Polyps Detection in Colonoscopies

Abstract

A colonic polyp is a growth in the lining of the colon or rectum and can be detected through colonoscopies. The efficiency of colonoscopies depends on the number of polyps detected. However, detecting and classifying polyps is difficult, tedious, and prone to error. Knowing that this process’s performance is far from perfect, the objective of this project is to help colonoscopists in the detection of polyps during the medical intervention, using Deep Learning (DL) alongside the image recognition capabilities of Convolutional Neural Networks (CNN) models that can process colonoscopy images at high speed in real-time. In this paper, were tested different state-of-the-art CNNs using a transfer learning approach, achieving an average accuracy of 95,70% in the polyp detection task. Multiple public datasets were used in this study to train, test, and evaluate the classifiers. The negative class included images representative of healthy tissue as well as other pathologies, so the models would not mistake other diseases as polyps.