Gastric Cancer Image Classification: A Comparative Analysis and Feature Fusion Strategies.

Area of Science:

• Oncology
• Computational Pathology
• Medical Imaging

Background:

• Gastric cancer is a leading cause of cancer death globally, with poor survival rates.
• Current histopathological diagnosis is limited by workload and potential errors, necessitating automated tools.
• Accurate prognostic prediction is crucial but challenging in gastric cancer management.

Purpose of the Study:

• To develop and evaluate Machine Learning (ML) and Deep Learning (DL) models for automated classification of gastric histopathological images.
• To compare the effectiveness of different feature extraction methods (handcrafted vs. deep features) and shallow learning classifiers.
• To assess the performance of these models on the GasHisSDB dataset without fine-tuning.

Main Methods:

• Utilized the GasHisSDB dataset comprising healthy and cancerous gastric histopathological images.
• Extracted both handcrafted and deep features from the images.
• Employed shallow learning classifiers, including Support Vector Machines (SVM), for image classification.
• Conducted comparative analysis of feature-classifier combinations and cross-magnification experiments.

Main Results:

• Achieved a high accuracy of 95% using the SVM classifier with feature fusion strategies.
• Demonstrated the effectiveness of combining different feature types for improved classification.
• Cross-magnification experiments showed promising results, with accuracies nearing 80% and 90% on images with varying resolutions.

Conclusions:

• Machine learning, particularly with feature fusion and SVM, offers a highly accurate automated solution for gastric cancer histopathological classification.
• The developed models show potential for improving diagnostic efficiency and accuracy, aiding pathologists.
• The approach demonstrates robustness across different image magnifications, suggesting broader applicability.