Gland segmentation in gastric histology images: detection of intestinal metaplasia

Abstract

Gastric cancer is one of the most frequent causes of cancer-related deaths worldwide. Gastric intestinal metaplasia (IM) of the mucosa of the stomach has been found to increase the risk of gastric cancer and is considered as one of the precancerous lesions. Therefore, early detection of IM may have a valuable role in histopathological risk assessment regarding the possibility of progression to cancer. Accurate segmentation and analysis of gastric glands from the histological images plays an important role in the diagnostic confirmation of IM. Thus, in this paper, we propose a framework for segmentation of gastric glands and detection of IM. More specifically, we propose the GAGL-Net for the segmentation of glands. Then, based on two features of the extracted glands we classify the tissues into normal and IM cases. The results showed that the proposed gland segmentation approach achieves an F1 score equal to 0.914. Furthermore, the proposed methodology shows great potential for the IM detection achieving an accuracy score equal to 96.6%. To evaluate the efficiency of the proposed methodology we used a publicly available dataset and we created the GAGL dataset consisting of 59 Whole Slide Images (WSI) including both IM and normal cases.