AI-Based Detection and Localization of Gastrointestinal Polyps by Using Deep Learning, Transfer Learning and the Fusion of These Techniques

Abstract

As per the World Health Organization, colorectal cancer (CRC) is the third most common cancer found in men and the second most common cancer among women. Colonoscopy is a brilliant norm for screening of CRC and for detecting and localizing polyps. Adenoma detection rate (ADR) – the measure of hit rate of endoscopist to locate and remove polyps – has become a key indicator for performance of endoscopist. Artificial intelligence-assisted examination for detection and localization of polyps may significantly control the high missed diagnosis rate. The convolutional neural network (CNN)-based diagnosis systems adapt exceptionally fast for the computer-aided diagnosis for detection, localization and grouping of polyps. In this chapter, we have made an attempt to study the outcomes independently for CNN and pre-trained model by Visual Geometry Group (VGG) with 16 layered architecture (VGG16), and then to study the outcomes of fusion of CNN and VGG16. However, we have obtained better results from the fusion of CNN and VGG16 compared to the results obtained from independent execution of these architectures.