Early Detection of Gastric Cancer Using an Efficient Learning AI Scheme Involving a Convolutional Neural Network Model Keisuke Hori 1,2 , Satoko Takemoto 3 , Hideo Yokota 3 , Hiroaki Ikematsu 1,2 , Tomonori Yano 1,4 1Department of Gastroenterology and Endoscopy, National Cancer Center Hospital East, Kashiwa, Japan 2Division of Science and Technology for Endoscopy, Exploratory Oncology Research & Clinical Trial Center, National Cancer Center East, Kashiwa, Japan 3Image Processing Research Team, Center for Advanced Photonics, RIKEN, Wako, Japan 4Medical Device Innovation Center, National Cancer Center Hospital East, Kashiwa, Japan Keyword: 内視鏡診断 , CAD , 畳み込みニューラルネットワーク , 早期胃癌 , データ拡張 pp.423-431
Published Date 2021/4/25
DOI https://doi.org/10.11477/mf.1403202297
  • Abstract
  • Look Inside
  • Reference

 The use of CAD(computer-aided diagnosis)in gastrointestinal endoscopic imaging is rapidly progressing because of the development of CNN(the convolutional neural network). The application of CAD based on CNN has been developed from the recognition of anatomical locations and diagnoses for gastritis. Recently, researchers have reported favorable qualitative and quantitative results using CNN in the detection and diagnosis of gastric cancer. Whereas a large amount of data is required to construct a CNN model for CAD, we have developed a novel scheme involving automatic EGC detection with segmented localization. From a limited training dataset of only 300 images, we randomly cropped uncompressed small images and used data augmentation for transfer learning. Sensitivity rates for cases of 137 patients with cancer for one consecutive year and their related 462 images was 87.2%(image based)and 97.8%(case based). Our proposed scheme to use highly efficient learning can support CAD utilization in the early detection and future diagnoses of gastric cancer.

Copyright © 2021, Igaku-Shoin Ltd. All rights reserved.


電子版ISSN 1882-1219 印刷版ISSN 0536-2180 医学書院