CNN-based Classification over Wireless Capsule Endoscopy Bleeding Images

To examine GI diseases and take painless images of the gut, wireless capsule endoscopy (WCE) is a useful technique. However, various issues, including effectiveness, tolerance, safety, and performance, make it difficult for widespread use and adaption. The purpose of this research is to develop a system that will analyse WCE photos, spot problems, and provide useful information to doctors. In order to categorise WCE bleedy photos, the authors of this paper recommend using a convolutional neural network (CNN). Precision, specificity, recall, F1 measure, and Cohen's kappa are used to assess the BIR's performance on a dataset consisting of 1650 WCE pictures during both training and testing. The obtained values of 0.993 for accuracy, 1.000 for precision, 0.994 for recall, 0.997 for the F1 score, and 0.995 for Cohen's kappa are encouraging. When applied to the Google-collected WCE picture dataset, the BIR model achieves an accuracy of 0.978, which is higher than that of state-of-the-art methods.