Related Experiment Video
Updated: Oct 13, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
682
Multiple instance convolutional neural network with modality-based attention and contextual multi-instance learning
Junming Jian1, Wei Xia2, Rui Zhang2
1Suzhou Institute of Biomedical Engineering and Technology, Chinese Academy of Sciences, Suzhou, Jiangsu 215163, China; Jinan Guoke Medical Engineering and Technology Development Co., Ltd., Jinan, Shandong 250109, China.
Artificial Intelligence in Medicine
|November 12, 2021
Summary
Differentiating malignant and borderline epithelial ovarian tumors (MEOTs and BEOTs) is vital for treatment. MAC-Net, a novel AI using multimodal MRI and multiple instance learning, shows promise in accurately distinguishing these tumors.
Area of Science:
- Oncology
- Radiology
- Artificial Intelligence
Background:
- Malignant epithelial ovarian tumors (MEOTs) are highly lethal, while borderline epithelial ovarian tumors (BEOTs) have a good prognosis.
- Accurate preoperative differentiation between MEOTs and BEOTs is critical for surgical planning and patient outcomes.
- Multimodal magnetic resonance imaging (MRI) is essential for diagnosis, but AI applications face resource limitations with 3D data.
Purpose of the Study:
- To develop and evaluate a novel AI method for differentiating between MEOTs and BEOTs using multimodal MRI.
- To improve the accuracy and efficiency of preoperative diagnosis for ovarian tumors.
Main Methods:
- Utilized a multiple instance learning (MIL) approach with a novel multiple instance convolutional neural network (MICNN) named MAC-Net.
- Incorporated a modality-based attention (MA) module for intelligent fusion of MRI data and a contextual MIL pooling layer (C-MPL) for enhanced prediction.
- Trained and tested MAC-Net on multimodal MRI datasets for BEOT/MEOT differentiation.
Main Results:
- MAC-Net achieved a superior area under the receiver operating characteristic curve (AUC) of 0.878.
- The proposed method outperformed several existing MICNN approaches in differentiating between BEOTs and MEOTs.
- The MA and C-MPL modules effectively improved the diagnostic performance by prioritizing MRI modalities and leveraging contextual information.
Conclusions:
- MAC-Net demonstrates high performance in differentiating malignant from borderline epithelial ovarian tumors using multimodal MRI.
- This AI-driven approach can serve as a valuable tool to assist clinicians in preoperative diagnosis, potentially improving patient management and quality of life.
