br Background and Aims According to guidelines endoscopic re
Background and Aims: According to guidelines, endoscopic resection should only be performed for patients whose early gastric cancer invasion depth is within the mucosa or submucosa of the stomach regardless of Riboflavin phosphate node involvement. The accurate prediction of invasion depth based on endoscopic images is crucial for screening pa-tients for endoscopic resection. We constructed a convolutional neural network computer-aided detection (CNN-CAD) system based on endoscopic images to determine invasion depth and screen patients for endoscopic resection.
Methods: Endoscopic images of gastric cancer tumors were obtained from the Endoscopy Center of Zhongshan Hospital. An artificial intelligence–based CNN-CAD system was developed through transfer learning leveraging a state-of-the-art pretrained CNN architecture, ResNet50. A total of 790 images served as a development dataset and another 203 images as a test dataset. We used the CNN-CAD system to determine the invasion depth of gastric cancer and evaluated the system’s classification accuracy by calculating its sensitivity, specificity, and area under the receiver operating characteristic curve.
Conclusions: We constructed a CNN-CAD system to determine the invasion depth of gastric cancer with high accuracy and specificity. This system distinguished early gastric cancer from deeper submucosal invasion and minimized overestimation of invasion depth, which could reduce unnecessary gastrectomy. (Gastrointest Endosc 2019;89:806-15.)
Abbreviations: CAD, computer-aided detection; CI, confidence interval; CNN, convolutional neural network; EGC, early gastric cancer; ESD, endoscopic submucosal dissection; M, mucosa; ResNet, residual network; SM, submucosa.
DISCLOSURE: All authors disclosed no financial relationships relevant to this publication. Research support for this study was provided by the National Natural Science Foundation of China Nos. 81873552 (Li QL), 81470811 (Zhou PH), 81570595 (Xu MD), and 81670483 (Zhou PH); Major Project of Shanghai Municipal Science and Technology Committee nos. 18ZR1406700 (Li QL) and 16411950400 (Zhou PH); Chen Guang Program of Shanghai Municipal Education Commission no. 15CG04 (Li QL), Outstanding Young Doctor Training Project of Shanghai Municipal Commission of Health and Family Planning no. 2017YQ026 (Li QL), and the Project of Shanghai Municipal Commission of Health and Family Planning no. SHDC12016203 (Zhou PH).
*Drs Zhu and Wang contributed equally to this article.
Current affiliations: Endoscopy Center and Endoscopy Research Institute, Zhongshan Hospital, Fudan University, Shanghai, China (1), Department of Computer Science, University of California, Irvine, Irvine, California, USA (2).
Reprint requests: Quan-Lin Li, MD, or Ping-Hong Zhou, MD, PhD, Endoscopy Center and Endoscopy Research Institute, Zhongshan Hospital, Fudan University, Shanghai 200032, China.
If you would like to chat with an author of this article, you may contact Dr Li at [email protected] or Dr Zhou at [email protected] sh.cn.
Zhu et al Applying a CNN-CAD system to determine invasion depth for endoscopic resection
With the development of endoscopic interventions for early gastric cancer (EGC), endoscopic submucosal dissec-tion (ESD) is preferred because it is minimally invasive and requires a shorter hospital stay.1,2 However, guidelines state that ESD should only be performed on patients whose gastric cancer invasion depth is within the mucosa
(M) or the superficial portion of the submucosa (SM1) of the stomach irrespective of lymph node involvement.3 Therefore, accurate prediction of invasion depth based on endoscopic images is crucial for screening patients for endoscopic resection. However, there is currently no reliable method of determining invasion depth. According to guidelines, determination of the invasion depth of EGC is generally carried out using conventional endoscopy with a recommendation for additional indigo carmine dye spraying.4 The overall accuracy of conventional endoscopy is 69% to 79%5,6 and depends on the physician’s experience in recognizing endoscopic features. r> Computer-aided detection (CAD) is used for the diag-nosis and differential diagnosis of human diseases.7 Recently, convolutional neural network (CNN) systems using artificial intelligence have been applied to detect colorectal cancer8 and classify breast histopathologic images.9 CNNs have remarkable visual recognition capabilities, with recently developed architectures performing better than humans. In the 2015 ImageNet Challenge,10 a large-scale visual recognition challenge, the champion classifier using a residual network (ResNet) architecture achieved 96.4% accuracy, which was superior to human accuracy of 94.9%.