Detecting Gastro-Intestinal Cancer from Wireless Capsule Endoscopy Images using Efficient Net Model

Recently polyps and ulcers have become a fatal form of gastrointestinal problem, and several researchers have looked into algorithms to diagnose gastrointestinal cancer lesions. This research work investigates the application of CNN to detect Gastro-Intestinal (GI) cancer from wireless capsule endoscopy (WCE) GI images. Five distinct CNN architectures EfficientNet, LeNet, GoogleNet, MobileNet-V2, and ResNet-50 have been studied, with two different setups i) original data with transfer learning ii) original data with data augmentation, and transfer learning. Simulations ran on 37,790 images with five different CNN architectures proved that the EfficientNet model exhibited superior performance of about 99.15% accuracy over all other CNN models, on the augmented dataset. Furthermore, the results from the experiments show that the EfficientNet model significantly outperforms previous methods in terms of classification accuracy.