Consensus statements on the current landscape of artificial intelligence applications in endoscopy, addressing roadblocks, and advancing artificial intelligence in gastroenterology.

BACKGROUND AND AIMS: The American Society for Gastrointestinal Endoscopy (ASGE) AI Task Force along with experts in endoscopy, technology space, regulatory authorities, and other medical subspecialties initiated a consensus process that analyzed the current literature, highlighted potential areas, and outlined the necessary research in artificial intelligence (AI) to allow a clearer understanding of AI as it pertains to endoscopy currently. METHODS: A modified Delphi process was used to develop these consensus statements. RESULTS: Statement 1: Current advances in AI allow for the development of AI-based algorithms that can be applied to endoscopy to augment endoscopist performance in detection and characterization of endoscopic lesions. Statement 2: Computer vision-based algorithms provide opportunities to redefine quality metrics in endoscopy using AI, which can be standardized and can reduce subjectivity in reporting quality metrics. Natural language processing-based algorithms can help with the data abstraction needed for reporting current quality metrics in GI endoscopy effortlessly. Statement 3: AI technologies can support smart endoscopy suites, which may help optimize workflows in the endoscopy suite, including automated documentation. Statement 4: Using AI and machine learning helps in predictive modeling, diagnosis, and prognostication. High-quality data with multidimensionality are needed for risk prediction, prognostication of specific clinical conditions, and their outcomes when using machine learning methods. Statement 5: Big data and cloud-based tools can help advance clinical research in gastroenterology. Multimodal data are key to understanding the maximal extent of the disease state and unlocking treatment options. Statement 6: Understanding how to evaluate AI algorithms in the gastroenterology literature and clinical trials is important for gastroenterologists, trainees, and researchers, and hence education efforts by GI societies are needed. Statement 7: Several challenges regarding integrating AI solutions into the clinical practice of endoscopy exist, including understanding the role of human-AI interaction. Transparency, interpretability, and explainability of AI algorithms play a key role in their clinical adoption in GI endoscopy. Developing appropriate AI governance, data procurement, and tools needed for the AI lifecycle are critical for the successful implementation of AI into clinical practice. Statement 8: For payment of AI in endoscopy, a thorough evaluation of the potential value proposition for AI systems may help guide purchasing decisions in endoscopy. Reliable cost-effectiveness studies to guide reimbursement are needed. Statement 9: Relevant clinical outcomes and performance metrics for AI in gastroenterology are currently not well defined. To improve the quality and interpretability of research in the field, steps need to be taken to define these evidence standards. Statement 10: A balanced view of AI technologies and active collaboration between the medical technology industry, computer scientists, gastroenterologists, and researchers are critical for the meaningful advancement of AI in gastroenterology. CONCLUSIONS: The consensus process led by the ASGE AI Task Force and experts from various disciplines has shed light on the potential of AI in endoscopy and gastroenterology. AI-based algorithms have shown promise in augmenting endoscopist performance, redefining quality metrics, optimizing workflows, and aiding in predictive modeling and diagnosis. However, challenges remain in evaluating AI algorithms, ensuring transparency and interpretability, addressing governance and data procurement, determining payment models, defining relevant clinical outcomes, and fostering collaboration between stakeholders. Addressing these challenges while maintaining a balanced perspective is crucial for the meaningful advancement of AI in gastroenterology.

Rates of Recurrent Intestinal Metaplasia and Dysplasia After Successful Endoscopic Therapy of Barrett's Neoplasia by Endoscopic Mucosal Resection vs Endoscopic Submucosal Dissection and Ablation: A Large North American Multicenter Cohort.

INTRODUCTION: Endoscopic eradication therapy (EET) combining endoscopic resection (ER) with endoscopic mucosal resection (EMR) or endoscopic submucosal dissection (ESD) followed by ablation is the standard of care for the treatment of dysplastic Barrett's esophagus (BE). We have previously shown comparable rates of complete remission of intestinal metaplasia (CRIM) with both approaches. However, data comparing recurrence after CRIM are lacking. We compared rates of recurrence after CRIM with both techniques in a multicenter cohort. METHODS: Patients undergoing EET achieving CRIM at 3 academic institutions were included. Demographic and clinical data were abstracted. Outcomes included rates and predictors of any BE and dysplastic BE recurrence in the 2 groups. Cox-proportional hazards models and inverse probability treatment weighting (IPTW) analysis were used for analysis. RESULTS: A total of 621 patients (514 EMR and 107 ESD) achieving CRIM were included in the recurrence analysis. The incidence of any BE (15.7, 5.7 per 100 patient-years) and dysplastic BE recurrence (7.3, 5.3 per 100 patient-years) were comparable in the EMR and ESD groups, respectively. On multivariable analyses, the chances of BE recurrence were not influenced by ER technique (hazard ratio 0.87; 95% confidence interval 0.51-1.49; P = 0.62), which was also confirmed by IPTW analysis (ESD vs EMR: hazard ratio 0.98; 95% confidence interval 0.56-1.73; P = 0.94). BE length, lesion size, and history of cigarette smoking were independent predictors of BE recurrence. DISCUSSION: Patients with BE dysplasia/neoplasia achieving CRIM, initially treated with EMR/ablation, had comparable recurrence rates to ESD/ablation. Randomized trials are needed to confirm these outcomes between the 2 ER techniques.

Algorithm Training and Testing for a Nonendoscopic Barrett's Esophagus Detection Test in Prospective Multicenter Cohorts.

BACKGROUND & AIMS: Endoscopic Barrett's esophagus (BE) and esophageal adenocarcinoma (EAC) detection is invasive and expensive. Nonendoscopic BE/EAC detection tools are guideline-endorsed alternatives. We previously described a 5-methylated DNA marker (MDM) panel assayed on encapsulated sponge cell collection device (CCD) specimens. We aimed to train a new algorithm using a 3-MDM panel and test its performance in an independent cohort. METHODS: Algorithm training and test samples were from 2 prospective multicenter cohorts. All BE cases had esophageal intestinal metaplasia (with or without dysplasia/EAC); control subjects had no endoscopic evidence of BE. The CCD procedure was followed by endoscopy. From CCD cell lysates, DNA was extracted, bisulfite treated, and MDMs were blindly assayed. The algorithm was set and locked using cross-validated logistic regression (training set) and its performance was assessed in an independent test set. RESULTS: Training (N = 352) and test (N = 125) set clinical characteristics were comparable. The final panel included 3 MDMs (NDRG4, VAV3, ZNF682). Overall sensitivity was 82% (95% CI, 68%-94%) at 90% (79%-98%) specificity and 88% (78%-94%) sensitivity at 84% (70%-93%) specificity in training and test sets, respectively. Sensitivity was 90% and 68% for all long- and short-segment BE, respectively. Sensitivity for BE with high-grade dysplasia and EAC was 100% in training and test sets. Overall sensitivity for nondysplastic BE was 82%. Areas under the receiver operating characteristic curves for BE detection were 0.92 and 0.94 in the training and test sets, respectively. CONCLUSIONS: A locked 3-MDM panel algorithm for BE/EAC detection using a nonendoscopic CCD demonstrated excellent sensitivity for high-risk BE cases in independent validation samples. (Clinical trials.gov: NCT02560623, NCT03060642.).

Physician perceptions on the current and future impact of artificial intelligence to the field of gastroenterology.

BACKGROUND AND AIMS: The use of artificial intelligence (AI) has transformative implications to the practice of gastroenterology and endoscopy. The aims of this study were to understand the perceptions of the gastroenterology community toward AI and to identify potential barriers for adoption. METHODS: A 16-question online survey exploring perceptions on the current and future implications of AI to the field of gastroenterology was developed by the American Society for Gastrointestinal Endoscopy AI Task Force and distributed to national and international society members. Participant demographic information including age, sex, experience level, and practice setting was collected. Descriptive statistics were used to summarize survey findings, and a Pearson χ2 analysis was performed to determine the association between participant demographic information and perceptions of AI. RESULTS: Of 10,162 invited gastroenterologists, 374 completed the survey. The mean age of participants was 46 years (standard deviation, 12), and 299 participants (80.0%) were men. One hundred seventy-nine participants (47.9%) had >10 years of practice experience, with nearly half working in the community setting. Only 25 participants (6.7%) reported the current use of AI in their clinical practice. Most participants (95.5%) believed that AI solutions will have a positive impact in their practice. One hundred seventy-six participants (47.1%) believed that AI will make clinical duties more technical but will also ease the burden of the electronic medical record (54.0%). The top 3 areas where AI was predicted to be most influential were endoscopic lesion detection (65.3%), endoscopic lesion characterization (65.8%), and quality metrics (32.6%). Participants voiced a desire for education on topics such as the clinical use of AI applications (64.4%), the advantages and limitations of AI applications (57.0%), and the technical methodology of AI (44.7%). Most participants (42.8%) expressed that the cost of AI implementation should be covered by their hospital. Demographic characteristics significantly associated with this perception included participants' years in practice and practice setting. CONCLUSIONS: Gastroenterologists have an overall positive perception regarding the use of AI in clinical practice but voiced concerns regarding its technical aspects and coverage of costs associated with implementation. Further education on the clinical use of AI applications with understanding of the advantages and limitations appears to be valuable in promoting adoption.

Prevalence and Predictors of Barrett's Esophagus After Negative Initial Endoscopy: Analysis From Two National Databases.

BACKGROUND & AIMS: Guidelines suggest a single screening esophagogastroduodenoscopy (EGD) in patients with multiple risk factors for Barrett's esophagus (BE). We aimed to determine BE prevalence and predictors on repeat EGD after a negative initial EGD, using 2 large national databases (GI Quality Improvement Consortium [GIQuIC] and TriNetX). METHODS: Patients who underwent at least 2 EGDs were included and those with BE or esophageal adenocarcinoma detected at initial EGD were excluded. Patient demographics and prevalence of BE on repeat EGD were collected. Multivariate logistic regression was performed to assess for independent risk factors for BE detected on the repeat EGD. RESULTS: In 214,318 and 153,445 patients undergoing at least 2 EGDs over a median follow-up of 28-35 months, the prevalence of BE on repeat EGD was 1.7% in GIQuIC and 3.4% in TriNetX, respectively (26%-45% of baseline BE prevalence). Most (89%) patients had nondysplastic BE. The prevalence of BE remained stable over time (from 1 to >5 years from negative initial EGD) but increased with increasing number of risk factors. BE prevalence in a high-risk population (gastroesophageal reflux disease plus ≥1 risk factor for BE) was 3%-4%. CONCLUSIONS: In this study of >350,000 patients, rates of BE on repeat EGD ranged from 1.7%-3.4%, and were higher in those with multiple risk factors. Most were likely missed at initial evaluation, underscoring the importance of a high-quality initial endoscopic examination. Although routine repeat endoscopic BE screening after a negative initial examination is not recommended, repeat screening may be considered in carefully selected patients with gastroesophageal reflux disease and ≥2 risk factors for BE, potentially using nonendoscopic tools.

Clinical Outcomes After Endoscopic Management of Low-Risk and High-Risk T1a Esophageal Adenocarcinoma: A Multicenter Study.

INTRODUCTION: Endoscopic eradication therapy (EET) is standard of care for T1a esophageal adenocarcinoma (EAC). However, data on outcomes in high-risk T1a EAC are limited. We assessed and compared outcomes after EET of low-risk and high-risk T1a EAC, including intraluminal EAC recurrence, extraesophageal metastases, and overall survival. METHODS: Patients who underwent EET for T1a EAC at 3 referral Barrett's esophagus endotherapy units between 1996 and 2022 were included. Patients with submucosal invasion, positive deep margins, or metastases at initial diagnosis were excluded. High-risk T1a EAC was defined as T1a EAC with poor differentiation and/or lymphovascular invasion, with low-risk disease being defined without these features. All pathology was systematically assessed by expert gastrointestinal pathologists. Baseline and follow-up endoscopy and pathology data were abstracted. Time-to-event analyses were performed to compare outcomes between groups. RESULTS: One hundred eighty-eight patients with T1a EAC were included (high risk, n = 45; low risk, n = 143) with a median age of 70 years, and 84% were men. Groups were comparable for age, sex, Barrett's esophagus length, lesion size, and EET technique. Rates of delayed extraesophageal metastases (11.1% vs 1.4%) were significantly higher in the high-risk group ( P = 0.02). There was no significant difference in the rates of intraluminal EAC recurrence ( P = 0.79) and overall survival ( P = 0.73) between the 2 groups. DISCUSSION: Patients with high-risk T1a EAC undergoing successful EET had a substantially higher rate of extraesophageal metastases compared with those with low-risk T1a EAC on long-term follow-up. These data should be factored into discussions with patients while selecting treatment approaches. Additional prospective data in this area are critical.

Population Based Time Trends in the Epidemiology and Mortality of Gastroesophageal Junction and Esophageal Adenocarcinoma.

BACKGROUND: Limited data are available on the epidemiology of gastroesophageal junction adenocarcinoma (GEJAC), particularly in comparison to esophageal adenocarcinoma (EAC). With the advent of molecular non-endoscopic Barrett's esophagus (BE) detection tests which sample the esophagus and gastroesophageal junction, early detection of EAC and GEJAC has become a possibility and their epidemiology has gained importance. AIMS: We sought to evaluate time trends in the epidemiology and survival of patients with EAC and GEJAC in a population-based cohort. METHODS: EAC and GEJAC patients from 1976 to 2019 were identified using ICD 9 and 10 diagnostic codes from the Rochester Epidemiology Project (REP). Clinical data and survival status were abstracted. Poisson regression was used to calculate incidence rate ratios (IRR). Survival analysis and Cox proportional models were used to assess predictors of survival. RESULTS: We included 443 patients (287 EAC,156 GEJAC). The incidence of EAC and GEJAC during 1976-2019 was 1.40 (CI 1.1-1.74) and 0.83 (CI 0.61-1.11) per 100,000 people, respectively. There was an increase in the incidence of EAC (IRR = 2.45, p = 0.011) and GEJAC (IRR = 3.17, p = 0.08) from 2000 to 2004 compared to 1995-1999, plateauing in later time periods. Most patients had associated BE and presented at advanced stages, leading to high 5-year mortality rates (66% in EAC and 59% in GEJAC). Age and stage at diagnosis were predictors of mortality. CONCLUSION: The rising incidence of EAC/GEJAC appears to have plateaued somewhat in the last decade. However, both cancers present at advanced stages with persistently poor survival, underscoring the need for early detection.

A novel blood-sensing capsule for rapid detection of upper GI bleeding: a prospective clinical trial.

BACKGROUND AND AIMS: Upper GI bleeding (UGIB) is a common medical emergency associated with high resource utilization, morbidity, and mortality. Timely EGD can be challenging from personnel, resource, and access perspectives. PillSense (EnteraSense Ltd, Galway, Ireland) is a novel swallowed bleeding sensor for the detection of UGIB, anticipated to aid in patient triage and guide clinical decision-making for individuals with suspected UGIB. METHODS: This prospective, open-label, single-arm comparative clinical trial of a novel bleeding sensor for patients with suspected UGIB was performed at a tertiary care center. The PillSense system consists of an optical sensor and an external receiver that processes and displays data from the capsule as "Blood Detected" or "No Blood Detected." Patients underwent EGD within 4 hours of capsule administration; participants were followed up for 21 days to confirm capsule passage. RESULTS: A total of 126 patients were accrued to the study (59.5% male; mean age, 62.4 ± 14.3 years). Sensitivity and specificity for detecting the presence of blood were 92.9% (P = .02) and 90.6% (P < .001), respectively. The capsule's positive and negative predictive values were 74.3% and 97.8%, and positive and negative likelihood ratios were 9.9 and .08. No adverse events or deaths occurred related to the PillSense system, and all capsules were excreted from patients on follow-up. CONCLUSIONS: The PillSense system is safe and effective for detecting the presence of blood in patients evaluated for UGIB before upper GI endoscopy. It is a rapidly deployed tool, with easy-to-interpret results that will affect the diagnosis and triage of patients with suspected UGIB. (Clinical trial registration number: NCT05385224.).

Development of Electronic Health Record-Based Machine Learning Models to Predict Barrett's Esophagus and Esophageal Adenocarcinoma Risk.

INTRODUCTION: Screening for Barrett's esophagus (BE) is suggested in those with risk factors, but remains underutilized. BE/esophageal adenocarcinoma (EAC) risk prediction tools integrating multiple risk factors have been described. However, accuracy remains modest (area under the receiver-operating curve [AUROC] ≤0.7), and clinical implementation has been challenging. We aimed to develop machine learning (ML) BE/EAC risk prediction models from an electronic health record (EHR) database. METHODS: The Clinical Data Analytics Platform, a deidentified EHR database of 6 million Mayo Clinic patients, was used to predict BE and EAC risk. BE and EAC cases and controls were identified using International Classification of Diseases codes and augmented curation (natural language processing) techniques applied to clinical, endoscopy, laboratory, and pathology notes. Cases were propensity score matched to 5 independent randomly selected control groups. An ensemble transformer-based ML model architecture was used to develop predictive models. RESULTS: We identified 8,476 BE cases, 1,539 EAC cases, and 252,276 controls. The BE ML transformer model had an overall sensitivity, specificity, and AUROC of 76%, 76%, and 0.84, respectively. The EAC ML transformer model had an overall sensitivity, specificity, and AUROC of 84%, 70%, and 0.84, respectively. Predictors of BE and EAC included conventional risk factors and additional novel factors, such as coronary artery disease, serum triglycerides, and electrolytes. DISCUSSION: ML models developed on an EHR database can predict incident BE and EAC risk with improved accuracy compared with conventional risk factor-based risk scores. Such a model may enable effective implementation of a minimally invasive screening technology.

Long-term outcomes following successful endoscopic treatment of T1 esophageal adenocarcinoma: a multicenter cohort study.

BACKGROUND AND AIMS: Endoscopic eradication therapy (EET) is guideline endorsed for management of early-stage (T1) esophageal adenocarcinoma (EAC). Patients with baseline high-grade dysplasia (HGD) and EAC are at highest risk of recurrence after successful EET, but limited data exist on long-term (>5 year) recurrence outcomes. Our aim was to assess the incidence and predictors of long-term recurrence in a multicenter cohort of patients with T1 EAC treated with EET. METHODS: Patients with T1 EAC achieving successful endoscopic cancer eradication with a minimum of 5 years' clinical follow-up were included. The primary outcome was neoplastic recurrence, defined as dysplasia or EAC, and it was characterized as early (<2 years), intermediate (2-5 years), or late (>5 years). Predictors of recurrence were assessed by time to event analysis. RESULTS: A total of 84 T1 EAC patients (75 T1a, 9 T1b) with a median 9.1 years (range, 5.1-18.3 years) of follow-up were included. The overall incidence of neoplastic recurrence was 2.0 per 100 person-years of follow-up. Seven recurrences (3 dysplasia, 4 EAC) occurred after 5 years of EAC remission. Overall, 88% of recurrences were treated successfully endoscopically. EAC recurrence-related mortality occurred in 3 patients at a median of 5.2 years from EAC remission. Complete eradication of intestinal metaplasia was independently associated with reduced recurrence (hazard ratio, .13). CONCLUSIONS: Following successful EET of T1 EAC, neoplastic recurrence occurred after 5 years in 8.3% of cases. Careful long-term surveillance should be continued in this patient population. Complete eradication of intestinal metaplasia should be the therapeutic end point for EET.

Framework and metrics for the clinical use and implementation of artificial intelligence algorithms into endoscopy practice: recommendations from the American Society for Gastrointestinal Endoscopy Artificial Intelligence Task Force.

In the past few years, we have seen a surge in the development of relevant artificial intelligence (AI) algorithms addressing a variety of needs in GI endoscopy. To accept AI algorithms into clinical practice, their effectiveness, clinical value, and reliability need to be rigorously assessed. In this article, we provide a guiding framework for all stakeholders in the endoscopy AI ecosystem regarding the standards, metrics, and evaluation methods for emerging and existing AI applications to aid in their clinical adoption and implementation. We also provide guidance and best practices for evaluation of AI technologies as they mature in the endoscopy space. Note, this is a living document; periodic updates will be published as progress is made and applications evolve in the field of AI in endoscopy.

Artificial intelligence-enhanced volumetric laser endomicroscopy improves dysplasia detection in Barrett's esophagus in a randomized cross-over study.

Volumetric laser endomicroscopy (VLE) is an advanced endoscopic imaging tool that can improve dysplasia detection in Barrett's esophagus (BE). However, VLE scans generate 1200 cross-sectional images that can make interpretation difficult. The impact of a new VLE artificial intelligence algorithm called Intelligent Real-time Image Segmentation (IRIS) is not well-characterized. This is a randomized prospective cross-over study of BE patients undergoing endoscopy who were randomized to IRIS-enhanced or unenhanced VLE first followed by the other (IRIS-VLE vs. VLE-IRIS, respectively) at expert BE centers. The primary outcome was image interpretation time, which served as a surrogate measure for ease of interpretation. The secondary outcome was diagnostic yield of dysplasia for each imaging modality. 133 patients were enrolled. 67 patients were randomized to VLE-IRIS and 66 to IRIS-VLE. Total interpretation time did not differ significantly between groups (7.8 min VLE-IRIS vs. 7 min IRIS-VLE, P = 0.1), however unenhanced VLE interpretation time was significantly shorter in the IRIS-VLE group (2.4 min vs. 3.8 min, P < 0.01). When IRIS was used first, 100% of dysplastic areas were identified, compared with 76.9% when VLE was the first interpretation modality (P = 0.06). IRIS-enhanced VLE reduced the time of subsequent unenhanced VLE interpretation, suggesting heightened efficiency and improved dysplasia detection. It was also able to identify all endoscopically non-visible dysplastic areas.

Acute Esophageal Necrosis: A Retrospective Cohort Study Highlighting the Mayo Clinic Experience.

OBJECTIVE: To describe the clinical, endoscopic, and histologic features in patients with acute esophageal necrosis (AEN). PATIENTS AND METHODS: In this retrospective cohort study, patients who were diagnosed as having AEN at Mayo Clinic sites in Minnesota, Florida, and Arizona between January 1, 1996, and January 31, 2021, were included. Data were collected on patient clinical characteristics and endoscopic and pathologic findings. RESULTS: The study included 79 patients with AEN with a median (range) age of 64 years (12 to 91 years); 53 (67.1%) were men. Predominant presenting symptoms were hematemesis (49 of 79 [62.0%]), abdominal pain (29 [36.7%]), and melena (20 [25.3%]). Shock was the triggering event for AEN in 49 (62.0%). The 30- and 90-day mortality were 24.0% (19 of 79) and 31.6% (25), respectively. The presence of coexisting infection or bacteremia was significantly associated with 90-day mortality (P<.01). Endoscopically, involvement of the distal third only, distal two-thirds only, and entire esophagus was observed in 31.6% (24 of 76), 39.5% (30), and 29.0% (22), respectively. The length of esophageal involvement correlated with duration of hospitalization (P=.05). The endoscopic appearance of the esophageal mucosa ranged from predominantly white (21 of 44 [47.7%]) to mixed white and black (13 [29.6%]) to predominantly black (10 [22.7%]), and sloughing was present in 18 (40.9%). In the 26 patients with histopathologic findings available for review, 25 (96.1%) had necrosis and/or ulceration with abundant pigmentation. Among the 79 patients, 39 (49.4%) had a follow-up esophagogastroduodenoscopy; 26 of these 39 patients (66.7%) had resolution while 5 had persistent AEN, 4 of whom had improvement. Esophageal strictures developed in 7 of the 39 patients (18.0%). CONCLUSION: Acute esophageal necrosis is a serious condition observed in critically ill patients. Its endoscopic appearance can be highly variable. In patients with an unclear diagnosis, esophageal biopsies may be helpful given the characteristic histologic findings.

Development of a deep learning model for the histologic diagnosis of dysplasia in Barrett's esophagus.

BACKGROUND AND AIMS: The risk of progression in Barrett's esophagus (BE) increases with development of dysplasia. There is a critical need to improve the diagnosis of BE dysplasia, given substantial interobserver disagreement among expert pathologists and overdiagnosis of dysplasia by community pathologists. We developed a deep learning model to predict dysplasia grade on whole-slide imaging. METHODS: We digitized nondysplastic BE (NDBE), low-grade dysplasia (LGD), and high-grade dysplasia (HGD) histology slides. Two expert pathologists confirmed all histology and digitally annotated areas of dysplasia. Training, validation, and test sets were created (by a random 70/20/10 split). We used an ensemble approach combining a "you only look once" model to identify regions of interest and histology class (NDBE, LGD, or HGD) followed by a ResNet101 model pretrained on ImageNet applied to the regions of interest. Diagnostic performance was determined for the whole slide. RESULTS: We included slides from 542 patients (164 NDBE, 226 LGD, and 152 HGD) yielding 8596 bounding boxes in the training set, 1946 bounding boxes in the validation set, and 840 boxes in the test set. When the ensemble model was used, sensitivity and specificity for LGD was 81.3% and 100%, respectively, and >90% for NDBE and HGD. The overall positive predictive value and sensitivity metric (calculated as F1 score) was .91 for NDBE, .90 for LGD, and 1.0 for HGD. CONCLUSIONS: We successfully trained and validated a deep learning model to accurately identify dysplasia on whole-slide images. This model can potentially help improve the histologic diagnosis of BE dysplasia and the appropriate application of endoscopic therapy.