Effective deep learning based segmentation and classification in wireless capsule endoscopy images

Wireless capsule endoscopy is a noninvasive wireless imaging method that has grown in popularity over the last several years. One of the efficient and effective ways for examining the gastrointestinal system is using WCE. It sends a huge number of images in a single examination cycle, making abnormality analysis and diagnosis extremely difficult and time-consuming. As a result, in this research, we provide the Expectation maximum (EM) algorithm, a revolutionary deep-learning-based segmentation approach for GI tract recognition in WCE images. DeepLap v3+ can extract a variety of features including colour, shape, and geometry, as well as SURF (speed-up robust features). Thus the Lenet 5 based classification can be made in the extracted images. The effectiveness of the performances is carried out on a publicly available Kvasir-V2 dataset, on which our proposed approach achieves 99.12% accuracy 98.79% of precision, 99.05% of recall and 98.49% of F1- score when compared to existing approaches. Effectiveness benefits are demonstrated over multiple current state-of-the-art competing techniques on all performance variables we evaluated, especially mean of Intersection Over Union (IoU), IoU for background, and IoU for the entire class.