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

TL;DR

This work tackles automatic gastric cancer histopathology classification (healthy vs tumor) using two classes on the GasHisSDB dataset. It compares handcrafted image descriptors and deep features from pre-trained CNNs as inputs to shallow classifiers (DT, kNN, SVM, RF) without fine-tuning. Deep features from DenseNet-201 and EfficientNet B0 consistently deliver the best results, with Random Forest achieving the top $F1$ around $93.4\%$, approaching end-to-end fine-tuned models. The study demonstrates that strong performance can be achieved with general-purpose features, suggesting practical utility for automated diagnostics and guiding future work in feature fusion and Vision Transformer approaches.

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.

Figures (1)

• Figure 1: Sample images from the GasHisSDB dataset.