Deep Learning Based Early Detection of Gastrointestinal Disorders

Abstract

Each year, around two million people die as a result of gastrointestinal problems worldwide. Medical experts use advanced video endoscopic imaging technology to diagnose and treat gastrointestinal issues such as ulcers, haemorrhaging, and polyps. However, due to the large volume of images produced by medical video endoscopy, healthcare personnel must devote significant time to assessing and interpreting these images. Given the difficulties of manual diagnosis, researchers are investigating computer-aided methods that can accurately and quickly identify all generated images. This research work compares three deep learning networks, GoogleNet, ResNet-50, and AlexNet, using the Kvasir dataset. AlexNet surpassed the other two deep learning networks, GoogleNet and ResNet-50, with a 97% accuracy rate, 96.8% sensitivity, 99.20% specificity, and 99.98% AUC (Area under the Curve).