Related Experiment Video
Updated: Jun 11, 2025
07:32
Author Spotlight: Investigating Immune Cell Dynamics in the Tumor Microenvironment — Challenges and Innovations in Cancer Prognosis
Published on: April 12, 2024
1.2K
Enhanced gastric cancer classification and quantification interpretable framework using digital histopathology
Muhammad Zubair1, Muhammad Owais2, Tahir Mahmood3
1Faculty of Information Technology & Computer Science, University of Central Punjab, Lahore, Punjab, Pakistan.
Scientific Reports
|September 28, 2024
Summary
A new gastric histology classification and segmentation (GHCS) framework improves computer-aided diagnosis for gastric cancer detection. This novel system enhances classification and segmentation of histopathology images, offering robust performance across datasets.
Area of Science:
- Digital pathology
- Computational oncology
- Medical image analysis
Background:
- Computer-aided diagnosis (CAD) systems are crucial for analyzing histopathology images in gastric cancer (GC) detection.
- Existing CAD models for GC classification and segmentation require further improvement for enhanced accuracy and robustness.
Purpose of the Study:
- To introduce a novel Gastric Histology Classification and Segmentation (GHCS) framework.
- To achieve modest yet meaningful improvements in GC classification and segmentation over current CAD models.
- To enhance the robustness and generalizability of CAD systems for gastric histopathology analysis.
Main Methods:
- Developed a novel GHCS framework incorporating an expectation-maximizing Naïve Bayes classifier with an updated Gaussian Mixture Model.
- Employed an improved Fuzzy c-means method for histopathology image segmentation.
- Utilized adaptive focusing on pertinent image characteristics for improved model performance.
- Incorporated Grad-CAM visualizations for model interpretability.
Main Results:
- Achieved high classification accuracies: 98.87% (validation) and 98.47% (test) on one dataset, and 97.28% (validation) and 97.31% (test) on another.
- Demonstrated slight but consistent improvement over existing techniques in gastric histopathology image classification.
- Outperformed state-of-the-art segmentation models with a Dice coefficient of 65.21% and a Jaccard index of 60.24% in GC histopathology image segmentation.
- The model showed robustness in handling variability and generalizing to different datasets.
Conclusions:
- The proposed GHCS framework offers improved performance in gastric cancer detection through enhanced classification and segmentation of histopathology images.
- The framework's robustness and ability to generalize across datasets suggest potential for improved clinical application.
- The interpretability provided by Grad-CAM enhances trust and understanding for clinicians using the CAD system.
