CONTRASTIVE LEARNING: A NOVEL APPROACH FOR GASTROINTESTINAL DISEASE CLASSIFICATION

Abstract

A wide range of serious health issues, such as pain, discomfort, poor digestion, and in certain situations, chronic difficulties, are brought on by structural or functional abnormalities of the digestive system, which are referred to as gastrointestinal (GI) diseases. The digestive system problems cause severe disruptions to people's quality of life. Early detection improves patient care and lowers medical costs, and it is strongly related to successful treatment. Through the examination of endoscopic images, artificial intelligence (AI), and intense learning, have shown outstanding potential in detecting gastrointestinal disorders. However, problems like incorrectly categorising conditions that are identical in appearance and limited model generalizability still present serious difficulties. Therefore, Further research is necessary to develop robust solutions that address these challenges. This work proposed a GI classification model using a self-supervised contrastive learning technique using the ResNet-18 as a CNN backbone. The proposed model performs exceptionally well in classifying GI disease on the publicly available Kvasir dataset. The proposed model efficiently classifies eight categories of GI disease with an excellent overall 100% accuracy. The placement of ROC curves for all classes close to the top-left corner of the graph and AUC ≈ 1 for all classes reflects the strong classification ability of the proposed model. The suggested model is a state-of-the-art method for precise and effective diagnosis, outperforming current AI models in the classification of gastrointestinal diseases.