Convolution Neural Network for Gastrointestinal Cancer Detection and Classification using Deep Learning
R. Sathishkumar, M. Nirmalraj, M. Govindarajan, J Jaisree, L. Haripriya, M. Santhiya
Abstract
Ailments of the gastrointestinal tract (GIT), including bleeding, ulcers, polyps, Crohn's disease, and cancer, are becoming more prevalent. Ulcers and bleeding in the small and large bowels are two of them that are particularly common. Medical experts find the manual diagnostic process to be both time-consuming and difficult. As a result, experts have suggested computerised techniques for the identification and classification of certain disorders. Medical video endoscopy produces a huge volume of images, thus it takes a long time for specialists to look over them all. Early detection and diagnosis of certain illnesses can result in successful therapy. SqueezeNet, ResNet-101, and DenseNet-169 are three deep learning-based models that we introduce in this paper and assess for their capacity to diagnose a dataset of lower gastrointestinal disorders. There are 5,000 images in the Kvasir dataset, similarly distributed among five different lower gastrointestinal conditions, including ulcerative colitis, polyps, normal cecum, and normal pylorus. The deep feature vector is processed using the softmax activation function, which divides the input images into five groups. Surprisingly, all of the convolutional neural network (CNN) models had exceptional performance, with DenseNet-169 reaching accuracy, specificity, precision, recall, and F1 Score values of 97.8%, 98.07%, 97.5%, 96.02% and 97.6%.