiomics and clinical data, particularly placental position, significantly boosted performance.

While the model is efficient—processing data in under two minutes—limitations include its reliance on a single MRI sequence (sagittal T2WI). Future work will explore multi-modal deep learning to incorporate additional views and sequences, and larger multi-center trials to improve robustness against hardware variations between 1.5T and 3.0T scanners.

Summary updated

Abstract

We aimed to establish a computerized diagnostic model to predict placenta accrete spectrum (PAS) disorders based on T2-weighted MR imaging. We recruited pregnant women with clinically suspected PAS disorders between January 2015 and December 2018 in our institution. All preoperative T2-weighted imaging (T2WI) MR images were manually outlined on the picture archive communication system terminal server. A nnU-Net network for automatic segmentation and the corresponding radiomics features extracted from the segmented region were applied to build a radiomics-clinical model for PAS disorders identification. Taking the surgical or pathological findings as the reference standard, we compared this computerized model's diagnostic performance in detecting PAS disorders. In the training cohort, our model combining both radiomics and clinical characteristics yielded an accuracy of 0.771, a sensitivity of 0.854, and a specificity of 0.750 in identifying PAS disorders. In the testing cohort, this model achieved a segmentation mean Dice coefficient of 0.890 and yielded an accuracy of 0.825, a sensitivity of 0.830 and a specificity of 0.822. In the external validation cohort, this computer-aided diagnostic model yielded an accuracy of 0.690, a sensitivity of 0.929 and a specificity of 0.467 in identifying placenta increta. In the present study, a machine learning model based on preoperative T2WI-based imaging had high accuracy in identifying PAS disorders in respect of surgical and histological findings.
BibTeX
@inproceedings{zhu2022computerized,
  author = {Hao Zhu and Xuan Yin and Haijie Wang and Yida Wang and Xuefen Liu and Chenglong Wang and Xiaotian Li and Yuanyuan Lu and Guang Yang and He Zhang},
  title = {{A computerized diagnostic model for automatically evaluating placenta accrete spectrum disorders based on the combined MR radiomics-clinical signatures}},
  booktitle = {Scientific},
  year = {2022},
  doi = {10.1038/s41598-022-14454-w},
}
Figure