Related Experiment Video
Updated: Jun 18, 2025
07:46
Author Spotlight: Advancing Reproductive Immunology with a Protocol for the Quantitative Evaluation of Endometrial Immune Cells
Published on: October 13, 2023
1.2K
A deep learning framework for predicting endometrial cancer from cytopathologic images with different staining styles
Ruijie Wang1, Qing Li2, Guizhi Shi3
1School of Automation Science and Engineering, Xi'an Jiaotong University, Xi'an, Shaanxi, P.R. China.
Plos One
|July 31, 2024
Summary
This study introduces an automated deep learning framework for endometrial cancer screening using cytopathology images. The novel approach effectively handles variations in staining styles, improving diagnostic accuracy and efficiency for early detection.
Area of Science:
- Computational pathology
- Medical imaging analysis
- Artificial intelligence in oncology
Background:
- Manual analysis of endometrial cytopathology images for cancer screening is time-consuming and subjective.
- Variations in staining styles across different medical centers hinder the generalization of deep learning models.
- Accurate and efficient screening methods are crucial for timely clinical treatment of endometrial cancer.
Purpose of the Study:
- To develop a robust automated screening framework for endometrial cancer adaptable to diverse staining styles in cytopathology images.
- To provide an objective diagnostic reference for cytopathologists, enhancing screening efficiency and accuracy.
- To introduce the XJTU-EC dataset, the first cytopathology dataset with segmentation and classification labels for endometrial cancer.
Main Methods:
- A two-stage deep learning framework was proposed: CM-UNet for cell clump segmentation and ECRNet for classification.
- CM-UNet incorporates channel attention (CA) and multi-level semantic supervision (MSS) to overcome staining variations.
- ECRNet utilizes contrastive learning with momentum-based updating and labeled memory banks to minimize false negatives.
Main Results:
- CM-UNet demonstrated excellent performance in segmenting cell clumps from images with varying staining styles.
- ECRNet achieved high accuracy (98.50%), precision (99.32%), and sensitivity (97.67%) on the XJTU-EC test set.
- The proposed framework outperformed existing classical models in endometrial cancer screening from cytopathologic images.
Conclusions:
- The developed automated framework robustly screens endometrial cancer from cytopathology images, irrespective of staining variations.
- This advancement contributes to improved early diagnosis and clinical treatment of endometrial cancer.
- The study provides a valuable dataset and a novel deep learning approach for computational pathology research.
