Gastrointestinal disease detection using multiclass CNN from endoscopy images

Abstract

Gastrointestinal diseases refer to diseases of the gastro intestinal tract and include; gastro esophagitis, ulcerative colitis, gastric lesions among others. These diseases are best diagnosed at an early stage, and therefore, the need for enhanced diagnostic methods that allow for same. This paper proposes the use of Convolutional Neural Networks (CNNs) in discerning gastrointestinal diseases from endoscopy images in an effort to make diagnosis more accurate as well as faster. The research assesses different CNN architectures – DenseNet201 and InceptionResNetV2, VGG19, and the CNN models: CNN01 and CNN02 developed by the authors specifically for this research investigation. A comprehensive set of images involving gastrointestinal diseases such as esophagitis, ulcerative colitis, dyed resection margins, normal pylorus were used for training and testing. The experimental findings also reveal that the built own CNN01 model yielded the highest average accuracy of 98.56%. These research outcomes show the effectiveness of the high-architectural CNN in achieving good classification of gastrointestinal diseases from endoscopic pictures. In light of the study of CNN, potential future applications of CNN are presented to demonstrate how CNN could improve diagnostic accuracy within gastroenterology and reduce the necessity of operation. The future of this kind of research will involve handling of a diverse dataset, working to reduce the class imbalance issue, and enhance the explain ability of the model.