Feature Extraction of Colorectal Endoscopic Images for Computer-Aided Diagnosis with CNN
Takumi Okamoto, Masayuki Odagawa, Tetsushi Koide, Shinji Tanaka, Toru Tamaki, Bisser Raytchev, Kazufumi Kaneda, Shigeto Yoshida, Hiroshi Mieno · 2019
This paper introduces a feature extraction method for Narrow-Band Imaging (NBI) colorectal endoscopic images with Convolutional Neural Network (CNN) for Support Vector Machine (SVM) as a Computer-Aided Diagnosis (CAD) system. The proposed method using the result of pre-learned CNN as a feature extraction module on Bag-of-Features (BoF) framework and SVM inputs the result for classification. We estimated identification accuracy compare with the BoF framework and the proposed method. As an estimation result, we achieved that the proposed method can identify cancer or not with about over 90% accuracy.