A Deep Learning Based Ensemble Approach for Gastrointestinal Disease Detection with XAI

Gastrointestinal (GI) diseases are one of the most prevalent health issues in many parts of the world. It contributes to a number of other severe illnesses, such as esophagitis, ulcers, polyps, diarrhea, abdominal pain, abdominal swelling, gastrointestinal hemorrhage, intestinal blockage, malabsorption, or malnutrition. The early diagnosis and treatment of individuals with GI diseases are extremely important for their recovery due to these reasons. Gastroenterologists often use endoscopic images to detect GI abnormalities and because of the growing amount of patients and data, it is getting harder and harder. A deep learning-based approach can be a viable solution in this regard. In order to identify three various types of GI disorders, this study aims to offer an ensemble technique based on a unique Convolutional Neural Network architecture and other pretrained models. The ensemble model exhibits training and validation accuracy on Kvasir, a multi-class image dataset for computer assisted gastrointestinal illness identification, of 98.95% and 98.00%, respectively. Moreover, the results have been visualized using Explainable AI (XAI) technique called Local Interpretable Model-agnostic Explanations (LIME).