PolypNet: A Lightweight CNN Framework for Early Detection of Colorectal Polyps Using Deep Learning

Abstract

Colorectal carcinoma is one of the most common reasons for carcinogenic death in the current world. Identifying the polyps that are present in the colon walls is one method to prevent this illness. However, a sparse number of research studies have been done to create a computer system that will detect the indisposition in the earlier stage. The enlargement of computer vision technology has accelerated the process by retrieving helpful information from the correlated data. Nonetheless, it is important to create an untrammelled system that will be able to sport colon polyps with better accuracy and training cost. In this research, we have delineated a Convolutional Neural Network (CNN) to emphasise Adenomatous, Hyperplastic and Serrated Lesions. The experiment of the network on the basis dataset has achieved an accuracy of 99.95% within a training time of only 18 minutes and 59 seconds. Stable learning efficiency was attained by the six-layer CNN with max-pooling and dropout regularisation.