Multi-class Gastrointestinal Diseases improved diagnosis based on Ensemble and Transfer Learning.

From the mouth to the anus, the digestive system is a long tube made up of multiple hollow organs. To heal, people with gastrointestinal illnesses need to receive the correct therapy and be diagnosed as soon as possible. In biomedical applications, there has been a recent surge in research on classifying endoscopic images for the identification of gastrointestinal tract disorders. Deep learning algorithms, particularly Deep Convolutional Neural Networks, are utilized to diagnose major gastrointestinal conditions like ulcerative colitis, polyps, and esophagitis..Multiple approaches were utilized to enhance the performance of the diagnosis system, but they are still not enough due to a lack of datasets and the complexity of designing new algorithms. So, to achieve this goal and overcome these problems, in this proposed work, at first data was preprocessed in manner training and testing, and data augmentation was applied to maximize training operation based on more data, then are four different experiments, the first two experiments were based on two novel different Convolution Neural Network models, which yielded accuracy values of 0.75125 and 0.99875, respectively; ensemble learning was used in the third experiment, yielding a total accuracy of 0.995, and transfer learning, which used the well-known pretrained model (VGG16), produced an accuracy of 0.9800 in the final experiment. Based on these four experimental methods, results have better performance evaluation parameters in accuracy, specificity, and F1 score than other recent related approaches for gastrointestinal diseases multi-class diagnosis. Also, these techniques can be applied to other multi-class diseases.