Bleeding Segmentation Based on a Bleeding Feature Engagement Module in Wireless Capsule Endoscopy

The capacity to visualize the gastrointestinal tract has been transformed by the diagnostic technique known as Wireless Capsule Endoscopy (WCE). Due to the enormous number of images produced, however, the time required to analyze these frames is relatively expensive. Rapider clinical treatments may be made possible by the computerized identification of abnormalities in these images. In this research, a U-Net model customized with a Bleeding Feature Engagement (BFE) module is suggested for bleeding segmentation in WCE images to decrease the strain on the physician. Every skip connection leverages the BFE module to create distinct bleeding borders. It is an inception-based structure that employs three various convolution layer types to produce a range of decoder feature sets. Comparing the suggested system to cutting-edge methods, its results demonstrate the best segmentation performance. On an openly available bleeding WCE dataset, the proposed model additionally provided dice scores of $\mathbf{9 1. 7 6 \%}$ and intersection over union (IoU) of 84.79%.