Related Experiment Videos
Deep learning for preoperative MRI-based endometrial cancer staging prediction
Caili Gong1, Yetong Qi1, Ying Su1
1School of Electronic Information Engineering, Inner Mongolia University, No. 24 Zhaojun Road, Hohhot, 010020, China.
Scientific Reports
|July 11, 2026
Summary
Researchers developed novel deep learning models, GCMF-UNet and MSFA-Net, to improve early-stage endometrial carcinoma diagnosis. These models enhance lesion segmentation and classification accuracy, aiding treatment planning and patient prognosis.
Area of Science:
- Oncology
- Medical Imaging
- Artificial Intelligence
Background:
- Endometrial carcinoma is a common female reproductive system malignancy.
- Accurate early-stage staging is crucial for treatment and prognosis.
- Traditional imaging and current deep learning models have limitations in precision and feature analysis.
Purpose of the Study:
- To enhance diagnostic precision for endometrial carcinoma.
- To overcome limitations of traditional imaging and existing deep learning models.
- To improve lesion segmentation and classification accuracy.
Main Methods:
- Proposed GCMF-UNet (Group Convolution and Multi-Scale Fusion U-Net) for segmentation.
- Introduced MSFA-Net (Multi-Scale Fusion Attention Network) for classification.
- MSFA-Net integrates ResNet-18, multi-scale feature aggregation, SE attention, and Swin Transformer.
Main Results:
- GCMF-UNet improved Accuracy from 90.1% to 94.2% and Recall from 89.3% to 94.8% over standard U-Net.
- MSFA-Net improved F1-score from 0.901 to 0.938 over baseline ResNet-18.
- Demonstrated enhanced capability in identifying critical lesion features.
Conclusions:
- GCMF-UNet and MSFA-Net effectively address limitations in conventional and deep learning diagnostics.
- The proposed architectures offer more accurate lesion segmentation and classification for endometrial carcinoma.
- These advancements provide a foundation for automated diagnosis and staging in endometrial carcinoma.