A Study on the Computer Aided Diagnosis System for Early Gastric Cancer Lesion Based on EfficientNetV2-L through Data Filtering

作     者：Lee, Han-Sung Cho, Hyun-Chong 

作者机构：Dept. of Electronics Engineering Interdisciplinary Graduate Program for BIT Medical Convergence Kangwon National University Korea Republic of Dept. of Interdisciplinary Graduate Program for BIT Medical Convergence Kangwon National University Korea Republic of 

出 版 物：《Transactions of the Korean Institute of Electrical Engineers》 (Trans. Korean Inst. Electr. Eng.)

年 卷 期：2022年第71卷第9期

页      面：1259-1265页

核心收录：

基　　金：This work was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MSIT) (No. 2017R1E1A1A03070297). This research was supported by Basic Science Research Program through the National Research Foundation of Korea(NRF) funded by the Ministry of Education (No. 2022R1I1A3053872) 

主　　题：Computer aided diagnosis 

摘      要：Gastric cancer is a common cancer worldwide, especially in Korea. Early diagnosis is very important to increase the full recovery rate. However, early gastric cancer has no special symptoms and is a disease that even experts find difficult to diagnose in gastroscopy. Therefore, in this paper proposed a computer-aided diagnosis(CADx) for early gastric cancer diagnosis using EfficientNetV2-L. Due to the nature of medical data, it is difficult to collect a large amount of data. The data used for training was augmented using Cifar10 policy of the Google s AutoAugment. Additionally, the augmented image was used as an input to the model trained with the original dataset and filtered according to the classification threshold. EfficientNetV2 is a classification network designed Training-NAS that can learn the feature of lesions with a small number of parameters. As a result, EfficientNetV2 set to the threshold value of 0.9 achieved the performance of accuracy 0.943 for early gastric cancer and abnormal image classification. The AUC value also increases from 0.972 to 0.991, showing that the data filtering method of this study was effective for improvement of classification performance. © 2022 Korean Institute of Electrical Engineers. All rights reserved.