Advertisement

Systematic review and meta-analysis: Artificial intelligence for the diagnosis of gastric precancerous lesions and Helicobacter pylori infection

  • E. Dilaghi
    Affiliations
    Department of Medical-Surgical Sciences and Translational Medicine, Sant'Andrea Hospital, Sapienza University of Rome, Via di Grottarossa, Roma 1035 - 00189, Italy
    Search for articles by this author
  • E. Lahner
    Affiliations
    Department of Medical-Surgical Sciences and Translational Medicine, Sant'Andrea Hospital, Sapienza University of Rome, Via di Grottarossa, Roma 1035 - 00189, Italy
    Search for articles by this author
  • B. Annibale
    Affiliations
    Department of Medical-Surgical Sciences and Translational Medicine, Sant'Andrea Hospital, Sapienza University of Rome, Via di Grottarossa, Roma 1035 - 00189, Italy
    Search for articles by this author
  • G. Esposito
    Correspondence
    Corresponding author.
    Affiliations
    Department of Medical-Surgical Sciences and Translational Medicine, Sant'Andrea Hospital, Sapienza University of Rome, Via di Grottarossa, Roma 1035 - 00189, Italy
    Search for articles by this author
Published:April 02, 2022DOI:https://doi.org/10.1016/j.dld.2022.03.007

      Abstract

      Background

      The endoscopic diagnosis of Helicobacter-pylori(H.pylori) infection and gastric precancerous lesions(GPL), namely atrophic-gastritis and intestinal-metaplasia, still remains challenging. Artificial intelligence(AI) may represent a powerful resource for the endoscopic recognition of these conditions.

      Aims

      To explore the diagnostic performance(DP) of AI in the diagnosis of GPL and H.pylori infection.

      Methods

      A systematic-review was performed by two independent authors up to September 2021. Inclusion criteria were studies focusing on the DP of AI-system in the diagnosis of GPL and H.pylori infection. The pooled accuracy of studies included was reported.

      Results

      Overall, 128 studies were found (PubMed-Embase-Cochrane Library) and four and nine studies were finally included regarding GPL and H.pylori infection, respectively. The pooled-accuracy(random effects model) was 90.3%(95%CI 84.3–94.9) and 79.6%(95%CI 66.7–90.0) with a significant heterogeneity[I2=90.4%(95%CI 78.5–95.7);I2=97.9%(97.2–98.6)] for GPL and H.pylori infection, respectively. The Begg's-test showed a significant publication-bias(p = 0.0371) only among studies regarding H.pylori infection. The pooled-accuracy(random-effects-model) was similar considering only studies using CNN-model for the diagnosis of H.pylori infection: 74.1%[(95%CI 51.6–91.3);I2=98.9%(95%CI 98.5–99.3)], Begg's-test(p = 0.1416) did not show publication-bias.

      Conclusion

      AI-system seems to be a good resource for an easier diagnosis of GPL and H.pylori infection, showing a pooled-diagnostic-accuracy of 90% and 80%, respectively. However, considering the high heterogeneity, these promising data need an external validation by randomized control trials and prospective real-time studies.

      Keywords

      1. Introduction

      Gastric cancer(GC) remains a major health concern since it is the sixth most common malignancy and the third cause of cancer-related death worldwide [
      • Bray F.
      • Ferlay J.
      • Soerjomataram I.
      • Siegel R.L.
      • Torre L.A.
      • Jemal A.
      Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries [published correction appears in CA Cancer J Clin. 2020 Jul;70(4):313].
      ]. As known, GC could be considered as the final step of the Correa's cascade [
      • Correa P.
      Human gastric carcinogenesis: a multistep and multifactorial process–first American cancer society award lecture on cancer epidemiology and prevention.
      ]. More specifically, the step-evolution begins from chronic Helicobacter pylori(H.pylori) infection, sequentially followed by atrophy, intestinal metaplasia(IM) (considered the “point-of-no-return”) and, thereafter, neoplastic conditions, such as dysplastic lesions and GC, may arise. For this reason, atrophic gastritis(AG) and IM are considered gastric precancerous lesions(GPL). Atrophy of the gastric mucosa is defined as the decrease or the disappearance of the original gastric glands, which may be replaced by pseudopyloric metaplasia [
      • Dilaghi E.
      • Baldaro F.
      • Pilozzi E.
      • et al.
      Pseudopyloric metaplasia is not associated with the development of gastric cancer.
      ] or IM. H.pylori is the main etiologic factor for chronic gastritis worldwide [

      IARC Helicobacter pylori Working Group, 2015. Helicobacter pylori eradication as a strategy for gastric cancer pre- vention. Lyon, France: International Agency for Research on Cancer (IARC Working Group Reports, No. 8). Avail- able at: http://www.iarc.fr/en/publications/pdfs-online/wrk/wrk8/index.php. Accessed on November 21, 2015.

      ], and longstanding infection may lead to progressive destruction of patches of gastric glands throughout the stomach described as multifocal AG.
      Sometimes, the diagnostic yield based on endoscopic features would be challenging, therefore, several technologies that use different light sources have been introduced in clinical practice of some referral centres offering alternative observation modes from white light imaging(WLI), such as electronic chromoendoscopy(EC), useful for IM detection and to highlight dysplastic-lesions, namely Narrow Band Imaging(NBI), Blue Light Imaging(BLI), and Linked Colour Imaging(LCI) [
      • Pimentel-Nunes P.
      • Libânio D.
      • Lage J.
      • et al.
      A multicenter prospective study of the real-time use of narrow-band imaging in the diagnosis of premalignant gastric conditions and lesions.
      ,
      • Rodríguez-Carrasco M.
      • Esposito G.
      • Libânio D.
      • Pimentel-Nunes P.
      • Dinis-Ribeiro M.
      Image-enhanced endoscopy for gastric preneoplastic conditions and neoplastic lesions: a systematic review and meta-analysis.
      ].
      Artificial intelligence(AI), with its efficient computational power and learning capacity, has caught considerable attention in the field of GPL and H.pylori infection. This powerful resource made possible using machines or computers based on human cognitive functions(learning and problem solving) related to human-thinking [
      • Russel S.
      • Norvig P.
      Artificial intelligence: a modern approach.
      ]. AI is originally based on machine-learning-algorithms, including random forest and support-vector-machines(SVM), that in the last years, were applied in various domains, especially in medicine. More recently, deep-learning (DL) had become the first approach adopted in much ongoing work, using multiple layers to gradually extract higher-level features from the originally input [
      • Niu P.H.
      • Zhao L.L.
      • Wu H.L.
      • Zhao D.B.
      • Chen Y.T.
      Artificial intelligence in gastric cancer: application and future perspectives.
      ]. The recent development of efficient hardware and computational power led to several AI models that could be applied in endoscopy, radiology, and pathology to improve the diagnosis, treatment, and prognosis of many gastrointestinal-conditions [
      • Hashimoto R.
      • Requa J.
      • Dao T.
      • et al.
      Artificial intelligence using convolutional neural networks for real-time detection of early esophageal neoplasia in Barrett's esophagus (with video).
      ,
      • Tang D.
      • Wang L.
      • Ling T.
      • et al.
      Development and validation of a real-time artificial intelligence-assisted system for detecting early gastric cancer: a multicentre retrospective diagnostic study.
      ,
      • Takenaka K.
      • Ohtsuka K.
      • Fujii T.
      • et al.
      Development and validation of a deep neural network for accurate evaluation of endoscopic images from patients with ulcerative colitis.
      ,
      • Byrne M.F.
      • Chapados N.
      • Soudan F.
      • et al.
      Real-time differentiation of adenomatous and hyperplastic diminutive colorectal polyps during analysis of unaltered videos of standard colonoscopy using a deep learning model.
      ,
      • Pecere S.
      • Milluzzo S.M.
      • Esposito G.
      • Dilaghi E.
      • Telese A.
      • Eusebi L.H.
      Applications of artificial intelligence for the diagnosis of gastrointestinal diseases.
      ].
      This systematic review and meta-analysis aims to report the diagnostic performance(DP) of AI application in the diagnosis of GPL and H.pylori infection.

      2. Materials and methods

      This systematic review and meta-analysis was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses(PRISMA) guidelines [
      • Moher D.
      • Liberati A.
      • Tetzlaff J.
      • Altman D.G.
      PRISMA Group
      Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement.
      ].
      To evaluate the DP of AI in the diagnosis of GPL (namely AG and IM) and H.pylori infection, a research in three electronic databases was performed in September 2021: PubMed, Embase, and the Cochrane Library. Study selection was conducted using the following search strings ((convolutional neural network)OR(artificial intelligence)OR(computer-based diagnosis)OR(computer-aided diagnosis)OR(support vector machine)) AND ((atrophic gastritis)OR(gastric atrophy)OR(gastric precancerous conditions)OR(gastric precancerous lesions)), ((convolutional neural network)OR(artificial intelligence)OR(computer-based diagnosis)OR(computer-aided diagnosis)OR(support vector machine)) AND ((intestinal metaplasia)OR(gastric precancerous conditions)OR(gastric precancerous lesions)), ((convolutional neural network)OR(artificial intelligence)OR(computer-based diagnosis)OR(computer-aided diagnosis)OR(support vector machine)) AND ((Helicobacter pylori)OR(Hp)).
      Furthermore, manual research of literature was conducted analysing the references of retrieved reviews and/or original articles already published on the topic under consideration.

      2.1 Study selection

      During the study selection process, original articles whose aim included accuracy/sensitivity/specificity of AI-system/s in the diagnosis of GPL and/or H.pylori infection were included. Letters, case reports, comments, reviews, congress abstracts and studies published in languages other than English, were excluded and duplicates were removed.
      Study selection was performed independently by two authors(ED and GE). First titles and subsequently abstracts were analysed and irrelevant studies were excluded. Therefore, the full-text papers were included or excluded following the inclusion and exclusion criteria, respectively. Disagreements were resolved through discussion with a third author(EL).

      2.2 Statistical analyses

      The diagnostic performance of AI application in the diagnosis of GPL and H.pylori infection was expressed as accuracy, sensitivity, specificity, positive (PPV), negative predictive values (NPV), positive likelihood ratio (LR+) and negative likelihood ratio (LR-). If these data were not present in the original study, they were calculated, where possible, by the authors.
      Of the included studies, a meta-analysis with Forest-plot was performed to evaluate the pooled diagnostic accuracy with 95% confidence intervals(CI) of AI-systems in the diagnosis of GPL and H.pylori infection. All studies were included in the meta-analyses irrespective of the applied AI system, and, in the case of diagnosis of H.pylori infection, a further meta-analysis with Forest-plot was performed considering only studies using convolutional-neural-network(CNN). Heterogeneity among studies was expressed as I2(inconsistency). In the case of heterogeneity among studies, the pooled diagnostic accuracy was expressed as a random effect model. Further, Egger's test and Funnel-plots were performed to assess potential publication-bias.

      3. Results

      The literature search resulted in 48 studies specifically focusing on the application of AI in endoscopic scenarios regarding GPL; 18 studies were removed as duplicates; after manually screening these titles and consequently the relative abstracts, 25 studies were included for full-text reading. Finally, 4 [
      • Zhang Y.
      • Li F.
      • Yuan F.
      • et al.
      Diagnosing chronic atrophic gastritis by gastroscopy using artificial intelligence.
      ,
      • Guimarães P.
      • Keller A.
      • Fehlmann T.
      • Lammert F.
      • Casper M.
      Deep-learning based detection of gastric precancerous conditions.
      ,
      • Yan T.
      • Wong P.K.
      • Choi I.C.
      • Vong C.M.
      • Yu H.H.
      Intelligent diagnosis of gastric intestinal metaplasia based on convolutional neural network and limited number of endoscopic images.
      ,
      • Xu M.
      • Zhou W.
      • Wu L.
      • et al.
      Artificial intelligence in the diagnosis of gastric precancerous conditions by image-enhanced endoscopy: a multicenter, diagnostic study (with video).
      ] studies met the inclusion criteria(Table 1).
      Table 1Baseline characteristics of the included studies on the diagnosis of gastric precancerous lesions.
      First Author, year of publicationCountryPatientsImages/ videosAI systemEndoscope systemCorrectly identifiedAccuracy (%)Sensitivity (%)Specificity (%)PPV
      Positive Predictive Value.
      (%)
      NPV
      Negative Predictive Value.
      (%)
      LR + 
      Positive Likelihood Ratio.
      LR -
      Negative Likelihood Ratio.
      Zhang Y, 2020China16995589 images (only antrum)CNN
      Convolutional Neural Network.
      WLI i-Scan mode
      White Light Imaging.
      160094.294.6 (95%CI 93.7–95.4)94.0 (95%CI 93.0–95.0)94.8 (95%CI 94.0–95.5)93.8 (95%CI 92.8–94.6)15.8 (95%CI 13.5–18.4)0.1 (95%CI 0.1–0.2)
      Guimar(a)es P, 2019Germany3570 images (corpus or fundus)DL
      Deep Learning.
      WLI6592.910087.585.71008.00.0
      Yan T, 2020China80477 imagesCNNNBI and magnifying-NBI
      Narrow Band Imaging.
      41186.292.6 (95%CI 88.3–95.4)79.6 (95%CI 73.7–84.4)82.4 (95%CI 78.3–85.8)91.2 (95%CI 86.9–94.2)4.5 (95%CI 3.5–5.9)0.1 (95%CI 0.1–0.2)
      Xu M, 2021China7798DCNN-ENDOANGEL
      Deep Convolutional Neural Network.
      Image-enhanced endoscopy8687.896.7 (95%CI 88.7–99.6)73.0 (95%CI 55.9–86.2)85.5 (95%CI 75.0–92.893.1 (95%CI 77.2–99.2)3.60.0
      First Author, year of publicationCountryPatientsImages/videosAI systemEndoscope systemCorrectly identifiedAccuracy (%)Sensitivity (%)Specificity (%)PPV
      Positive Predictive Value.
      (%)
      NPV
      Negative Predictive Value.
      (%)
      PLR
      Positive Likelihood Ratio.
      NLR
      Negative Likelihood Ratio.
      Zhang Y, 2020China16991641 images (only antrum)CNN
      Convolutional Neural Network.
      WLI i-Scan mode
      White Light Imaging.
      154694.294.694.015.70.1
      Guimar(a)es P, 2019Germany3570 images (corpus or fundus)DL
      Deep Learning.
      WLI6592.910087.585.71008.00.0
      Yan T, 2020China80477 imagesCNNNBI and magnifying-NBI
      Narrow Band Imaging.
      41186.2 (95%CI 82.7–89.1)92.6 (95%CI 88.3–95.4)79.6 (95%CI 73.7–84.4)82.4 (95%CI 78.3–85.8)91.2 (95%CI 86.9–94.2)4.5 (95%CI 3.5–5.9)0.1 (95%CI 0.1–0.2)
      Xu M, 2021China7798DCNN-ENDOANGEL
      Deep Convolutional Neural Network.
      Image-enhanced endoscopy8687.8 (95%CI 79.6–93.5)96.7 (95%CI 88.7–99.6)73.0 (95%CI 55.9–86.2)85.5 (95%CI 75.0–92.893.1 (95%CI 77.2–99.2)3.60.0
      a Convolutional Neural Network.
      b Deep Convolutional Neural Network.
      c Deep Learning.
      d White Light Imaging.
      e Narrow Band Imaging.
      f Positive Predictive Value.
      g Negative Predictive Value.
      h Positive Likelihood Ratio.
      i Negative Likelihood Ratio.
      A total of 80 studies were initially selected regarding the application of AI in the evaluation of H.pylori infection; however, 16 studies were excluded as duplicates, and 12 studies were excluded after the title review. In the next stage(abstract revision), 41 studies were not considered pertinent to the aim of the systematic review. Finally, 9 [
      • Zheng W.
      • Zhang X.
      • Kim J.J.
      • et al.
      High accuracy of convolutional neural network for evaluation of helicobacter pylori infection based on endoscopic images: preliminary experience.
      ,
      • Nakashima H.
      • Kawahira H.
      • Kawachi H.
      • Sakaki N.
      Artificial intelligence diagnosis of Helicobacter pylori infection using blue laser imaging-bright and linked color imaging: a single-center prospective study.
      ,
      • Shichijo S.
      • Endo Y.
      • Aoyama K.
      • et al.
      Application of convolutional neural networks for evaluating Helicobacter pylori infection status on the basis of endoscopic images.
      ,
      • Shichijo S.
      • Nomura S.
      • Aoyama K.
      • et al.
      Application of convolutional neural networks in the diagnosis of helicobacter pylori infection based on endoscopic images.
      ,
      • Itoh T.
      • Kawahira H.
      • Nakashima H.
      • Yata N.
      Deep learning analyzes Helicobacter pylori infection by upper gastrointestinal endoscopy images.
      ,
      • Huang C.R.
      • Sheu B.S.
      • Chung P.C.
      • Yang H.B.
      Computerized diagnosis of Helicobacter pylori infection and associated gastric inflammation from endoscopic images by refined feature selection using a neural network.
      ,
      • Huang C.R.
      • Chung P.C.
      • Sheu B.S.
      • Kuo H.J.
      • Popper M.
      Helicobacter pylori-related gastric histology classification using support-vector-machine-based feature selection.
      ,
      • Yasuda T.
      • Hiroyasu T.
      • Hiwa S.
      • et al.
      Potential of automatic diagnosis system with linked color imaging for diagnosis of Helicobacter pylori infection.
      ,
      • Nakashima H.
      • Kawahira H.
      • Kawachi H.
      • Sakaki N.
      Endoscopic three-categorical diagnosis of Helicobacter pylori infection using linked color imaging and deep learning: a single-center prospective study (with video).
      ] studies were selected and included in the metanalysis(Table 2). The PRISMA flowcharts was reported(Fig. 1A-B). Quality assessment exploring the methodological qualities was provided(Fig. 2A-B): all studies assessing AI-system in the diagnosis of GPL had a high-quality score; regarding studies on H.pylori, one study [
      • Nakashima H.
      • Kawahira H.
      • Kawachi H.
      • Sakaki N.
      Artificial intelligence diagnosis of Helicobacter pylori infection using blue laser imaging-bright and linked color imaging: a single-center prospective study.
      ] showed a low risk of bias, and two studies [
      • Itoh T.
      • Kawahira H.
      • Nakashima H.
      • Yata N.
      Deep learning analyzes Helicobacter pylori infection by upper gastrointestinal endoscopy images.
      ,
      • Nakashima H.
      • Kawahira H.
      • Kawachi H.
      • Sakaki N.
      Endoscopic three-categorical diagnosis of Helicobacter pylori infection using linked color imaging and deep learning: a single-center prospective study (with video).
      ] showed a high risk of bias.
      Table 2Baseline characteristics of the included studies on the diagnosis of Helicobacter pylori infection.
      First Author, year of publicationCountryPatientsImagesAI systemEndoscope systemCorrectly identifiedAccuracy (%)Sensitivity (%)Specificity (%)PPV
      Positive Predictive Value.
      (%)
      NPV
      Negative Predictive Value.
      (%)
      LR+
      Positive Likelihood Ratio.
      LR-
      Negative Likelihood Ratio.
      Huang CR, 2008Taiwan236Not availableSFFS
      Sequential Forward Floating Selection.
      based on SVM
      WLI
      White Light Imaging.
      21390.398.5 (95%CI 95.6–99–5)80.2 (95%CI 74.4–85.0)85.9 (95%CI 80.7–90.0)97.7 (95%CI 94.6–99.1)5.0 (95%CI 3.4–7.3)0.0 (95%CI 0.0–0.1)
      Zheng W, 2019China4523755CNN
      Convolutional Neural Network.
      WLI352293.891.6 (95%CI 90.7–92.5)98.6 (95%CI 98.1–98.9)99.3 (95%CI 98.9–99.5)84.3 (95%CI 83.1–85.5)63.6 (95%CI 39.7–102.0)0.1 (95%CI 0.1–0.1)
      Nakashima H, 2018Japan60648CNNWLI, BLI
      Blue Light Imaging.
      , and LCI
      Linked Color Imaging.
      3863.366.7 (95%CI 53.2–78.0)60.0 (95%CI 46.5–72.2)62.5 (95%CI 49.0–74.4)64.3 (95%CI 50.0–75.9)1.7 (95%CI 1.0–2.8)0.6 (95%CI 0.3–1.0)
      Shichijo S, 2017Japan39711,481CNNWLI34887.788.9 (95%CI 85.3–91.7)87.4 (95%CI 83.6–90.4)61.0 (95%CI 55.9–65.7)97.3 (95%CI 95.0–98.6)7.1 (95%CI 5.2–9.5)0.1 (95%CI 0.1–0.2)
      Shichijo S, 2019Japan84723,699CNNWLI77491.462.9 (95%CI 59.5–66.1)94.0 (95%CI 92.1–95.4)48.4 (95%CI 44.9–51.8)96.6 (95% CI 95.0–97.6)10.4 (95%CI 7.5–14.5)0.4 (95%CI 0.3–0.5)
      Itoh T, 2018Japan13930CNNWLI2686.786.7 (95%CI 68.4–95.6)86.7 (95%CI 68.4–95.6)86.7 (95%CI 68.4–95.6)86.7 (95%CI 68.4–95.6)6.5 (95%CI 1.8–24.0)0.2 (95%CI 0.0–0.6)
      Huang CR, 2004Taiwan74222RFSNN
      Refined Feature Selection with Neural Network.
      WLI6587.885.4 (95%CI 74.8–92.2)90.9 (95%CI 91.4–96.0)92.1 (95%CI 82.8–96.8)83.3 (95%CI 72.5–90.6)9.4 (95%CI 3.2–27.8)0.2 (95%CI 0.1–0.3)
      Nakashima H, 2020Japan120120 WLI imaging and 120 LCI imagingCAD system
      Computer-Aided Diagnosis.
      WLI and LCI77.560.0 (95%CI 44.3.75.1)86.2 (95%CI 76.7–92.9)68.8 (95%CI 44.3–75.1)4.30.5
      Yasuda T, 2019Japan105525SVM
      Support Vector Machine.
      LCI9287.690.5 (95%CI 82.8–95.1)85.7 (95%CI 77.2–91.5)80.9 (95%CI 71.8–87.6)93.1 (95%CI 86.0–96.9)6.3 (95%CI 3.4–11.7)0.1 (95%CI 0.0–0.3)
      a Convolutional Neural Network.
      b Computer-Aided Diagnosis.
      c Refined Feature Selection with Neural Network.
      d Sequential Forward Floating Selection.
      e Support Vector Machine.
      f White Light Imaging.
      g Blue Light Imaging.
      h Linked Color Imaging.
      i Positive Predictive Value.
      j Negative Predictive Value.
      k Positive Likelihood Ratio.
      l Negative Likelihood Ratio.
      Fig. 1
      Fig. 1(A-B): The PRISMA flowcharts regarding the selection of studies focused on the diagnosis of gastric precancerous lesions (A) and H. pylori infection(B).
      Fig. 2
      Fig. 2(A-B): Quality Assessment of AI-system for Diagnostic Performance in the diagnosis of gastric precancerous conditions (A) and H. pylori infection (B).

      3.1 AI and gastric precancerous lesions

      Four studies were selected: two studies aimed to recognize AG [
      • Shichijo S.
      • Nomura S.
      • Aoyama K.
      • et al.
      Application of convolutional neural networks in the diagnosis of helicobacter pylori infection based on endoscopic images.
      ,
      • Itoh T.
      • Kawahira H.
      • Nakashima H.
      • Yata N.
      Deep learning analyzes Helicobacter pylori infection by upper gastrointestinal endoscopy images.
      ], one study focused on the diagnosis of gastric IM [
      • Huang C.R.
      • Sheu B.S.
      • Chung P.C.
      • Yang H.B.
      Computerized diagnosis of Helicobacter pylori infection and associated gastric inflammation from endoscopic images by refined feature selection using a neural network.
      ], whilst the last aimed to detect both, AG and IM [
      • Huang C.R.
      • Chung P.C.
      • Sheu B.S.
      • Kuo H.J.
      • Popper M.
      Helicobacter pylori-related gastric histology classification using support-vector-machine-based feature selection.
      ].
      A study of 2020 by Zhang et al. [
      • Zhang Y.
      • Li F.
      • Yuan F.
      • et al.
      Diagnosing chronic atrophic gastritis by gastroscopy using artificial intelligence.
      ] aimed to construct a CNN to improve the diagnostic-rate of AG. A total of 5470 images of the antrum were collected, of which 3042 presented AG. The diagnostic accuracy of the CNN-AG model for AG was 94.2%, the sensitivity was 94.6%, and specificity was 94.0%, a LR+ of 15.8 and a LR- of 0.1.
      In 2020, Guimaraes et al. [
      • Guimarães P.
      • Keller A.
      • Fehlmann T.
      • Lammert F.
      • Casper M.
      Deep-learning based detection of gastric precancerous conditions.
      ] developed a DL-system for the diagnosis of corpus AG with non-standardized real-world WLI. A first data set with 100 AG images of 37 patients and 100 non-AG images of 64 patients, and a second data-set, used for independent testing, of 30 atrophic images and 40 non-atrophic images, were built. The DL-based algorithm showed, in the first data-set, an accuracy, sensitivity, specificity, and area under the curve(AUC) of 93.5%, 93.0%, 94.0%, and 0.98, respectively. The second data-set was evaluated by 3 expert endoscopists and 3 non-expert endoscopists. No significant difference between the expert and the non-expert group was found. The accuracy, sensitivity, specificity, PPV, NPV, LR+, LR- and AUC were 92.9%, 100%, 87.5%, 85.7%, 100%, 8.0, 0.0 and 0.98, respectively.
      Concerning gastric IM, Yan et al. in 2020 [
      • Yan T.
      • Wong P.K.
      • Choi I.C.
      • Vong C.M.
      • Yu H.H.
      Intelligent diagnosis of gastric intestinal metaplasia based on convolutional neural network and limited number of endoscopic images.
      ] aimed to develop an intelligent diagnostic (ID)-system using a CNN to assist in gastric IM diagnosis. Images were retrospectively collected and 158 patients with a diagnosis of gastric IM and 178 patients without IM, were included. Regarding patients with IM, 622 images with magnified NBI(M-NBI) and 426 images with NBI were collected. Regarding patients without IM, 462 M-NBI-images and 370 NBI-images from 178 were collected. A per-image and a per-patient analysis were performed. Concerning the per-images analysis, NBI-images and M-NBI-images were analysed. Concerning the NBI-images, the system showed a sensitivity, specificity and accuracy of 91.2%, 71.1%, and 82.7%, respectively. Concerning the M-NBI-images, the ID-system showed a sensitivity, specificity and accuracy of 93.8%, 84.2%, and 88.6%, respectively. The AUC for NBI-images and M-NBI-images were 0.907 and 0.941, respectively. At a per-patient analysis, the ID-system was compared to four expert endoscopists. A total of 80 patients were enrolled and one image for patient was used. The expert endoscopists showed a sensitivity of 86.5%, a specificity of 81.4% and an accuracy of 83.8%. The ID-system showed a sensitivity of 91.9%, a specificity of 86.0% and an accuracy of 88.8%.
      Recently, Xu M et al. in 2021 [
      • Xu M.
      • Zhou W.
      • Wu L.
      • et al.
      Artificial intelligence in the diagnosis of gastric precancerous conditions by image-enhanced endoscopy: a multicenter, diagnostic study (with video).
      ] developed a deep convolutional neural network(DCNN) system, named ENDOANGEL, for detecting in real-time endoscopic images(EI) of GPL acquired through image-enhanced endoscopy(IEE) technology, such as NBI and BLI. EI were retrospectively collected for the development, validation, and internal and external test of the system; whilst, prospective consecutive patients receiving IEE were enrolled from a single center, to access in real-time the applicability of the proposed AI-technology. For the prospective video test set, 98 video clips from 77 patients undergoing IEE were enrolled. ENDOANGEL had an accuracy, sensitivity, specificity, PPV, NPV, LR+, LR-, of 87.8%, 96.7%, and 73.0%, 85.5%, 93.1%, 3.6, 0.0, respectively for the AG-model. For detecting IM, the AI-system showed an accuracy of 89.8%, a sensitivity of 94.6%, and a specificity of 83.7%. Compared with non-expert endoscopists, for identifying AG, ENDOANGEL had a better performance regarding the accuracy, specificity, and positive predictive value(PPV), whereas the DCNN system was comparable with the experts regarding the accuracy, specificity, and PPV. For identifying IM, the DCNN system, compared with non-experts, had better performance for the accuracy, PPV, and negative predictive value(NPV), whereas ENDOANGEL was comparable with the experts for the accuracy, PPV, and NPV.

      3.1.1 Meta-analysis

      In Fig. 3A-B the Forest-plot and the Funnel-plot of the DP of AI in the diagnosis of GPL are given. Considering the significant level of heterogeneity among the included studies [I290.4%, p<0.0001 (95%CI 78.5–95.7)], the total random effect was considered, and a pooled accuracy of AI-systems on precancerous-conditions of 90.3%(95%CI 84.3–94.9) was obtained. Egger's test, showing Intercept of −2.9 (95%CI −15.9–10.0; p = 0.43), and Begg's test (p = 1.00) did not show publication-bias.
      Fig. 3
      Fig. 3(A-B): The Forest plot (A) and the Funnel plot (B) of the diagnostic performance of AI in the diagnosis of gastric precancerous lesions.

      3.1.2 AI and Helicobacter pylori infection

      Regarding the application of AI in endoscopic-scenarios for H.pylori diagnosis, 9 studies[19–27] and 2 meta-analysis [
      • Mohan B.P.
      • Khan S.R.
      • Kassab L.L.
      • et al.
      Convolutional neural networks in the computer-aided diagnosis of Helicobacter pylori infection and non-causal comparison to physician endoscopists: a systematic review with meta-analysis.
      ,
      • Bang C.S.
      • Lee J.J.
      • Baik G.H.
      Artificial intelligence for the prediction of Helicobacter pylori infection in endoscopic images: systematic review and meta-analysis of diagnostic test accuracy.
      ] were selected. Eight out of the nine studies included in this systematic review were already included in the results of the two previously published meta-analyses, while one additional study was retrieved in the current systematic review.
      Zheng et al. [
      • Zheng W.
      • Zhang X.
      • Kim J.J.
      • et al.
      High accuracy of convolutional neural network for evaluation of helicobacter pylori infection based on endoscopic images: preliminary experience.
      ] in 2019, aimed to examine the accuracy of CNN using EI for evaluating H.pylori infection. A total of 1959 patients were retrospectively included and of those, 59% had H.pylori infection documented on gastric biopsies and/or H.pylori breath-test. The derivation-cohort, for machine-learning, consisted of 1507 patients(11,729 gastric images), and the validation-cohort, for evaluating the accuracy of the AI-system, included a total of 452 patients(3755 gastric images). The DP of CNN for single gastric images showed an AUC, sensitivity, specificity, and accuracy of 0.93, 81.4%, 90.1%, 84.5%, respectively. Applying the CNN-model for multiple gastric-images per patient, the AUC, sensitivity, specificity, PPV, NPV, and accuracy were: 0.97, 91.6%, 98.6%, 99.3%, 84.3%, and 93.8%, respectively. The CNN-model using multiple gastric images had a higher AUC compared with a single gastric image(p<0.001).
      Nakashima et al. [
      • Nakashima H.
      • Kawahira H.
      • Kawachi H.
      • Sakaki N.
      Artificial intelligence diagnosis of Helicobacter pylori infection using blue laser imaging-bright and linked color imaging: a single-center prospective study.
      ], in a prospective study, aimed to establish an AI-model for the prediction of H.pylori infection using EI. Sixty patients were included in the test group to evaluate the diagnostic accuracy of the AI. The authors focused on EI of the lesser curvature of the gastric corpus, and endoscopists captured 3 still images at the same position in all subjects using WLI, BLI-bright, and LCI. The H.pylori IgG antibody titer was considered as the gold-standard for infection status. The AUC for WLI was 0.66. In contrast, the AUC for BLI-bright and LCI was 0.96 and 0.95, respectively. The AUC obtained for BLI-bright and LCI were significantly larger than those for WLI(p<0.01).
      In the retrospective study, by Shichijo et al. [
      • Shichijo S.
      • Nomura S.
      • Aoyama K.
      • et al.
      Application of convolutional neural networks in the diagnosis of helicobacter pylori infection based on endoscopic images.
      ], a total of 1750 patients were reviewed and a total of 32,208 images were considered for the development data-set. To evaluate the diagnostic accuracy of the constructed CNN, a separate test data-set was prepared, 587 patients for a total of 11,481 images were considered. The authors, firstly, constructed the CNN using all images together. Second, another CNN-model was proposed using the images classified according to 8 different locations (cardia, upper-corpus, middle-corpus, lesser-curvature, angulus, lower-corpus, antrum, pylorus). The DP of the CNN constructed using unclassified images showed an AUC of 0.89; sensitivity, specificity, and accuracy were 81.9%, 83.4%, and 83.1%, respectively. Whereas, the CNN-model prepared using classified images showed an AUC of 0.93, and the DP showed a sensitivity, specificity, and accuracy of 88.9%, 87.4%, and 87.7%, respectively.
      In a later study, Shichijo et al. [
      • Shichijo S.
      • Endo Y.
      • Aoyama K.
      • et al.
      Application of convolutional neural networks for evaluating Helicobacter pylori infection status on the basis of endoscopic images.
      ] in 2019, aimed to evaluate the performance of a CNN-model in the diagnosis of the H.pylori infection(not only H.pylori-positive, or -negative, but also H.pylori eradicated patients). A test data-set was prepared to evaluate the diagnostic accuracy of the proposed CNN, and a total of 847 patients(23,699 images) were included. The CNN diagnosed 418 of 23,699 images as positive(1.8%), 23,034(97.2%) as negative, and 247(1.0%) as eradicated. Authors reported that these disproportionate findings may have been attributable to a large proportion of negative images in the development data-set, or difficulty in picking up characteristic findings of positive or eradicated status.
      Itoh et al. [
      • Itoh T.
      • Kawahira H.
      • Nakashima H.
      • Yata N.
      Deep learning analyzes Helicobacter pylori infection by upper gastrointestinal endoscopy images.
      ] in 2018, conducted a prospective study, evaluating a CNN-model optimized to diagnose H.pylori infection. A total of 139 patients were included and from those, 179 WLI EI were obtained of the lesser curvature of the stomach. For CNN learning, 149 images were used, and the remaining 30 were set aside to be used as test images. The CNN-model showed a sensitivity and a specificity of 86.7%, and the AUC was 0.956.
      Huang et al. [
      • Huang C.R.
      • Sheu B.S.
      • Chung P.C.
      • Yang H.B.
      Computerized diagnosis of Helicobacter pylori infection and associated gastric inflammation from endoscopic images by refined feature selection using a neural network.
      ] in 2004, proposed an AI-system by constructing a refined feature selection with neural network(RFSNN), based on EI from patients with or without H.pylori infection. One-hundred-four patients were prospectively included for diagnostic upper-endoscopy for dyspeptic symptoms. H.pylori infection was histologically diagnosed. Seventy-four patients were included for predicting the presence of H.pylori infection. RFSNN showed a sensitivity, specificity, and accuracy of 85.4%, 90.9%, and >80%, respectively. In 2008, Huang et al. [
      • Huang C.R.
      • Chung P.C.
      • Sheu B.S.
      • Kuo H.J.
      • Popper M.
      Helicobacter pylori-related gastric histology classification using support-vector-machine-based feature selection.
      ] proposed a computer-aided diagnosis system using sequential forward floating selection(SFFS) based on SVM, to diagnose gastric histology of H.pylori from EI. A total of 130 H.pylori-infected patients were included, and the SFFS showed an accuracy >90%, regarding DP in acute and chronic inflammation, related to H.pylori infection.
      Yasuda et al. [
      • Yasuda T.
      • Hiroyasu T.
      • Hiwa S.
      • et al.
      Potential of automatic diagnosis system with linked color imaging for diagnosis of Helicobacter pylori infection.
      ] in 2019, constructed a machine-learning-based algorithm for an automatic diagnosis system for H.pylori infection using LCI-images. A SVM system was used. In the validation test, a total of 105 consecutive patients were retrospectively included. The proposed AI-model showed an accuracy, sensitivity, and specificity of 87.6%, 90.5%, and 85.7%, respectively.
      In 2020 by Nakashima et al. [
      • Nakashima H.
      • Kawahira H.
      • Kawachi H.
      • Sakaki N.
      Endoscopic three-categorical diagnosis of Helicobacter pylori infection using linked color imaging and deep learning: a single-center prospective study (with video).
      ]. In this study, the authors aimed to develop a computer-aided-diagnosis(CAD) system to distinguish H.pylori-infected by non-infected subjects. The endoscopic data were used to develop two different CAD-systems, one for LCI(LCI-CAD) and one for WLI(WLI-CAD) images. The diagnosis accuracy of the LCI-CAD system was 84.2% for “uninfected”, 82.5% for infected, and 79.2% for “post-eradication” subjects.

      3.1.3 Meta-analysis

      In Fig. 4A-B the Forest-plot and Funnel-plot of the DP of AI in the diagnosis of H.pylori infection are given. Heterogeneity among studies considered reached a significant level, I2 was 97.9%[95%CI 97.2–98.5(p<0.0001)]. Consequently, a totally random effect was considered, and the pooled accuracy, taking in count all the 9 studies on the DP of AI in the diagnosis of H.pylori infection, was 79.6(95%CI 66.7–90.0). The Egger's test showed Intercept of −8.6(95%CI −20.8–3.5; p = 0.14). The Begg's test showed a significant publication-bias(p = 0.0371).
      Fig. 4
      Fig. 4(A-B): The Forest plot (A) and Funnel plot (B) of the diagnostic performance of AI in the diagnosis of H. pylori infection.
      We conducted a second meta-analysis(Supplementary Fig. 1A-B), considering only the studies that proposed a CNN-model [
      • Zheng W.
      • Zhang X.
      • Kim J.J.
      • et al.
      High accuracy of convolutional neural network for evaluation of helicobacter pylori infection based on endoscopic images: preliminary experience.
      ,
      • Nakashima H.
      • Kawahira H.
      • Kawachi H.
      • Sakaki N.
      Artificial intelligence diagnosis of Helicobacter pylori infection using blue laser imaging-bright and linked color imaging: a single-center prospective study.
      ,
      • Shichijo S.
      • Endo Y.
      • Aoyama K.
      • et al.
      Application of convolutional neural networks for evaluating Helicobacter pylori infection status on the basis of endoscopic images.
      ,
      • Shichijo S.
      • Nomura S.
      • Aoyama K.
      • et al.
      Application of convolutional neural networks in the diagnosis of helicobacter pylori infection based on endoscopic images.
      ,
      • Itoh T.
      • Kawahira H.
      • Nakashima H.
      • Yata N.
      Deep learning analyzes Helicobacter pylori infection by upper gastrointestinal endoscopy images.
      ]. The pooled accuracy did not substantially change: 74.10%(95%CI 51.59–91.35, total random effect,I2 98.9%(95%CI 98.5–99.3,p<0.0001). The Egger's test showed Intercept of −16.5(95%CI −44.9–11.8; p = 0.16) and Begg's test(p = 0.14) did not show significant publication-bias.

      4. Discussion

      To our best knowledge, this is the first meta-analysis specifically dedicated to GPL. Heterogeneity among studies was significant, anyway, the pooled accuracy of 90.3%, may let us consider AI a powerful tool in the diagnosis of AG or gastric IM, as precancerous-conditions.
      A lower pooled accuracy was reported, regarding studies focused on H.pylori infection, in particular accuracy, considering all studies [
      • Zheng W.
      • Zhang X.
      • Kim J.J.
      • et al.
      High accuracy of convolutional neural network for evaluation of helicobacter pylori infection based on endoscopic images: preliminary experience.
      ,
      • Nakashima H.
      • Kawahira H.
      • Kawachi H.
      • Sakaki N.
      Artificial intelligence diagnosis of Helicobacter pylori infection using blue laser imaging-bright and linked color imaging: a single-center prospective study.
      ,
      • Shichijo S.
      • Endo Y.
      • Aoyama K.
      • et al.
      Application of convolutional neural networks for evaluating Helicobacter pylori infection status on the basis of endoscopic images.
      ,
      • Shichijo S.
      • Nomura S.
      • Aoyama K.
      • et al.
      Application of convolutional neural networks in the diagnosis of helicobacter pylori infection based on endoscopic images.
      ,
      • Itoh T.
      • Kawahira H.
      • Nakashima H.
      • Yata N.
      Deep learning analyzes Helicobacter pylori infection by upper gastrointestinal endoscopy images.
      ,
      • Huang C.R.
      • Sheu B.S.
      • Chung P.C.
      • Yang H.B.
      Computerized diagnosis of Helicobacter pylori infection and associated gastric inflammation from endoscopic images by refined feature selection using a neural network.
      ,
      • Huang C.R.
      • Chung P.C.
      • Sheu B.S.
      • Kuo H.J.
      • Popper M.
      Helicobacter pylori-related gastric histology classification using support-vector-machine-based feature selection.
      ,
      • Yasuda T.
      • Hiroyasu T.
      • Hiwa S.
      • et al.
      Potential of automatic diagnosis system with linked color imaging for diagnosis of Helicobacter pylori infection.
      ,
      • Nakashima H.
      • Kawahira H.
      • Kawachi H.
      • Sakaki N.
      Endoscopic three-categorical diagnosis of Helicobacter pylori infection using linked color imaging and deep learning: a single-center prospective study (with video).
      ] and those in which a CNN-model was proposed [
      • Zheng W.
      • Zhang X.
      • Kim J.J.
      • et al.
      High accuracy of convolutional neural network for evaluation of helicobacter pylori infection based on endoscopic images: preliminary experience.
      ,
      • Nakashima H.
      • Kawahira H.
      • Kawachi H.
      • Sakaki N.
      Artificial intelligence diagnosis of Helicobacter pylori infection using blue laser imaging-bright and linked color imaging: a single-center prospective study.
      ,
      • Shichijo S.
      • Endo Y.
      • Aoyama K.
      • et al.
      Application of convolutional neural networks for evaluating Helicobacter pylori infection status on the basis of endoscopic images.
      ,
      • Shichijo S.
      • Nomura S.
      • Aoyama K.
      • et al.
      Application of convolutional neural networks in the diagnosis of helicobacter pylori infection based on endoscopic images.
      ,
      • Itoh T.
      • Kawahira H.
      • Nakashima H.
      • Yata N.
      Deep learning analyzes Helicobacter pylori infection by upper gastrointestinal endoscopy images.
      ], was 79.64 and 74.1, respectively. Despite these pooled results, we confidentially may consider the AI-system a worthwhile resource for endoscopists. Indeed, our data are supported by the previous systematic review and meta-analysis, as mentioned before. Mohan et al. [
      • Mohan B.P.
      • Khan S.R.
      • Kassab L.L.
      • et al.
      Convolutional neural networks in the computer-aided diagnosis of Helicobacter pylori infection and non-causal comparison to physician endoscopists: a systematic review with meta-analysis.
      ] in 2020, reported a pooled accuracy of 87.1% for detecting H.pylori infection. Similarly, Bang et al. [
      • Bang C.S.
      • Lee J.J.
      • Baik G.H.
      Artificial intelligence for the prediction of Helicobacter pylori infection in endoscopic images: systematic review and meta-analysis of diagnostic test accuracy.
      ] in the same year, showed how AI-systems reached a real good DP for the prediction of H.pylori infection: the pooled sensitivity and specificity were 87% and 86%, respectively.
      GPL and H.pylori infection are two well-established risk factors for GC [
      • Correa P.
      Human gastric carcinogenesis: a multistep and multifactorial process–first American cancer society award lecture on cancer epidemiology and prevention.
      ,
      • Kuipers E.J.
      • Uyterlinde A.M.
      • Peña A.S.
      • et al.
      Long-term sequelae of Helicobacter pylori gastritis.
      ,
      • Annibale B.
      • Negrini R.
      • Caruana P.
      • et al.
      Two-thirds of atrophic body gastritis patients have evidence of Helicobacter pylori infection.
      ,
      • Sugano K.
      Effect of Helicobacter pylori eradication on the incidence of gastric cancer: a systematic review and meta-analysis.
      ,
      • Uemura N.
      • Okamoto S.
      • Yamamoto S.
      • et al.
      Helicobacter pylori infection and the development of gastric cancer.
      ]. Quantification of GC risk in patients with GPL is unclear, and literature studies reported conflicting results. The cumulative 5-year incidence of GC has been shown to vary from 0.7% to 10% in patients with AG and from 5.3% to 9.8% in those with IM [
      • Shichijo S.
      • Hirata Y.
      • Niikura R.
      • et al.
      Histologic intestinal metaplasia and endoscopic atrophy are predictors of gastric cancer development after Helicobacter pylori eradication.
      ]. Reported progression rates to GC varied between 0% to 2% per year for AG; for IM, progression rates to GC varied widely from 0% to 10% [
      • de Vries A.C.
      • Haringsma J.
      • Kuipers E.J.
      The detection, surveillance and treatment of premalignant gastric lesions related to Helicobacter pylori infection.
      ]. For this reason, European guidelines [
      • Pimentel-Nunes P.
      • et al.
      Management of epithelial precancerous conditions and lesions in the stomach (MAPS II): european society of gastrointestinal en- doscopy (ESGE), European helicobacter and microbiota study group (EHMSG), European society of pathology (ESP), and sociedade Portuguesa de endoscopia digestiva (SPED) guideline update 2019.
      ] suggest a 3-years endoscopic follow-up for patients with GPL [
      • Esposito G.
      • Dilaghi E.
      • Cazzato M.
      • Pilozzi E.
      • Conti L.
      • Carabotti M.
      • et al.
      Endoscopic surveillance at 3 years after diagnosis, according to European guidelines, seems safe in patients with atrophic gastritis in a low-risk region.
      ], or shorter if risk factors are present. H. pylori infection is widely known as a risk factor for GC [
      • Uemura N.
      • Okamoto S.
      • Yamamoto S.
      • et al.
      Helicobacter pylori infection and the development of gastric cancer.
      ,
      • Shichijo S.
      • Hirata Y.
      • Niikura R.
      • et al.
      Histologic intestinal metaplasia and endoscopic atrophy are predictors of gastric cancer development after Helicobacter pylori eradication.
      ]. Indeed, approximately 50% of the world population is infected by H.pylori, and it is estimated that 89% of non-cardia gastric cancers, which represents 78% of the GC diagnosis, are attributed to H.pylori infection [

      IARC Helicobacter pylori Working Group, 2015. Helicobacter pylori eradication as a strategy for gastric cancer prevention. Lyon, France: International Agency for Research on Cancer (IARC Working Group Reports, No. 8). Avail- able at: http://www.iarc.fr/en/publications/pdfs-online/wrk/wrk8/index.php. Accessed on November 21,2015.

      ,
      Schistosomes, liver flukes and Helicobacter pylori
      IARC working group on the evaluation of carcinogenic risks to humans. Lyon, 7-14 June 1994.
      ,
      • de Martel C.
      • Ferlay J.
      • Franceschi S.
      • et al.
      Global burden of cancers attributable to infections in 2008: a review and synthetic analysis.
      ].
      The current systematic review showed that AI could be a substantial resource in the detection of GPL and H.pylori infection. Studies on GPL showed a good DP of each AI model considered, with a pooled accuracy of 90.3% and a sensibility in the diagnosis of AG and IM of about 90%. Moreover, 3 studies reported a superior DP of AI when compared to endoscopists, especially when compared to less experienced endoscopists. Similarly, AI applied for H.pylori detection during gastroscopy, revealed an excellent potentiality, showing a pooled diagnostic accuracy of 79.6% but studies showed a significant level of heterogeneity(I2=97.9,p<0.0001).
      Several limitations would attempt the robustness of the studies considered. The performance of AI can only be valid for the population under evaluation and depends on the prevalence of target conditions for the selected population(known as spectrum bias or class imbalance). External(possibly, prospective) validation would be considered mandatory. Moreover, the majority of the studies selected were conducted at a single center, which limits the possibility to generalize the results obtained. Another issue to take into consideration, regarding studies focused on H.pylori infection, is that there were only a few data regarding post-eradication images, thus increasing the difficulty of the analysis of performance in the discrimination of uninfected and post-eradicated images of H.pylori infection. All studies but one, included in the systematic review, were conducted in Asia. Mostly, it was not considered an external validation confirming the diagnostic validity of the results reached, making it difficult to generalize the results, which for this reason should be received with caution. Heterogeneity was present taking into account the methodology adopted from studies considered: variability was present in terms of the number of patients included, not all studies considered the same reference test, in particular concerning H.pylori infection, patients could undergo histology assessment to test H.pylori status, or serum/urine H.pylori IgG antibodies, fecal-antigen-test, or urease-breath-test, as well. Images used to create the training or the validation data-set were obtained from different regions of the stomach, and the endoscopic technology applied was represented from a different source of light: some studies considered only WLI, other only EC, such as NBI, or BLI, or LCI images, other studies considered a combination of these technologies. Meta-analytic data confirmed heterogeneity in both groups of studies exploring DP in GPL and H.pylori infection and in studies specifically focused on H.pylori infection, publication-bias was present according to Egger's and Bang's tests, as reported above.
      Despite these lacks, AI might be considered a potentially reliable tool for physicians in the next future; moreover, it could be considered a potential resource for trainee endoscopists, after appropriate practical training.
      In conclusion, AI-system seems to be a good resource to easier diagnose GPL and H.pylori infection showing a pooled-diagnostic-accuracy of 90% and 80%,respectively. This resource seems to have great potential in the field of GPL and H.pylori infection, being able to help patients and physicians to implement the possibility to diagnose such conditions, making eventually possible and optimizing the effort in the prevention of GC. Moreover, the future availability and the eventual widespread diffusion of AI in the reference centres and beyond could make possible to reach uniformity in the diagnosis of such conditions. AI could reduce the procedure time and the inter-variability between different operators. The spread of AI in the Endoscopic Units could be helpful to overcome the diagnosis gap between referral centres and primary care hospitals, making homogeneous the patient's health-care, in respect of early diagnosis, prevention and therapeutic possibilities. Future prospective and multicentre studies are needed, also considering an external validation and a real-time application. AI is expected to allow a greater agreement in the evaluation of H.pylori infection, and to significantly reduce the miss rate of AG and IM diagnosis, and ultimately to improve the care of patients.

      Funding

      This study was funded in part by Sapienza University Rome, Funds “Ateneo2019–2020″.

      Declaration of Competing Interest

      None declared.

      Appendix. Supplementary materials

      References

        • Bray F.
        • Ferlay J.
        • Soerjomataram I.
        • Siegel R.L.
        • Torre L.A.
        • Jemal A.
        Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries [published correction appears in CA Cancer J Clin. 2020 Jul;70(4):313].
        CA Cancer J Clin. 2018; 68: 394-424https://doi.org/10.3322/caac.21492
        • Correa P.
        Human gastric carcinogenesis: a multistep and multifactorial process–first American cancer society award lecture on cancer epidemiology and prevention.
        Cancer Res. 1992; 52: 6735-6740
        • Dilaghi E.
        • Baldaro F.
        • Pilozzi E.
        • et al.
        Pseudopyloric metaplasia is not associated with the development of gastric cancer.
        Am J Gastroenterol. 2021; 116: 1859-1867https://doi.org/10.14309/ajg.0000000000001390
      1. IARC Helicobacter pylori Working Group, 2015. Helicobacter pylori eradication as a strategy for gastric cancer pre- vention. Lyon, France: International Agency for Research on Cancer (IARC Working Group Reports, No. 8). Avail- able at: http://www.iarc.fr/en/publications/pdfs-online/wrk/wrk8/index.php. Accessed on November 21, 2015.

        • Pimentel-Nunes P.
        • Libânio D.
        • Lage J.
        • et al.
        A multicenter prospective study of the real-time use of narrow-band imaging in the diagnosis of premalignant gastric conditions and lesions.
        Endoscopy. 2016; 48: 723-730https://doi.org/10.1055/s-0042-108435
        • Rodríguez-Carrasco M.
        • Esposito G.
        • Libânio D.
        • Pimentel-Nunes P.
        • Dinis-Ribeiro M.
        Image-enhanced endoscopy for gastric preneoplastic conditions and neoplastic lesions: a systematic review and meta-analysis.
        Endoscopy. 2020; 52: 1048-1065https://doi.org/10.1055/a-1205-0570
        • Russel S.
        • Norvig P.
        Artificial intelligence: a modern approach.
        2th ed. Pearson Education, 2003
        • Niu P.H.
        • Zhao L.L.
        • Wu H.L.
        • Zhao D.B.
        • Chen Y.T.
        Artificial intelligence in gastric cancer: application and future perspectives.
        World J Gastroenterol. 2020; 26: 5408-5419https://doi.org/10.3748/wjg.v26.i36.5408
        • Hashimoto R.
        • Requa J.
        • Dao T.
        • et al.
        Artificial intelligence using convolutional neural networks for real-time detection of early esophageal neoplasia in Barrett's esophagus (with video).
        Gastrointest Endosc. 2020; 91: 1264-1271.e1https://doi.org/10.1016/j.gie.2019.12.049
        • Tang D.
        • Wang L.
        • Ling T.
        • et al.
        Development and validation of a real-time artificial intelligence-assisted system for detecting early gastric cancer: a multicentre retrospective diagnostic study.
        EBioMedicine. 2020; 62103146https://doi.org/10.1016/j.ebiom.2020.103146
        • Takenaka K.
        • Ohtsuka K.
        • Fujii T.
        • et al.
        Development and validation of a deep neural network for accurate evaluation of endoscopic images from patients with ulcerative colitis.
        Gastroenterology. 2020; 158: 2150-2157https://doi.org/10.1053/j.gastro.2020.02.012
        • Byrne M.F.
        • Chapados N.
        • Soudan F.
        • et al.
        Real-time differentiation of adenomatous and hyperplastic diminutive colorectal polyps during analysis of unaltered videos of standard colonoscopy using a deep learning model.
        Gut. 2019; 68: 94-100https://doi.org/10.1136/gutjnl-2017-314547
        • Pecere S.
        • Milluzzo S.M.
        • Esposito G.
        • Dilaghi E.
        • Telese A.
        • Eusebi L.H.
        Applications of artificial intelligence for the diagnosis of gastrointestinal diseases.
        Diagnostics (Basel). 2021 Aug 30; 11: 1575
        • Moher D.
        • Liberati A.
        • Tetzlaff J.
        • Altman D.G.
        • PRISMA Group
        Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement.
        PLoS Med. 2009; 6e1000097https://doi.org/10.1371/journal.pmed.1000097
        • Zhang Y.
        • Li F.
        • Yuan F.
        • et al.
        Diagnosing chronic atrophic gastritis by gastroscopy using artificial intelligence.
        Dig Liver Dis. 2020; 52: 566-572https://doi.org/10.1016/j.dld.2019.12.146
        • Guimarães P.
        • Keller A.
        • Fehlmann T.
        • Lammert F.
        • Casper M.
        Deep-learning based detection of gastric precancerous conditions.
        Gut. 2020; 69: 4-6https://doi.org/10.1136/gutjnl-2019-319347
        • Yan T.
        • Wong P.K.
        • Choi I.C.
        • Vong C.M.
        • Yu H.H.
        Intelligent diagnosis of gastric intestinal metaplasia based on convolutional neural network and limited number of endoscopic images.
        Comput Biol Med. 2020; 126104026https://doi.org/10.1016/j.compbiomed.2020.104026
        • Xu M.
        • Zhou W.
        • Wu L.
        • et al.
        Artificial intelligence in the diagnosis of gastric precancerous conditions by image-enhanced endoscopy: a multicenter, diagnostic study (with video).
        Gastrointest Endosc. 2021; 94: 540-548.e4https://doi.org/10.1016/j.gie.2021.03.013
        • Zheng W.
        • Zhang X.
        • Kim J.J.
        • et al.
        High accuracy of convolutional neural network for evaluation of helicobacter pylori infection based on endoscopic images: preliminary experience.
        Clin Transl Gastroenterol. 2019; 10: e00109https://doi.org/10.14309/ctg.0000000000000109
        • Nakashima H.
        • Kawahira H.
        • Kawachi H.
        • Sakaki N.
        Artificial intelligence diagnosis of Helicobacter pylori infection using blue laser imaging-bright and linked color imaging: a single-center prospective study.
        Ann Gastroenterol. 2018; 31: 462-468https://doi.org/10.20524/aog.2018.0269
        • Shichijo S.
        • Endo Y.
        • Aoyama K.
        • et al.
        Application of convolutional neural networks for evaluating Helicobacter pylori infection status on the basis of endoscopic images.
        Scand J Gastroenterol. 2019; 54: 158-163https://doi.org/10.1080/00365521.2019.1577486
        • Shichijo S.
        • Nomura S.
        • Aoyama K.
        • et al.
        Application of convolutional neural networks in the diagnosis of helicobacter pylori infection based on endoscopic images.
        E Bio Med. 2017; 25: 106-111https://doi.org/10.1016/j.ebiom.2017.10.014
        • Itoh T.
        • Kawahira H.
        • Nakashima H.
        • Yata N.
        Deep learning analyzes Helicobacter pylori infection by upper gastrointestinal endoscopy images.
        Endosc Int Open. 2018; 6: E139-E144https://doi.org/10.1055/s-0043-120830
        • Huang C.R.
        • Sheu B.S.
        • Chung P.C.
        • Yang H.B.
        Computerized diagnosis of Helicobacter pylori infection and associated gastric inflammation from endoscopic images by refined feature selection using a neural network.
        Endoscopy. 2004; 36: 601-608https://doi.org/10.1055/s-2004-814519
        • Huang C.R.
        • Chung P.C.
        • Sheu B.S.
        • Kuo H.J.
        • Popper M.
        Helicobacter pylori-related gastric histology classification using support-vector-machine-based feature selection.
        IEEE Trans Inf Technol Biomed. 2008; 12: 523-531https://doi.org/10.1109/TITB.2007.913128
        • Yasuda T.
        • Hiroyasu T.
        • Hiwa S.
        • et al.
        Potential of automatic diagnosis system with linked color imaging for diagnosis of Helicobacter pylori infection.
        Dig Endosc. 2020; 32: 373-381https://doi.org/10.1111/den.13509
        • Nakashima H.
        • Kawahira H.
        • Kawachi H.
        • Sakaki N.
        Endoscopic three-categorical diagnosis of Helicobacter pylori infection using linked color imaging and deep learning: a single-center prospective study (with video).
        Gastric Cancer. 2020; 23: 1033-1040https://doi.org/10.1007/s10120-020-01077-1
        • Mohan B.P.
        • Khan S.R.
        • Kassab L.L.
        • et al.
        Convolutional neural networks in the computer-aided diagnosis of Helicobacter pylori infection and non-causal comparison to physician endoscopists: a systematic review with meta-analysis.
        Ann Gastroenterol. 2021; 34: 20-25https://doi.org/10.20524/aog.2020.0542
        • Bang C.S.
        • Lee J.J.
        • Baik G.H.
        Artificial intelligence for the prediction of Helicobacter pylori infection in endoscopic images: systematic review and meta-analysis of diagnostic test accuracy.
        J Med Internet Res. 2020; 22 (Published 2020 Sep 16): e21983https://doi.org/10.2196/21983
        • Kuipers E.J.
        • Uyterlinde A.M.
        • Peña A.S.
        • et al.
        Long-term sequelae of Helicobacter pylori gastritis.
        Lancet. 1995; 345: 1525-1528https://doi.org/10.1016/s0140-6736(95)91084-0
        • Annibale B.
        • Negrini R.
        • Caruana P.
        • et al.
        Two-thirds of atrophic body gastritis patients have evidence of Helicobacter pylori infection.
        Helicobacter. 2001; 6: 225-233https://doi.org/10.1046/j.1083-4389.2001.00032.x
        • Sugano K.
        Effect of Helicobacter pylori eradication on the incidence of gastric cancer: a systematic review and meta-analysis.
        Gastric Cancer. 2019; 22: 435-445https://doi.org/10.1007/s10120-018-0876-0
        • Uemura N.
        • Okamoto S.
        • Yamamoto S.
        • et al.
        Helicobacter pylori infection and the development of gastric cancer.
        N Engl J Med. 2001; 345: 784-789https://doi.org/10.1056/NEJMoa001999
        • Shichijo S.
        • Hirata Y.
        • Niikura R.
        • et al.
        Histologic intestinal metaplasia and endoscopic atrophy are predictors of gastric cancer development after Helicobacter pylori eradication.
        Gastrointest Endosc. 2016; 84: 618-624https://doi.org/10.1016/j.gie.2016.03.791
        • de Vries A.C.
        • Haringsma J.
        • Kuipers E.J.
        The detection, surveillance and treatment of premalignant gastric lesions related to Helicobacter pylori infection.
        Helicobacter. 2007; 12: 1-15https://doi.org/10.1111/j.1523-5378.2007.00475.x
        • Pimentel-Nunes P.
        • et al.
        Management of epithelial precancerous conditions and lesions in the stomach (MAPS II): european society of gastrointestinal en- doscopy (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
        • Esposito G.
        • Dilaghi E.
        • Cazzato M.
        • Pilozzi E.
        • Conti L.
        • Carabotti M.
        • et al.
        Endoscopic surveillance at 3 years after diagnosis, according to European guidelines, seems safe in patients with atrophic gastritis in a low-risk region.
        Dig Liver Dis. Apr 2021; 53: 467-473
      2. IARC Helicobacter pylori Working Group, 2015. Helicobacter pylori eradication as a strategy for gastric cancer prevention. Lyon, France: International Agency for Research on Cancer (IARC Working Group Reports, No. 8). Avail- able at: http://www.iarc.fr/en/publications/pdfs-online/wrk/wrk8/index.php. Accessed on November 21,2015.

        • Schistosomes, liver flukes and Helicobacter pylori
        IARC working group on the evaluation of carcinogenic risks to humans. Lyon, 7-14 June 1994.
        IARC Monogr Eval Carcinog Risks Hum. 1994; 61: 1-241
        • de Martel C.
        • Ferlay J.
        • Franceschi S.
        • et al.
        Global burden of cancers attributable to infections in 2008: a review and synthetic analysis.
        Lancet Oncol. 2012; 13: 607-615https://doi.org/10.1016/S1470-2045(12)70137-7