Deep Learning based Automatic Detection of Intestinal Hemorrhage Using Wireless Capsule Endoscopy Images

Abstract

The development of computer-aided diagnosis (CAD) systems for detecting GI bleeding in Wireless Capsule Endoscopy (WCE) image videos has become a hot topic in science, with the goal of reducing physician workload. Due to their constrained feature representation capacity, existing methods provide inadequate precision for bleeding detection. CNN is efficient and a generalised robust system is created using attribute selection and ensemble learning. A supervised learning ensemble will be built in this paper to detect the bleeding in WCE images. With this model, the best possible combination of attributes needed to identify bleeding symptoms in endoscopy images will be discovered. Both the public and the private datasets are trained and tested in our work and our model produced an accuracy of 95.7% with 97.1% sensitivity and 94.6% specificity.