Development of Polyp Detection Technology by Analyzing Deep-learning

Abstract

This paper proposes a technique of guiding a clinician by recognizing colon polyp in patients in real time at the time of endoscopic examination. Automatic recognition of colon polyp by endoscopic image has applied Res-net’s advanced technique for calculating weight, existing deep learning method to enhance accuracy of recognizing colon polyp. Drawn result values provides information on shape of colon polyp guiding whether there is colon polyp and location of colon polyp in real time. Dataset used in developing analysis technique was collected from 5,000 cases from normal group and 5,000 cases from patients with colon polyp. For verification, accuracy was drawn by calculating confusion matrix with 10-fold cross validation. Accuracy of proposed technique was verified by drawing AUC of 0.96. Dataset was tested with data from control group and patient group in the ratio of 50%. We plan to develop deep learning technique that calculates the ratio of risk of occurrence of colorectal cancer. It is important to secure various test data such as images of colon polyp by type and whether there is adenocarcinoma. We plan to develop service that can predict ratio of risk of occurrence of colorectal cancer by applying relevant techniques automatically.