Exponential Chef based Optimization Enabled Deep Learning for Polyp Frame Detection and Polyp Segmentation using Colonoscopy Videos

Abstract

Identifying and describing the abnormal develop ment of tissues as they appear in medical frame images of endoscopy or colonoscopy is known as polyp segmentation. Accuracy of Polyp segmentation depends on high-quality imaging data, which may not always be available resulting in misiden tification and compromisation in segmentation precision. The most effective method of detecting colorectal cancer is still a colonoscopy. Nonetheless, notable miss rates for polyps have been seen, especially in cases with many tiny adenomas. This offers a chance to make use of computer-aided technologies to assist physicians in their work and lower the amount of polyps overlooked .In order to address these issues the Exponential chef-based optimization (ECBO) algorithm is introduced as it will help to train the Deep Maxout Network and Lightweight TransUNet weights to increase the accuracy of segmenting polyps from video frames. We first collect the thousand-frame image dataset known as (KVasir-SEG), and we preprocess the data by using Adaptive median filter with CLAHE i.e. Contrast Limit adaptive histogram equalization (CLAHE). CLAHE enhances local contrast, making the image more suitable for subsequent processing like edge preservation by the adaptive median filter also Adaptive median filtering can effectively reduces the noise. On the KVasir-SEG dataset, A polyp frame detection is carried out by Deep-Maxout Network. The experimental results show that ECBO based Lightweight TransUNet performs better than earlier methods in terms of Dice, Interception over union (IoU), Recall, and Precision. These results demonstrate the strong gen eralization strength and learning capability of the proposed that ECBO based Lightweight TransUNet approach, which makes it a desirable substitute for real-world applications with large data variations