Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Placenta Accreta Spectrum Detection using Multimodal Deep Learning

Sumaiya Ali, Areej Alhothali, Sameera Albasri, Ohoud Alzamzami, Ahmed Abduljabbar, Muhammad Alwazzan

TL;DR

This study addresses the need for accurate prenatal PAS detection by developing a multimodal deep learning framework that fuses 3D MRI and 2D ultrasound information at the feature level. The approach uses a hybrid 3D DenseNet121–ViT MRI backbone and a 2D ResNet50 US backbone, with an intermediate fusion that concatenates modality-specific features and passes them through an MLP to predict PAS. On a held-out test set of patient-maired MRI-US pairs, the multimodal model achieves 92.5% accuracy and AUC 0.927, outperforming MRI-only and US-only models, demonstrating that MRI and US provide complementary diagnostic cues. The work contributes a dual-modality dataset and end-to-end fusion framework, showing potential to enhance prenatal PAS risk stratification and guide clinical management, while acknowledging limitations such as single-center data and a relatively small multimodal sample for fusion.

Abstract

Placenta Accreta Spectrum (PAS) is a life-threatening obstetric complication involving abnormal placental invasion into the uterine wall. Early and accurate prenatal diagnosis is essential to reduce maternal and neonatal risks. This study aimed to develop and validate a deep learning framework that enhances PAS detection by integrating multiple imaging modalities. A multimodal deep learning model was designed using an intermediate feature-level fusion architecture combining 3D Magnetic Resonance Imaging (MRI) and 2D Ultrasound (US) scans. Unimodal feature extractors, a 3D DenseNet121-Vision Transformer for MRI and a 2D ResNet50 for US, were selected after systematic comparative analysis. Curated datasets comprising 1,293 MRI and 1,143 US scans were used to train the unimodal models and paired samples of patient-matched MRI-US scans was isolated for multimodal model development and evaluation. On an independent test set, the multimodal fusion model achieved superior performance, with an accuracy of 92.5% and an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.927, outperforming the MRI-only (82.5%, AUC 0.825) and US-only (87.5%, AUC 0.879) models. Integrating MRI and US features provides complementary diagnostic information, demonstrating strong potential to enhance prenatal risk assessment and improve patient outcomes.

Placenta Accreta Spectrum Detection using Multimodal Deep Learning

TL;DR

This study addresses the need for accurate prenatal PAS detection by developing a multimodal deep learning framework that fuses 3D MRI and 2D ultrasound information at the feature level. The approach uses a hybrid 3D DenseNet121–ViT MRI backbone and a 2D ResNet50 US backbone, with an intermediate fusion that concatenates modality-specific features and passes them through an MLP to predict PAS. On a held-out test set of patient-maired MRI-US pairs, the multimodal model achieves 92.5% accuracy and AUC 0.927, outperforming MRI-only and US-only models, demonstrating that MRI and US provide complementary diagnostic cues. The work contributes a dual-modality dataset and end-to-end fusion framework, showing potential to enhance prenatal PAS risk stratification and guide clinical management, while acknowledging limitations such as single-center data and a relatively small multimodal sample for fusion.

Abstract

Placenta Accreta Spectrum (PAS) is a life-threatening obstetric complication involving abnormal placental invasion into the uterine wall. Early and accurate prenatal diagnosis is essential to reduce maternal and neonatal risks. This study aimed to develop and validate a deep learning framework that enhances PAS detection by integrating multiple imaging modalities. A multimodal deep learning model was designed using an intermediate feature-level fusion architecture combining 3D Magnetic Resonance Imaging (MRI) and 2D Ultrasound (US) scans. Unimodal feature extractors, a 3D DenseNet121-Vision Transformer for MRI and a 2D ResNet50 for US, were selected after systematic comparative analysis. Curated datasets comprising 1,293 MRI and 1,143 US scans were used to train the unimodal models and paired samples of patient-matched MRI-US scans was isolated for multimodal model development and evaluation. On an independent test set, the multimodal fusion model achieved superior performance, with an accuracy of 92.5% and an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.927, outperforming the MRI-only (82.5%, AUC 0.825) and US-only (87.5%, AUC 0.879) models. Integrating MRI and US features provides complementary diagnostic information, demonstrating strong potential to enhance prenatal risk assessment and improve patient outcomes.
Paper Structure (29 sections, 21 equations, 10 figures, 7 tables)

This paper contains 29 sections, 21 equations, 10 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Data Collection
  4. Data Preparation and Preprocessing
  5. MRI Preprocessing
  6. Data Distribution and Augmentation
  7. Unimodal MRI Dataset
  8. Unimodal US Dataset
  9. Multimodal Dataset
  10. Data Augmentation
  11. Deep Learning Model Architectures
  12. MRI Feature Extraction Model
  13. Ultrasound Feature Extraction Model
  14. Multimodal Intermediate Fusion Architecture
  15. Experimental Setup and Training
  16. ...and 14 more sections

Figures (10)

  • Figure 1: Example of non-PAS MRI (A), PAS-positive MRI (B), non-PAS US (C), PAS-positive US (D), non-PAS multimodal pair (E), and PAS-positive multimodal pair (F).
  • Figure 2: Preprocessing pipeline for standardizing MRI scans. Ali_2025
  • Figure 3: Preprocessing pipeline for standardizing US images.
  • Figure 4: The selected MRI model architecture of 3D DenseNet121-ViT. Ali_2025
  • Figure 5: The selected US model architecture of 2D ResNet50.
  • ...and 5 more figures