Classification of Endoscopic Images and Identification of Gastrointestinal diseases

Every year around 20,00,000 people die because of diseases in the GI (Gastrointestinal) tract. To detect problems in the GI tract, endoscopy is used. Many Gastrointestinal diseases such as polyps, pylorus, colitis, etc. can be detected using video endoscopy. In this endoscopy, the doctor inserts a long tube (also known as an endoscope) through our throat and this tube gives us a lot of images of the GI tract which is then manually studied by the doctors to detect the disease. Therefore, through this paper, we are trying to automate the detection of diseases in the GI tract with the help of using neural networks with high accuracy and without the manual intervention of doctors.We have used Kvasir Dataset v2 here for our study. Here for feature extraction, we have used Global Averaged Concatenated MaxPooled features from ResNet50 and EfficientNetB7 models pre-trained on ImageNet. In the Classification stage, we employed a Voting Classifier, and we were able to obtain a maximum accuracy of 88.1875%. We also compared the results of Random Forest and Wildwood on the Kvasir Dataset in this paper.