A Comparative Analysis of Image Descriptors for Histopathological Classification of Gastric Cancer

2025-03-21

AI Summary

Gastric cancer is highly prevalent worldwide and associated with poor prognosis (approximately 20% five-year survival rate); current histopathological diagnosis relies heavily on manual assessment, resulting in low efficiency and high inter-observer variability—highlighting an urgent need for lightweight, robust automated decision-support tools. This study systematically evaluates a fine-tuning-free feature engineering paradigm for binary classification of gastric cancer tissue histopathology images on the GasHisSDB dataset. We compare handcrafted features—including LBP, GLCM, and HOG—with deep features extracted from pre-trained VGG and AlexNet models, coupled with shallow classifiers such as Random Forest (RF) and SVM. To our knowledge, this is the first comprehensive investigation into the generalizability and robustness of diverse feature–classifier combinations in gastric cancer histopathology. RF with handcrafted features achieves an F1-score of 93.4%, significantly outperforming existing fine-tuning-free methods and demonstrating the clinical viability and practical utility of lightweight feature engineering for deployable辅助 diagnostic systems.

Technology Category

• Machine Learning: Feature Construction/Reformulation
• Computer Vision: Other Foundations of Computer Vision
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Web Mining and Content Analysis: Normalization, clustering, classification, and summarization of Web text
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

Gastric cancer ranks as the fifth most common and fourth most lethal cancer globally, with a dismal 5-year survival rate of approximately 20%. Despite extensive research on its pathobiology, the prognostic predictability remains inadequate, compounded by pathologists' high workload and potential diagnostic errors. Thus, automated, accurate histopathological diagnosis tools are crucial. This study employs Machine Learning and Deep Learning techniques to classify histopathological images into healthy and cancerous categories. Using handcrafted and deep features with shallow learning classifiers on the GasHisSDB dataset, we offer a comparative analysis and insights into the most robust and high-performing combinations of features and classifiers for distinguishing between normal and abnormal histopathological images without fine-tuning strategies. With the RF classifier, our approach can reach F1 of 93.4%, demonstrating its validity.

Problem

• Classify gastric cancer histopathological images accurately
• Compare handcrafted and deep features for classification
• Improve diagnostic accuracy and reduce pathologists' workload

Innovation

• Machine Learning and Deep Learning techniques
• Handcrafted and deep features combination
• RF classifier achieves 93.4% F1 score