Classification and localization of gastric cancer using Multi-Information Fusion Network

Abstract

Diagnosing and differentiating gastric cancer cells from stomach ulcers requires high-domain expertise and is time-consuming. Furthermore, medical image processing requires extremely high segmentation accuracy, which may lack interpretability and therefore, cannot be trusted by professional doctors for clinical application Hence, a diagnosis support system is proposed to aid pathologists and gastroenterologists that employs a deep learning network known as Multi-Information Fusion Network (MIFNET) for detecting cancer cells from histopathological images. The proposed system will offer assistance to gastroenterologists by displaying the severity level of the cancer, thereby helping them in choosing the appropriate treatment to increase the patient survival rates for those with stomach cancer. In contrast to conventional MIFNET segmentation functionality, the Fusion net layer of MIFNET was modified with VGG16 for classifying the malignant lesions in the gastric lining. The proposed MIFNET-based model results in a segmentation accuracy of 83% and a classification accuracy of 99%. Moreover, it is proven that the model achieved better prediction accuracy than other deep learning models such as U-Net and V-Net.

Keywords

• gastric cancer
• mifnet
• deep learning
• image segmentation
• u-net
• v-net
• classification
• histopathological images