Cost Reduction of Creating Likelihood Map for Automatic Polyp Detection Using Image Pyramid

Abstract

Endoscopes are widely used for polyp diagnosis in medical scene. Automatic polyp detection is needed for reducing the burden on doctors and patients.This paper proposes an automatic polyp detection method. First, polyp candidate regions are extracted by creating likelihood maps using edge and color information of the endoscope image. Second, features of candidate regions are extracted by using Convolutional Neural Network as a feature extractor. Finally, polyps are detected by classifying polyp existence of the candidate region using Support Vector Machine as a classifier.Evaluation of the proposed method is conducted using actual endoscope images.