Histopathological Gastric Cancer Detection on GasHisSDB Dataset Using Deep Ensemble Learning

Area of Science:

• Oncology
• Medical Imaging
• Computer Science

Background:

• Gastric cancer is a major global health concern, with early detection crucial for improving patient outcomes.
• Current histopathological analysis for gastric cancer is manual, time-consuming, and prone to variability.
• Deep learning offers potential for automated analysis but individual models have feature extraction limitations.

Purpose of the Study:

• To develop and evaluate ensemble deep learning models for improved gastric cancer detection.
• To overcome the limitations of single deep learning models in feature extraction for classification.
• To enhance the accuracy and efficiency of computer-aided diagnosis in gastric cancer histopathology.

Main Methods:

• Proposed ensemble models combining decisions from multiple deep learning models.
• Utilized the publicly available Gastric Histopathology Sub-size Image Database for evaluation.
• Tested performance across different sub-databases, including varying patch sizes.

Main Results:

• The top 5 ensemble model achieved state-of-the-art detection accuracy across all tested sub-databases.
• A highest detection accuracy of 99.20% was recorded on the 160 × 160 pixels sub-database.
• Demonstrated that ensemble models effectively extract critical features from smaller image patches.

Conclusions:

• Ensemble deep learning models show significant promise for accurate gastric cancer detection.
• This approach can assist pathologists, leading to more efficient and reliable histopathological analysis.
• The study contributes to advancing early gastric cancer detection, potentially improving patient survival rates.