GINet: A Deep Learning Approach to Gastrointestinal Image Classification

The gastrointestinal (GI) tract is one of the important parts of our body, performing the primary functions of digestion and absorption of food and nutrients. Many types of diseases hamper the normal operation of the GI tract and cause life-threatening conditions like cancer. Early detection of these diseases from endoscopic images is an essential step towards human well-being. A computer-aided system capable of automatically detecting diseases can reduce the pressure on doctors in underdeveloped countries who currently do this manually. This work proposes a newly modified deep convolutional neural network model, gastrointestinal network (GINet), which is trained to perform this task, achieving training and validation accuracies of 92.22 % and 87.41 %, respectively, on an 8 -class dataset. The model was trained on the Kvasir V2 dataset, containing 8,000 images, with 1,000 images per class. Our goal is to classify GI images automatically with the least error. Based on the results, it is evident that this model is well-suited to handle GI tract data, enabling early detection of abnormalities.