Advertisement

A deep learning method to assist with chronic atrophic gastritis diagnosis using white light images

      Abstract

      Background

      Chronic atrophic gastritis is a common preneoplastic condition of the stomach with a low detection rate during endoscopy.

      Aims

      This study aimed to develop two deep learning models to improve the diagnostic rate.

      Methods

      We collected 10,593 images from 4005 patients including 2280 patients with chronic atrophic gastritis and 1725 patients with chronic non-atrophic gastritis from two tertiary hospitals. Two deep learning models were developed to detect chronic atrophic gastritis using ResNet50. The detection ability of the deep learning model was compared with that of three expert endoscopists.

      Results

      In the external test set, the diagnostic accuracy of model 1 for detecting gastric antrum atrophy was 0.890. The identification accuracies for the severity of gastric antrum atrophy were 0.773 and 0.590 in the internal and external test sets, respectively. In the other two external sets, the detection accuracies of model 2 for chronic atrophic gastritis were 0.854 and 0.916, respectively. Deep learning model 1′s ability to identify gastric antrum atrophy was comparable to that of human experts.

      Conclusion

      Deep-learning-based models can detect chronic atrophic gastritis with good performance, which may greatly reduce the burden on endoscopists, relieve patient suffering, and improve the disease's detection rate in primary hospitals.

      Keywords

      Abbreviations:

      CAG (Chronic Atrophic Gastritis), CNAG (Chronic Non-Atrophic Gastritis), AI (Artificial Intelligence), DL (Deep Learning), CNN (Convolutional Neural Network), ADR (Adenoma Detection Rate), OLGA (Operative Link on Gastritis Assessment)

      1. Introduction

      Chronic atrophic gastritis (CAG) is a common gastric disease that has been recognized as a precancerous lesion [
      • Hatakeyama M.
      Helicobacter pylori CagA and gastric cancer: a paradigm for hit and-run carcinogenesis.
      ]. The population prevalence of CAG generally ranged from 2.1% to 8.2% worldwide. In China, the prevalence of CAG was 55.7 per 1000 patients with gastric cancer [
      • Wang R.
      • Chen X.Z.
      Prevalence of atrophic gastritis in southwest China and predictive strength of serum gastrin-17: a cross-sectional study (SIGES).
      ]. Patients with CAG have an increased risk of gastric adenocarcinoma [
      • Banks M.
      • Graham D.
      • Jansen M.
      • et al.
      British Society of Gastroenterology guidelines on the diagnosis and management of patients at risk of gastric adenocarcinoma.
      ]. Increased severity of atrophy and extent of intestinal metaplasia are associated with an increased risk of cancer [
      • Banks M.
      • Graham D.
      • Jansen M.
      • et al.
      British Society of Gastroenterology guidelines on the diagnosis and management of patients at risk of gastric adenocarcinoma.
      ,
      • Rugge M.
      Gastric Cancer Risk in Patients with Helicobacter pylori Infection and Following Its Eradication.
      . Therefore, early detection of CAG and the evaluation of the severity of atrophy would be helpful in identifying populations at high risk for gastric cancer.
      Gastric atrophy and intestinal metaplasia are usually diagnosed by endoscopic examination and subsequent biopsy. However, the procedure has several limitations. First, making a diagnosis using standard endoscopy alone (white light endoscopy) is reliant on highly specialized doctors; second, the false-negative rate of biopsy is high; and third, some patients have contraindications for biopsy. A multicenter survey in China showed that the biopsy sensitivity of pathologic atrophic gastritis was only 42% [
      • Du Y.
      • Bai Y.
      • Xie P.
      • et al.
      Chronic gastritis in China: a national multi-center survey.
      ]. Although multi-point biopsy could increase the positivity rate, it would also likely increase the risk of hemorrhage and other complications.
      Artificial intelligence (AI) has profoundly transformed the way we live our lives, and this has been especially evident in the field of medical science. In recent years, with the accumulation of abundant medical data and improvement in computer algorithms, AI development has made great strides in assisting diagnoses and predicting prognosis – for example, AI-based cardiovascular disease risk identification systems and diabetic retinopathy detection systems have been created [
      • Du T.
      • Xie L.
      • Zhang H.
      • et al.
      Training and validation of a deep learning architecture for the automatic analysis of coronary angiography.
      ,
      • Gulshan V.
      • Peng L.
      • Coram M.
      • et al.
      Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs.
      .
      Deep learning (DL), a branch of AI, allows computational models composed of multiple processing layers to learn representations of data with multiple levels of abstraction. Convolutional Neural Network (CNN) is the most important basic unit in the field of DL. The typical architecture of a CNN usually includes a convolutional layer, pooling layer, rectified linear unit (ReLU) layer, and a fully connected layer. When images and video files are input into a CNN, the CNN can extract useful image features automatically and exploit the softmax function (also known as multinomial logistic regression) for classification [
      • LeCun Y.
      • Bengio Y.
      • Hinton G.
      Deep learning.
      ].
      These methods have dramatically improved the state-of-the-art speech recognition, visual object recognition, object detection, and many other domains, such as drug discovery and genomics. DL discovers intricate structures in large datasets using the back-propagation algorithm in order to dictate how a machine should change its internal parameters. These parameters compute the representation in each layer from the representation in the previous layer. DL have brought about breakthroughs in image, video, speech, and audio processing [
      • Zhou L.
      • Li Q.
      • Huo G.
      • et al.
      Image Classification Using Biomimetic Pattern Recognition with Convolutional Neural Networks Features.
      ].
      Image and video data, such as CT, MRI and endoscopic examinations, are crucial for diagnosing digestive system diseases. Hence, the application of DL has significant prospects for development. Progress has already been made in the study of digestive system diseases. DL-based methods can detect not only early esophageal cancer and early gastric cancer but also H. pylori infections [
      • Guo L.
      • Xiao X.
      • Wu C.
      • et al.
      Real-time automated diagnosis of precancerous lesions and early esophageal squamous cell carcinoma using a deep learning model (with videos).
      ,
      • Ikenoyama Y.
      • Hirasawa T.
      • Ishioka M.
      • et al.
      Detecting early gastric cancer: comparison between the diagnostic ability of convolutional neural networks and endoscopists.
      ,
      • Nakashima H.
      • Kawahira H.
      • Kawachi H.
      • et al.
      Helicobacter pylori Artificial intelligence diagnosis of infection using blue laser imaging-bright and linked color imaging: a single-center prospective study.
      ]. By utilizing colonoscopy images, a DL-based computer-assisted diagnosis system could help doctors significantly increase the adenoma detection rate (ADR) [
      • Wang P.
      • Berzin T.M.
      • Glissen Brown J.R.
      • et al.
      Real-time automatic detection system increases colonoscopic polyp and adenoma detection rates: a prospective randomised controlled study.
      ,
      • Wang P.
      • Liu X.
      • Berzin T.M.
      • et al.
      Effect of a deep-learning computer-aided detection system on adenoma detection during colonoscopy (CADe-DB trial): a double-blind randomised study.
      . The DL model can detect and grade ulcers in Crohn's disease and ulcerative colitis based on capsule endoscopy images and videos [
      • Stidham R.W.
      • Liu W.
      • Bishu S.
      • et al.
      Performance of a Deep Learning Model vs Human Reviewers in Grading Endoscopic Disease Severity of Patients with Ulcerative Colitis.
      ,
      • Klang E.
      • Barash Y.
      • Margalit R.Y.
      • et al.
      Deep learning algorithms for automated detection of Crohn's disease ulcers by video capsule endoscopy.
      ,
      • Barash Y.
      • Azaria L.
      • Soffer S.
      • et al.
      Ulcer severity grading in video capsule images of patients with Crohn's disease: an ordinal neural network solution.
      ]. However, there are few reports of DL being used for CAG detection. Thus, our study aimed to use the DL method to improve the diagnostic rate of CAG using traditional white light images.

      2. Materials and methods

      This study conformed to the ethical guidelines of the Declaration of Helsinki and was approved by the ethics committee of the Third Xiangya Hospital of Central South University (No.21160). According to national legislation and institutional requirements, informed consent was waived by the Ethics Committee of the Third Xiangya Hospital of Central South University and Changsha Central Hospital due to the retrospective nature of this study.

      2.1 Datasets and preprocessing

      This retrospective study was conducted at the Gastroenterology Department of the Third Xiangya Hospital of Central South University and Changsha Central Hospital. Endoscopic images of patients with CAG and chronic non-atrophic gastritis (CNAG) were collected between January 2015 and December 2020. We collected 10,593 images of 4005 patients, including 2280 patients with CAG and 1725 patients with CNAG, from two hospitals. All CAG cases were confirmed by pathological examination. First, the biopsies were scored semi-quantitatively by two pathologists with >10 years of experience each. The scoring was conducted according to the updated Sydney classification system and the Operative Link on Gastritis Assessment (OLGA) method, which combines the degree and range of intestinal metaplasia and gastric mucosa. Second, the degree of atrophy was calculated on the basis of the degree of gland reduction: mild atrophy - the number of glands was reduced by less than 1/3, moderate atrophy - the number of glands was reduced between 1/3 and 2/3, severe atrophy - the number of glands was reduced by more than 2/3, with only a few remaining glands or their complete disappearance. Cases were excluded when the two pathologists did not reach an agreement.
      All gastroscopic reports of CAG and CNAG were reviewed by two experts with more than 10 years of experience in gastroscopy each. All reports included a series of endoscopic images and text descriptions. We first reviewed the text descriptions to determine the location of the performed biopsy and then extracted the matched images. Images were excluded if they were of poor quality or depicted CAG or CNAG with other lesions, such as ulcers and cancerous lesions.
      Images were collected and classified according to the hospital in which the procedure took place and the location, atrophy, and severity of the atrophy. A total of 3885 images were collected from 2280 CAG patients, including 2931 images of the antrum, 590 images of the angle, 309 images of the corpus, 45 images of the fundus, 1825 images of mild atrophic gastritis, 1902 images of moderate atrophic gastritis, and 38 images of severe atrophic gastritis. A total of 6708 images were collected from 1725 CNAG patients, including 1650 images of the antrum, 1595 images of the angle, 1729 images of the corpus, and 1734 images of the fundus. Information about the gastroscopic image dataset is shown in Supplementary Table 1.
      All gastroscopic images were obtained by doctors during their daily clinical operations. Four different devices were used: Olympus Evis Lucera 260/290 (Tokyo, Japan) and FUJIFILM EG-530WR/601 WR (Tokyo, Japan). The gastric mucosa images were obtained using the white light model; all images were in JPG format with ordinary resolution, and the size of each image was 50–300 kb.
      In order to identify gastric antrum atrophy and its severity, we established two datasets to develop model 1. Dataset 1 contained all antrum images of CAG and CNAG from the Third Xiangya Hospital. Three hundred antrum images were randomly selected from the Changsha Central Hospital as dataset 2. The details of these two datasets are summarized in Supplementary Table 2.
      In order to identify CAG, we established three datasets for developing model 2. Dataset 3 contained all CAG and CNAG images from the Third Xiangya Hospital. We employed several data augmentation methods to increase the size of dataset 3, including flip and 90° rotation. Two hundred CAG and CANG images were randomly selected from Changsha Central Hospital as dataset 4. A total of 132 images without a gastric antrum were randomly selected from Changsha Central Hospital as Dataset 5. The details of these three datasets are summarized in Supplementary Table 3.
      The relevant codes and models can be freely accessed at https://github.com/philiplaw1984/chronic-atrophic-gastritis/.

      2.2 Deep learning networks

      Our study tested the current mainstream architectures, namely Vgg-16, ResNet-50, DenseNet169, and Inception_V3 networks. Resnet-50 showed the best performance for CAG identification. Therefore, we selected Resnet-50 as the basic network to develop models 1 and 2.
      Resnet-50 is a classic CNN. The ``50'' in the name ResNet-50 refers to an architecture with 7 × 7 convolutional layers and 16 building blocks (each building block includes 3 convolutional layers), forming a total of 48 convolutional layers and a fully connected layer. One of the most important features of ResNet-50 is the shortcut connections, which skip one or more layers to solve the vanishing gradient problem in deep neural networks by allowing the gradient to flow through the layer. Fig. 1 shows the structure of the ResNet-50.
      Fig. 1
      Fig. 1The simple structure of the ResNet-50.
      1a - the structure of ResNet-50;
      1b - the structure of ID BLOCK;
      1c - the structure of CONV BLOCK.

      2.3 Experimental settings and evaluation values

      All experiments using deep learning for model training were performed on matpool, a GPU cloud platform, with an NVIDIA GeForce RTX 3090 GPU. The experimental environment was Python 3.8, CUDA 11.0, cuDNN 8.0, TensorFlow 2.4, Keras2.3.1, NVCC, and Ubuntu 18.04. The batch size was 64 and the number of epochs was 500. The main evaluation values were accuracy, sensitivity, and specificity.

      2.4 AI vs doctors

      To verify the performance of the DL model, we designed an AI-doctors comparison experiment. Fifty images were randomly selected from the test-set of datasets 1 and 2 as the test set of AI-doctors competition. We invited three gastroenterologists to counter DL model 1. All three experts had more than five years of experience in gastroscopy. We then compared the performances of the three experts with the performance of the AI on this small test set.

      3. Result

      3.1 Model 1 recognized gastric antrum atrophy and its severity

      In the test set of dataset 1, the diagnostic accuracy, sensitivity, and specificity of model 1 for gastric antrum atrophy were 0.902, 0.891, and 0.915, respectively. The accuracy of identifying the severity of gastric antrum atrophy was 0.773. To test the robustness and universality, we tested the performance of model 1 in dataset 2, an external test set. Accuracy, sensitivity, and specificity for gastric antrum atrophy were 0.890, 0.905, and 0.890, respectively. The accuracy of identifying the severity of gastric antrum atrophy was 0.590 in the external test set. The details of the results are shown in Fig. 2. The detection rates of CNAG and mild, moderate, and severe antrum atrophy in the internal test set were 91.5, 87.9, 90.4, and 91.7%, respectively. The detection rates of CNAG and mild, moderate, and severe antrum atrophy in the external test set were 89.0, 93.0, 87.8, and 90.0%, respectively.
      Fig. 2
      Fig. 2The details of model 1 recognize gastric antrum atrophy and its severity.

      3.2 Model 2 recognized chronic atrophic gastritis

      Our results showed that model 1 could recognize gastric antrum atrophy and its severity with a first-rate performance. However, chronic atrophic gastritis is not only limited to the antrum but also occurs in other locations in the stomach, including the angle, corpus, and fundus. To overcome this limitation, we developed model 2 to recognize CAG. In the test set of dataset 3, the diagnostic accuracy, sensitivity, and specificity of model 2 for CAG were 0.859, 0.875, and 0.854, respectively. To test the robustness and universality, we tested the performance of model 2 on dataset 4, an external test set; the accuracy, sensitivity, and specificity for CAG were 0.854, 0.870, and 0.850, respectively. We also tested the performance of model 2 in dataset 5, an external test set containing all the images of all gastric segments apart from the antrum. Accuracy, sensitivity, and specificity were 0.916, 0.912, and 0.920, respectively. The detailed results of the two external test sets are shown in Fig. 3.
      Fig. 3
      Fig. 3The details of model 2 recognize chronic atrophic gastritis.

      3.3 AI vs doctors

      The AI-doctors competition experiment showed that the ability of our DL model to identify gastric antrum atrophy was almost equal to those of the doctors. The results of the three experts and our DL model 1 of the AI-doctors competition test set are shown in Figs. 4 and 5.
      Fig. 4
      Fig. 4The details of three experts and deep learning model 1 on the artificial intellingence-doctors competition test-set.
      Fig. 5
      Fig. 5The artificial intelligence-doctors comparison results.

      4. Discussion

      Research on deep learning and chronic atrophic gastritis is relatively scarce. With the help of the DL model based on DenseNet, CAG can be diagnosed using high-resolution gastric atrophy images [
      • Zhang Y.
      • Li F.
      • Yuan F.
      • et al.
      Diagnosing chronic atrophic gastritis by gastroscopy using artificial intelligence.
      ]. A small study from Germany indicated that DL could diagnose atrophic gastritis with high accuracy [
      • Guimarães P.
      • Keller A.
      • Fehlmann T.
      • et al.
      Deep-learning based detection of gastric precancerous conditions.
      ]. Using gastric X-ray images, a Japanese research team developed a DL model for automatic CAG detection [
      • Kanai M.
      • Togo R.
      • Ogawa T.
      • et al.
      Chronic atrophic gastritis detection with a convolutional neural network considering stomach regions.
      ]. However, the previous studies also have several limitations. First, to achieve high accuracy, studies are prone to using high-resolution images or magnifying narrow-band images. Only a few studies have focused on ordinary white light images, which are the images most widely used in clinical practice. Second, current studies either focused on the gastric antrum or on the fundus and corpus. Few studies have focused on all stomach regions where chronic atrophic gastritis occurs. Third, most of the published studies were single-center ones. Due to limited data, many studies lacked external testing with which to evaluate the performance of the models.
      To overcome these limitations, we first collected abundant ordinary white light images of the antrum, angle, corpus, and fundus from two different medical centers. Subsequently, we established five independent datasets for developing and testing two DL models and one small dataset for AI-doctors competition. Then, we developed two DL models, one for recognizing gastric antrum atrophy and its severity and another for recognizing CAG. Both DL models could detect CAG with good performance. Finally, we verified that the ability of our DL model to identify gastric antrum atrophy was similar to that of trained endoscopists.
      Helicobacter pylori causes CAG with predominant localization in the gastric antrum [
      • Kuipers E.J.
      • Uyterlinde A.M.
      • Peña A.S.
      • et al.
      Increase of Helicobacter pylori-associated corpus gastritis during acid suppressive therapy: implications for long-term safety.
      ]. Many lines of evidence indicate that advanced atrophy is a risk factor for gastric cancer [
      • Kaji K.
      • Hashiba A.
      • Uotani C.
      • et al.
      Grading of Atrophic Gastritis is Useful for Risk Stratification in Endoscopic Screening for Gastric Cancer.
      ,
      • Choi I.J.
      • Kook M.C.
      • Kim Y.I.
      • et al.
      Helicobacter pylori Therapy for the Prevention of Metachronous Gastric Cancer.
      . Therefore, grading atrophic gastritis is useful for the early detection and treatment of gastric cancer. Our study developed a DL model with good performance for the detection and grading of gastric antrum atrophy. China has a high prevalence of Helicobacter pylori infection, resulting in a high incidence of gastric atrophy, gastric intestinal metaplasia, and gastric cancer [
      • Hooi J.K.
      • Lai W.Y.
      • Ng W.K.
      • et al.
      Global Prevalence of Helicobacter pylori Infection: systematic Review and Meta-Analysis.
      ,
      • Pan K.
      • Zhang L.
      • Gerhard M.
      • et al.
      A large randomised controlled intervention trial to prevent gastric cancer by eradication of Helicobacter pylori in Linqu County, China: baseline results and factors affecting the eradication.
      . For this reason, the Chinese Society of Gastroenterology has suggested that doctors should follow OLGA in clinical practice. The Chinese Society of Gastroenterology also recommends that patients with moderate and severe gastric atrophy with gastric intestinal metaplasia should be followed annually [
      The Chinese Society of Gastroenterology. China consensus report on chronic gastritis (2017, Shanghai).
      ]. Our initial assumption was that by including the assistance of the DL model, we could detect moderate and severe atrophic gastritis as early as possible. Moreover, if the follow-up biopsy confirmed that the patient had intestinal metaplasia, we could identify the high-risk gastric cancer population earlier. If the DL model predicts mild gastric atrophy, biopsy might be avoided.
      However, CAG not only occurs in the gastric antrum but also in other locations, including the angle, corpus, and fundus. Patients with CAG should be followed up with high quality endoscopy, meaning that doctors should carefully examine all gastric locations [
      • Pimentel-Nunes P.
      • Libânio D.
      • Marcos-Pinto R.
      • et al.
      Management of epithelial precancerous conditions and lesions in the stomach (MAPS II): european Society of Gastrointestinal Endoscopy (ESGE), European Helicobacter and Microbiota Study Group (EHMSG), European Society of Pathology (ESP), and Sociedade Portuguesa de Endoscopia Digestiva (SPED) guideline update 2019.
      ]. DL model 2 could detect CAG in any region of the stomach with excellent performance. This may greatly reduce the burden on doctors and improve the CAG detection rate.
      Conventional white light endoscopy has moderate sensitivity and specificity, as well as high inter-observer variability, and is therefore not sufficient for reliably diagnosing gastric atrophy [
      • Pimentel-Nunes P.
      • Libânio D.
      • Marcos-Pinto R.
      • et al.
      Management of epithelial precancerous conditions and lesions in the stomach (MAPS II): european Society of Gastrointestinal Endoscopy (ESGE), European Helicobacter and Microbiota Study Group (EHMSG), European Society of Pathology (ESP), and Sociedade Portuguesa de Endoscopia Digestiva (SPED) guideline update 2019.
      ,
      • Redéen S.
      • Petersson F.
      • Jönsson K.-.A.
      • et al.
      Relationship of gastroscopic features to histological findings in gastritis and Helicobacter pylori infection in a general population sample.
      . While advanced endoscopic techniques such as chromoendoscopy, magnification endoscopy, and confocal laser endomicroscopy could increase the detection rate, primary hospitals are often hindered by technical availability and high costs. However, using only conventional white light images, our DL models could detect CAG with outstanding sensitivity and specificity. Moreover, the computer hardware requirements of our DL models were very cost-effective since the input images were small and had low resolutions. With their good performance and relatively low cost, the DL models we developed may be more suitable for primary hospitals.
      Our DL model performed equally well in detecting CAG in two different hospitals. The results indicate that our model may be easily transferred from one hospital to another. While the accuracy of grading atrophic gastritis was decreased, this may largely be because endoscopy images from different hospitals were acquired by different devices and operators. It is unrealistic to expect every hospital, especially primary hospitals, to invent their own DL model. A possible solution to this issue is to use our pre-trained DL model and add small-sized images to train their own models. It is our future work to prospectively validate our methods and help other hospitals build their own models with our pre-trained model.
      In this study, ResNet-50 was selected as the framework since we required a deep network to extract the hidden features from endoscopic images, which are more challenging than other images. Macroscopically, as atrophy in CAG progresses, the gastric folds disappear. This loss of gastric rugae, combined with mucosal pallor and increased visibility of mucosal vessels, constitutes the main endoscopic features of atrophic gastritis [
      • Uedo N.
      • Yao K.
      Endoluminal Diagnosis of Early Gastric Cancer and Its Precursors: bridging the Gap Between Endoscopy and Pathology.
      ,
      • Kanemitsu T.
      • Yao K.
      • Nagahama T.
      • et al.
      Extending magnifying NBI diagnosis of intestinal metaplasia in the stomach: the white opaque substance marker.
      . On white light endoscopy, intestinal metaplasia typically appears as small, gray-white, elevated plaques surrounded by a mix of patchy pink and pale areas of mucosa, causing an irregular surface appearance [
      • Uedo N.
      Advanced Imaging in the Diagnosis of Gastric Intestinal Metaplasia: the Expert’s Approach.
      ]. ResNet-50 is suitable for extracting endoscopic findings of this complexity, since it can detect multiple features including texture, shape, and color [
      • Wei W.
      • Can T.
      • Xin W.
      • et al.
      Image Object Recognition via Deep Feature-Based Adaptive Joint Sparse Representation.
      ].
      Our study has several limitations. First, the accuracy of DL model 1 in grading atrophic gastritis, especially in the external test set, was unsatisfactory. To improve the performance of DL model, more training data from different medical centers and a more appropriate deep learning network architecture are required. Second, due to the lack of sufficient training data, DL model 2 was unable to grade the atrophic severity of the angle, corpus, and fundus. Further studies are needed to collect more atrophic gastritis images of these segments, especially images of severe atrophy. Finally, our research was a retrospective study, and more prospective studies are needed to verify the diagnostic ability of the DL model.

      5. Conclusion

      In conclusion, DL-based models could assist endoscopists in diagnosing chronic atrophic gastritis with high efficacy and low cost. This may greatly reduce the burden on endoscopists, relieve the suffering of patients, and allow for better detection of atrophic gastritis in primary hospitals.

      Declaration of Competing Interest

      The authors declare no competing interests.

      Availability of data and materials

      To protect patient privacy, the original data used to support the findings of this study could not be shared.

      Ethics approval and consent to participate

      This retrospective study conformed to the ethical guidelines of the Declaration of Helsinki and was approved by the ethics committee of the Third Xiangya Hospital of Central South University (No.21160). According to national legislation and institutional requirements, informed consent was waived by the Ethics Committee of the Third Xiangya Hospital of Central South University and Changsha Central Hospital due to the retrospective nature of this study.

      Consent for publication

      Not applicable.

      Funding

      Not applicable.

      Author contributions

      Ju Luo and Suo Cao wrote the manuscript's main text. Canxia Xu and Xin Liao contributed to the conception and design of the study. Ju Luo and Lin Peng analyzed the data and developed the deep learning model. Ning Ding created the tables and figures. Canxia Xu and Lin Peng revised some of the chapters with constructive comments. All authors have approved the final draft of the manuscript and have agreed to its submission for publication.

      Acknowledgment

      We thank Elsevier for their help in the English language editing of this manuscript. We also thank Wei Chen, jiangfang Chui and Chaiwei He for their help in the artificial Intelligence-doctors competition test-set.

      Appendix. Supplementary materials

      References

        • Hatakeyama M.
        Helicobacter pylori CagA and gastric cancer: a paradigm for hit and-run carcinogenesis.
        Cell Host Microbe. 2014; 15: 306-316
        • Wang R.
        • Chen X.Z.
        Prevalence of atrophic gastritis in southwest China and predictive strength of serum gastrin-17: a cross-sectional study (SIGES).
        Sci Rep. 2020; 10: 4523
        • Banks M.
        • Graham D.
        • Jansen M.
        • et al.
        British Society of Gastroenterology guidelines on the diagnosis and management of patients at risk of gastric adenocarcinoma.
        Gut. 2019; 68: 1545-1575
        • Rugge M.
        Gastric Cancer Risk in Patients with Helicobacter pylori Infection and Following Its Eradication.
        Gastroenterol Clin North Am. 2015; 44: 609-624
        • Du Y.
        • Bai Y.
        • Xie P.
        • et al.
        Chronic gastritis in China: a national multi-center survey.
        BMC Gastroenterol. 2014; 14: 21
        • Du T.
        • Xie L.
        • Zhang H.
        • et al.
        Training and validation of a deep learning architecture for the automatic analysis of coronary angiography.
        EuroIntervention. 2021; 17: 32-40
        • Gulshan V.
        • Peng L.
        • Coram M.
        • et al.
        Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs.
        JAMA. 2016; 316: 2402-2410
        • Zhou L.
        • Li Q.
        • Huo G.
        • et al.
        Image Classification Using Biomimetic Pattern Recognition with Convolutional Neural Networks Features.
        Comput Intell Neurosci. 2017; 20173792805https://doi.org/10.1155/2017/3792805
        • LeCun Y.
        • Bengio Y.
        • Hinton G.
        Deep learning.
        Nature. 2015; 521 (436–444)
        • Guo L.
        • Xiao X.
        • Wu C.
        • et al.
        Real-time automated diagnosis of precancerous lesions and early esophageal squamous cell carcinoma using a deep learning model (with videos).
        Gastrointest Endosc. 2020; 91: 41-51
        • Ikenoyama Y.
        • Hirasawa T.
        • Ishioka M.
        • et al.
        Detecting early gastric cancer: comparison between the diagnostic ability of convolutional neural networks and endoscopists.
        Dig Endosc. 2020; 33: 141-150
        • Nakashima H.
        • Kawahira H.
        • Kawachi H.
        • et al.
        Helicobacter pylori Artificial intelligence diagnosis of infection using blue laser imaging-bright and linked color imaging: a single-center prospective study.
        Ann Gastroenterol. 2018; 31: 462-468
        • Wang P.
        • Berzin T.M.
        • Glissen Brown J.R.
        • et al.
        Real-time automatic detection system increases colonoscopic polyp and adenoma detection rates: a prospective randomised controlled study.
        Gut. 2019; 68: 1813-1819
        • Wang P.
        • Liu X.
        • Berzin T.M.
        • et al.
        Effect of a deep-learning computer-aided detection system on adenoma detection during colonoscopy (CADe-DB trial): a double-blind randomised study.
        Lancet Gastroenterol Hepatol. 2020; 5: 343-351
        • Stidham R.W.
        • Liu W.
        • Bishu S.
        • et al.
        Performance of a Deep Learning Model vs Human Reviewers in Grading Endoscopic Disease Severity of Patients with Ulcerative Colitis.
        JAMA Netw Open. 2019; 2e193963
        • Klang E.
        • Barash Y.
        • Margalit R.Y.
        • et al.
        Deep learning algorithms for automated detection of Crohn's disease ulcers by video capsule endoscopy.
        Gastrointest Endosc. 2020; 91 (e2): 606-613
        • Barash Y.
        • Azaria L.
        • Soffer S.
        • et al.
        Ulcer severity grading in video capsule images of patients with Crohn's disease: an ordinal neural network solution.
        Gastrointest Endosc. 2021; 93: 187-192
        • Zhang Y.
        • Li F.
        • Yuan F.
        • et al.
        Diagnosing chronic atrophic gastritis by gastroscopy using artificial intelligence.
        Dig Liver Dis. 2020; 52: 566-572
        • Guimarães P.
        • Keller A.
        • Fehlmann T.
        • et al.
        Deep-learning based detection of gastric precancerous conditions.
        Gut. 2020; 69: 4-6
        • Kanai M.
        • Togo R.
        • Ogawa T.
        • et al.
        Chronic atrophic gastritis detection with a convolutional neural network considering stomach regions.
        World J Gastroenterol. 2020; 26: 3650-3659
        • Kuipers E.J.
        • Uyterlinde A.M.
        • Peña A.S.
        • et al.
        Increase of Helicobacter pylori-associated corpus gastritis during acid suppressive therapy: implications for long-term safety.
        Am J Gastroenterol. 1995; 90: 1401-1406
        • Kaji K.
        • Hashiba A.
        • Uotani C.
        • et al.
        Grading of Atrophic Gastritis is Useful for Risk Stratification in Endoscopic Screening for Gastric Cancer.
        Am J Gastroenterol. 2019; 114: 71-79
        • Choi I.J.
        • Kook M.C.
        • Kim Y.I.
        • et al.
        Helicobacter pylori Therapy for the Prevention of Metachronous Gastric Cancer.
        N Engl J Med. 2018; 378: 1085-1095
        • Hooi J.K.
        • Lai W.Y.
        • Ng W.K.
        • et al.
        Global Prevalence of Helicobacter pylori Infection: systematic Review and Meta-Analysis.
        Gastroenterology. 2017; 153: 420-429
        • Pan K.
        • Zhang L.
        • Gerhard M.
        • et al.
        A large randomised controlled intervention trial to prevent gastric cancer by eradication of Helicobacter pylori in Linqu County, China: baseline results and factors affecting the eradication.
        Gut. 2016; 65: 9-18
      1. The Chinese Society of Gastroenterology. China consensus report on chronic gastritis (2017, Shanghai).
        Chin J Gastroenterol. 2017; 22: 670-687
        • Pimentel-Nunes P.
        • Libânio D.
        • Marcos-Pinto R.
        • et al.
        Management of epithelial precancerous conditions and lesions in the stomach (MAPS II): european Society of Gastrointestinal Endoscopy (ESGE), European Helicobacter and Microbiota Study Group (EHMSG), European Society of Pathology (ESP), and Sociedade Portuguesa de Endoscopia Digestiva (SPED) guideline update 2019.
        Endoscopy. 2019; 51: 365-388
        • Redéen S.
        • Petersson F.
        • Jönsson K.-.A.
        • et al.
        Relationship of gastroscopic features to histological findings in gastritis and Helicobacter pylori infection in a general population sample.
        Endoscopy. 2003; 35: 946-950
        • Uedo N.
        • Yao K.
        Endoluminal Diagnosis of Early Gastric Cancer and Its Precursors: bridging the Gap Between Endoscopy and Pathology.
        Adv Exp Med Biol. 2016; 908: 293-316
        • Kanemitsu T.
        • Yao K.
        • Nagahama T.
        • et al.
        Extending magnifying NBI diagnosis of intestinal metaplasia in the stomach: the white opaque substance marker.
        Endoscopy. 2017; 49: 529-535
        • Uedo N.
        Advanced Imaging in the Diagnosis of Gastric Intestinal Metaplasia: the Expert’s Approach.
        Video Journal and Encyclopedia of GI Endoscopy. 2013; 1: 112-114
        • Wei W.
        • Can T.
        • Xin W.
        • et al.
        Image Object Recognition via Deep Feature-Based Adaptive Joint Sparse Representation.
        Comput Intell Neurosci. 2019; 20198258275