LGFFN-GHI: A Local-Global Feature Fuse Network for Gastric Histopathological Image Classification

Abstract

Gastric cancer remains the third most common cause of cancer-related death. Histopathological examination of gastric cancer is the gold standard for the diagnosis of gastric cancer. However, manual pathology examination is time-consuming and laborious. Computer-aided diagnosis (CAD) systems can assist pathologists in diagnosing pathological images, thus improving the efficiency of disease diagnosis. In this paper, we propose a two-branch network named LGFFN-GHI, which can classify histopathological images of gastric cancer into two categories: normal and abnormal. LGFFN-GHI consists of two parallel networks, ResNet18 and Pvt-Tiny, which extract local and global features of microscopic gastric tissue images, respectively. We propose a feature blending module (FFB) that fuses local and global features at the same resolution in a cross-attention manner. This enables ResNet18 to acquire the global features extracted by Pvt-Tiny, while enabling Pvt-Tiny to acquire the local features extracted by ResNet18. We conducted experiments on a novel publicly available subsize image database of gastric histopathology (GasHisSDB). The experimental results show that LGFFN-GHI achieves an accuracy of 96.814%, which is 2.388% and 3.918% better than the baseline methods ResNet18 and Pvt-Tiny, respectively. Our proposed network exhibits high classification performance, demonstrating its effectiveness and future potential for the gastric histopathology image classification (GHIC) task.