Integration of AI in Barrett’s Esophagus Clinical Practice: A New Way Forward

Abstract

Barrett’s esophagus is the only known precursor for esophageal adenocarcinoma. The stepwise progression of nondysplastic Barrett’s to low-grade dysplasia, high-grade dysplasia, intramucosal carcinoma, and invasive cancer provides an opportunity for endoscopic surveillance, with the goal to detect and effectively treat early neoplastic changes endoscopically, in order to avoid surgery and chemoradiation. Multiple limitations in the current endoscopic surveillance practice have led to missed early neoplasia and therefore reduced the effectiveness of the current surveillance practice. Artificial intelligence (AI)-based clinical decision support systems have been developed to provide additional assistance to physicians performing diagnostic and therapeutic gastrointestinal endoscopy. In this article, we review the current endoscopic surveillance practice for Barrett’s esophagus, its limitations, the potential role of AI to improve these limitations, and our suggested framework to integrate AI into Barrett’s clinical practice.

Introduction

Barrett’s esophagus (BE), defined as the replacement of the squamous epithelium of the distal esophagus by intestinal metaplasia, is the only known precursor for esophagus adenocarcinoma (EAC).¹ Patients with BE have a 30–40-fold higher risk of developing EAC as compared with the general population.² The incidence of EAC has risen significantly in the past decade in Western countries. Although EAC and BE are known to be the disease of elderly White males, recent studies have reported an alarming 40% rise in the EAC rate among the 45–64 age group and 50% rise in the prevalence of BE (from 304 to 466 per 100,000 patients).³ The majority of patients with EAC are still diagnosed at advanced stages, with 5-year survival rates remaining below 20%.⁴ The stepwise progression of nondysplastic BE (NDBE) to low-grade dysplasia (LGD), high-grade dysplasia (HGD), intramucosal carcinoma, and invasive cancer provides an opportunity for screening and surveillance.⁵ Currently, all gastrointestinal societies recommend endoscopic surveillance once a diagnosis of BE has been established.^1,5–7 The goal of endoscopic surveillance is to detect early neoplastic changes that can be treated endoscopically, without the need for chemotherapy, radiation, or surgical resection. As such, endoscopic surveillance is an integral part of the EAC prevention paradigm. Despite this, multiple limitations in the current approach exist that result in missing early neoplasia and therefore reduced effectiveness of the current surveillance practice.

In recent years, artificial intelligence (AI) has emerged as a promising tool to improve the field of BE, some include detection of subtle dysplasia and cancer, quality control in upper endoscopy, and reducing the considerable interobserver variability in identifying and distinguishing dysplasia histologically. Here, we review the current endoscopic surveillance practice for BE and its limitations, the potential role of AI to improve these limitations, and our suggested framework to integrate AI into Barrett’s clinical practice.

Current Status of BE Surveillance and Limitations

High-quality examination of the BE mucosa is the foundation of endoscopic surveillance, which entails washing and cleaning of the surface mucosa, taking adequate time for inspection using both high-definition white-light imaging and chromoendoscopy, either dye-based (using methylene blue or acetic acid) or electronic (narrow band imaging [NBI], blue laser imaging, or i-Scan optical enhancement [i-Scan], Fujinon intelligence chromoendoscopy), identification of landmarks such as the squamocolumnar junction, gastroesophageal junction (GEJ), and the diaphragmatic hiatus, as well as defining the BE segment with the Prague C (circumferential) and M (maximal) criteria for standardized BE reporting.⁸ Although it has been shown that there is a direct correlation between the inspection time and dysplasia detection, due to the time-consuming and challenging nature of the BE exam, endoscopists may not adhere to these protocols, especially in busy practices.⁹

Once the BE mucosa has been carefully examined, sampling should be performed. Visible lesions should be categorized based on morphological appearance in accordance with the Paris classification, sampled separately (either by the biopsy forceps or endoscopic resection) and placed in a separate bottle.¹⁰ Due to the subtle nature of the dysplastic lesions, it can be difficult, sometimes even for Barrett’s experts, to identify these lesions.^11,12 At the same time, it is known that the majority of surveillance endoscopies are conducted by non-Barrett’s experts in community settings.¹³ It has been reported that the sensitivity and specificity of Barrett’s nonexperts are 79% and 75% for the detection of Barrett’s-related neoplasia, and up to 26% of neoplastic lesions are missed on initial endoscopic evaluation.^15,16

The targeted biopsy/resection of the visible lesions should be followed by the Seattle protocol, defined as nontargeted four-quadrant biopsies every 2 cm (in the absence of dysplasia) or every 1 cm (with a history of dysplasia). The Seattle protocol biopsies are required because only 13% of early neoplasia appears as macroscopically visible nodules on endoscopic examination.¹⁷ This protocol has been shown to increase the yield of dysplasia detection.¹⁸ However, adherence to this protocol is time-consuming, with data showing that only 51.2% of endoscopists are compliant with the Seattle protocol, with rates further decreasing as the length of the BE segment increases.¹⁹ In addition, even with full compliance with the Seattle protocol, due to the patchy distribution of dysplasia and EAC, these areas can be missed. As a result of the limitations, a significant proportion of dysplastic lesions are missed on the initial endoscopic evaluation. One systematic review and meta-analysis showed that up to 26.6% of lesions with HGD and EAC can be missed.¹⁵ Another study reported a 26% rate for post-endoscopy esophageal carcinoma.¹⁶

Histology is critical in guiding the time intervals at which surveillance endoscopy should be performed after the initial evaluation. For patients with NDBE, it is recommended to perform surveillance endoscopies every 3–5 years, depending on the society guidelines and the length of the Barrett’s segment.^1,5–7 Despite the criteria for the diagnosis of dysplasia being clearly defined, there is high interobserver variability among even expert pathologists, particularly in the diagnosis of LGD.²⁰ If LGD is detected, it should be confirmed by expert pathology review, given that most LGD diagnosed in the community is downgraded to NDBE when reviewed by experienced gastrointestinal pathologists.²⁰ Recent guidelines suggest that endoscopic eradication therapy (EET) can be considered for confirmed LGD cases. However, if EET is not pursued, surveillance in 6 months done by a Barrett’s expert using advanced imaging followed by annual surveillance is also reasonable. All major gastroenterology societies recommend EET for confirmed HGD, given the high rates of progression to EAC, and the efficacy and cost-effectiveness of EET in the treatment of HGD.^1,5–7,21

Early-stage EAC (also known as superficial EAC) includes Tis (also known as HGD), T1a (limited to mucosa), and T1b (limited to submucosa). Cancers deeper than these should be treated with surgical resection and lymph node dissection. In recent years, endoscopic therapy has emerged as an alternative to esophagectomy for the treatment of some superficial EAC, therefore, determining the depth of tumor invasion is extremely valuable as it can guide the optimal treatment plan. Most T1 cancers can be completely resected endoscopically, which can provide precise staging as well as potentially curative resection. Lesions limited to mucosa (T1a) can be effectively treated endoscopically via endoscopic mucosal resection (EMR) with excellent long-term outcomes reported in the literature.²² However, the optimal treatment approach for T1b lesions is less well defined, as the risk of lymph node metastasis in T1b lesions can be significant. In general, in patients with T1b lesions who are fit for surgery, the standard of care has been esophagectomy with regional lymphadenectomy. Some studies have reported promising survival data for endoscopic resection in patients with T1b EAC with “low-risk features,” defined as lesions <2 cm, with moderately or well-differentiated histology, and superficial submucosal invasion (<500 μ or T1b SM1).^23,24 The American Gastroenterological Association (AGA) and European Society of Gastrointestinal Endoscopy (ESGE) guidelines recommend EMR for most smaller esophageal lesions, reserving endoscopic submucosal dissection for selected cases, such as lesions larger than 15 mm, poorly lifting lesions, and lesions at risk for carcinoma with submucosal invasion.^25,26 While determining the depth of tumor invasion is extremely valuable for guiding the optimal treatment approach, even under trained eyes, it is extremely difficult to differentiate the different depths, which can result in choosing an inappropriate resection plan.

Overall, the aforementioned challenges have led to modest effectiveness of endoscopic surveillance and highlight the profound need for innovative solutions to improve the effectiveness of current surveillance strategies.

AI in Gastroenterology: A Brief Review

In the past few years, AI has been increasingly utilized in medicine in order to improve patients’ outcome and physician performance. AI comprises the use of computers or machines to perform critical analysis in cognitive tasks, similar to humans.²⁷ Machine learning (ML), as a subdiscipline of AI, describes algorithms employed to learn from preexisting data. Deep learning (DL) is a subtype of ML in which a convolutional neural network (CNN) receives input, learns specific pattern, and processes this information through the multilayered network to produce an output.²⁸ In gastrointestinal endoscopy, this output has been the mainstay of the development and expansion in the field of computer-aided detection (CAD) and computer-aided diagnosis (CADx). CAD systems are trained using still images to discern characteristics of neoplastic lesions, eventually progressing to video recordings, and then to live endoscopy. AI has the potential to improve several aspects of endoscopic surveillance for BE.

AI in BE

AI for the diagnosis of BE-related neoplasia (Table 1)

AI comprises the use of computers or machines to perform critical analysis in cognitive tasks, similar to human beings.²⁷ Its use in medicine has exploded, improving dictation and electronic medical record (EMR) software in clinics. It has been incorporated into hardware, creating intelligent prostheses and surgical assistant robots.²⁹ Neural networks are a subset of AI that involves multiple nodes containing an input and output layer.²⁸ When the output of one node reaches a certain threshold, it activates and passes its information to the next node. DL is a neural network of three or more nodes that can analyze unstructured information in the form of text or images and categorize it based on structured training.²⁸ DL was first applied in endoscopy as an adjunct in colorectal cancer surveillance.³⁰ Computer-aided diagnosis (CADx) systems were trained using still images to discern characteristics of neoplastic lesions, eventually progressing to video recordings, and then to live endoscopy. These CADx systems could predict histological diagnoses of lesions depicted on NBI with a sensitivity and specificity of 93%.³¹ Given the current performance deficiencies in Barrett’s surveillance, application of DL methods was not only needed but also welcomed. AI systems can improve the detection of early neoplastic lesions suitable for minimally invasive endoscopic treatment, which is the goal of endoscopic surveillance. This improved diagnostic accuracy in the community has the potential to reduce the number of referral cases and associated costs, especially once a standardized CNN is offered to community sites. In addition, the AI powered decision support can increase the community’s physician’s confidence, limiting the unnecessary repeated upper endoscopies for avoiding adverse events and improved costs. With AI, random biopsies (Seattle protocol) can be replaced by “high-quality targeted biopsies.” This can improve the sampling sensitivity and reduce missed lesions, speed up examination time, limit the use of staining resulting in shorter test time and improved patient experience, and reduce the number of biopsies and associated costs.

Initial Barrett’s AI studies utilized small datasets of retrospectively collected still images and demonstrated impressive diagnostic performance of CNN, with sensitivity, specificity, and accuracy of 83–96%, 42–100%, and 85–92%.^14,32–46

Table 1.

Studies Applying AI in the Detection of Neoplasia in Patients with Barrett’s Esophagus

	Author	Year	Aim	Algorithm design	Data-set acquisition method	Training & testing set size	Performance
CADx system	Van der Sommen et al.¹⁴	2016	Efficacy of CAD in detecting early neoplasia in BE	Image-based	Retrospective	100 images: 60 images from EAC and HGD, 40 images without dysplasia	86% sensitivity, 87% specificity
	De Groof et al.³²	2019	Developing a CAD system to identify BERN	Image-based	Retrospective	60 slides with either EAC or HGD, 20 slides with non-dysplastic Barrett’s mucosa.	92% accuracy, 95% sensitivity, 85% specificity
	Ebigbo et al.³³	2019	Developing a CAD to identify BE and early EAC	Image-based	Retrospective	248 images; 50 of EAC, 63 of non-neoplastic Barrett’s mucosa	Sen 92%, Spec 100%; WLI: Sen 97%, Spec 88%; NBI: Sen 94%, Spec 80%
	Ghatwary et al.³⁴	2019	Evaluating deep learning methods in identifying EAC	Image-based	Retrospective	50 images of EAC, 50 images of normal mucosa.	96% sensitivity, 92% specificity
	Hashimoto et al.³⁹	2020	Efficacy of a CNN in recognition of early esophageal neoplasia in BE	Image-based	Retrospective	916 images of HGD and T1 adenocarcinoma, 916 control images	96.4% sensitivity, 94.2% specificity, 95% accuracy
	Liu et al.³⁵	2020	Applying a CNN to identify EC	Image-based	Retrospective	354 images of esophageal cancer, 387 of prem alignancy, and 531 of normal mucosa.	85% accuracy, 94% sensitivity, 95% specificity
	De Groof et al.³⁶	2020	Developing a CNN to detect early neoplasia in patients with BE	Image-based	Retrospective	Three datasets were involved in the performance. Dataset 3: 297 images of early BE neoplasia and non-dysplastic BE. Dataset 4: 40 images of early BE neoplasia and 40 images of non-dysplastic BE. Dataset 5: 40 images of subtle BE neoplasia, and 40 images of non-dysplastic BE.	89% accuracy, 90% sensitivity, 88% specificity. CAD outperformed general endoscopists (88% vs 73% accuracy, 93% vs 72% sensitivity, 83% vs 74% specificity).
	Ebigbo et al.⁴⁶	2020	Analyzing the efficacy of CAD in differentiating normal BE and early EAC.	Image-based	Prospective	36 images of early EAC, 26 of normal BE.	Sensitivity 83.7%, specificity 100%, accuracy 90%.
	Van der Putten et al.³⁴	2020	Applying CNN in detecting dysplastic lesions in BE.	Image-based	Retrospective and prospective	Two datasets. Dataset 1: 33 images of early EAC, 41 areas of non-neoplastic BE mucosa. Dataset 2: 17 images of early EAC, 22 images of non-neoplastic BE mucosa.	95% sensitivity, 95% specificity. AI outperformed endoscopist by 5%.
	Van der Putten et al.³⁷	2020	Development and validation of CNN in identifying BE dysplasia	Image-based	Prospective	80 images of HGD or early EAC, 80 images of non-dysplastic BE mucosa.	Accuracy 89%, sensitivity 90%, specificity 87.5%.
	Iwagami et al.⁴⁵	2021	Developing a computerized image analysis system using AI to diagnose EAC	Image-based	Retrospective	36 images of cancer, 43 controls	94% sensitivity, 42% specificity, 66% accuracy
	De Souza et al.³⁸	2021	Detection of Barrett esophagus and EAC using various CNN	Image-based	Retrospective	50 images of early EAC, 50 images of non-dysplastic BE	84% sensitivity, 85% specificity (VGG16 model); 86% accuracy, 88% sensitivity (ResNet50 model)
	Hussein et al.⁵⁰	2022	Applying CNN to identify early neoplasia in BE	Image-based	Retrospective	28 images of dysplasia, 33 non-dysplastic mucosa.	91% sensitivity, 79% specificity, AUROC 93%.
	Van der Laan et al.⁴⁴	2023	Applying AI in optical biopsy in BE	Image-based	Prospective	710 images of EC metaplasia, neoplasia, or non-dysplastic BE mucosa.	AI improved sensitivity and accuracy by 20% and 12.2%, respectively.
	Tsai et al.⁴³	2023	Detection of histological BE using AI.	Image-based	Retrospective	70 images of BE, 90 images of normal mucosa.	Accuracy of 94.4%, sensitivity of 94.3 and specificity and 94.4%
	Fockens et al.⁴²	2023	Creating a CAD system to assist in BERN detection	Image-based	Prospective	Dataset 1: 50 neoplastic images, 150 non-dysplastic BE images. Dataset 2: 50 neoplastic images, 50 non-dysplastic BE images. Dataset 3: 50 neoplastic images, 150 non-dysplastic BE images.	88% sensitivity, 66% specificity.
	Takeda et al.⁴¹	2024	Applying AI in identification of BE.	Image-based	Retrospective	624 patients (542 WLI); 643 CLI)77 images of short segment BE, 49 images of non-short segment BE	WLI: 84.1% accuracy, 89.6% sensitivity, 75.5% specificityCLI: 90.5% accuracy, 90.1% sensitivity, 91.1% specificity
	Van der Putten et al.⁴⁷	2020	Applying a deep learning system in identifying BE with NBI — zoom imagery	Video-based	Retrospective	59 videos of dysplasia, 100 videos of non-dysplastic Barrett’s esophagus	84% accuracy, 83% sensitivity, 84% specificity
Real-time videos analysis	De Groof et al.⁴⁸	2020	Determining the diagnostic accuracy of a CAD system in identifying BE	Video-based	Prospective	1. Validation-only dataset: 80 HDWLE images of 80 patients2. Dataset for validation and comparison to 53 nonexpert endoscopists: 80 HD-WLE images of 80 patients	1. CADx: sensitivity, 90%; specificity, 88%; accuracy, 89% CADe: optimal biopsy spot in 97%2. CADx: sensitivity, 93%; specificity, 83%; accuracy, 88% CADe: optimal biopsy spot in 92% Endoscopists: sensitivity, 72%; specificity, 74%; accuracy, 73%
	Struyvenberg et al.⁴⁹	2021	Diagnostic performance of CAD system in identifying BE.	Image and video-based	Retrospective	59 neoplastic videos, 98 non-dysplastic BE videos.	Images: 84% accuracy, 88% sensitivity, 78% specificityVideo: 83% accuracy, 85% sensitivity, 83% specificity
	Ebigbo et al.⁵⁴	2022	Detection and Delineation of BE-related neoplasia in real-time endoscopic video	Video-based	Prospective	2 videos, both with BE-related neoplasia.	N/A
	Kusters et al.⁵²	2022	Developing an endoscopy driven CAD system in identifying BE	Image and video-based	Retrospective and prospective	50 images of BE neoplasia, 111 images of non-dysplastic BE. 40 videos of BE neoplasia, 83 videos of non-dysplastic BE.	Images: Accuracy 92.5%, Sensitivity 91.5%, Specificity 92.9%Video: Accuracy 93.8%, Sensitivity 96%, Specificity 92.8%.
	Abdelrahim et al.⁵⁵	20222023	Developing and externally validating a CAD system to identify and classify BE.	Image and video-based	Retrospective and Prospective	107 images of neoplastic BE, 364 images of non-dysplastic BE. 32 videos of neoplastic BE, 43 with non-dysplastic BE.	Images: 95% sensitivity, 94% specificity, 95% accuracy.Video: 94% sensitivity, 91% specificity, 95% accuracy.
	Hussein et al.⁵³	2023	Developing a CAD model to diagnose BERN	Video-based	Retrospective	34 cases of dysplasia (19 HGD, 15 IMC), 23 cases with non-dysplastic BE.	Images: 94% sensitivity, 86% specificity, AUROC 96%.Videos: 92% sensitivity, 82% specificity, AUROC 95%.
	Fockens et al.⁵⁸	2023	Efficacy of CAD system in differentiating BE from NDBE during live endoscopy	Video-based	Prospective	15 neoplastic cases, 15 non-dysplastic BE cases.	100% sensitivity
	Fockens et al.⁵¹	2023	Developing a CNN to detect early neoplasia in BE.	Image and video-based	Images and videos	82 neoplastic images, 327 non-dysplastic BE images. 71 neoplastic videos, 180 non-dysplastic BE videos.	Accuracy of 90% for images, 91% for videos. CNN outperformed endoscopists in accuracy (90% vs 74% in images, 91% vs 67% in videos).
	Jukema et al.⁵⁹	2024	Applying AI in endoscopic identification of early BE neoplasia.	Image and video-based	Prospective	45 neoplastic images, 83 non-dysplastic BE images.	Images: 100% sensitivity, 98% specificityVideo: 93% sensitivity, 96% specificity
	Meinikheim et al.⁵⁶	2024	Development of CNN in assisting in diagnosis of BE	Video-based	Prospective	45 videos of non-dysplastic BE, five cases of LGD, seven cases of HGD, 36 cases of T1a adenocarcinoma, three cases of T1b adenocarcinoma	92% sensitivity, 69% specificity, 81% accuracy.

AI, artificial intelligence; AUROC, area under the receiver operator curve; BE, Barrett’s esophagus; BERN, Barrett’s esophagus-related neoplasia; CAD, computer-aided detection; CLI, color-linked imaging; CNN, convoluted neural network; EAC, esophageal adenocarcinoma; EC, esophageal cancer; GE, gastroesophageal; H&E, hematoxylin and eosin; HGD, high-grade dysplasia; LGD, low-grade dysplasia; NBI, narrow band imaging; NDBE, nondysplastic Barrett’s esophagus; NLP, natural language processing; WLI, white light imaging.

The performance of CNN in these early studies inspired further research, integrating CNN in real-time video analysis and direct comparison with general endoscopists. Video-based studies trained CNN to assess the lumen in real time, breaking down the screen into 60 × 60 pixel frames, labeling them as either “neoplastic” or “non-neoplastic.” Studies investigating CNN using real-time video identified BE lesions from video frames with accuracy, sensitivity, and specificity ranges of 84–90%, 83–92%, and 82–89%.^47–56 Next, studies bolstered their datasets to enhance neural network training. The Barrett's Oesophagus Imaging for Artificial Intelligence (BONS-AI) consortium, for example, used data from 15 international centers demonstrating a synergistic effect of CADx during video, improving endoscopist sensitivity from 67% to 79% without compromising specificity.⁵¹ Recent studies utilized prospective data to diversify the dataset in which CNN is tested, with sensitivity and specificity of 84–100% and 79–100%, respectively.^{36,45,46,57,58} These studies directly assess the performance of CNN using images of early BE neoplasia and non-BE dysplasia, benchmarked by performance by general endoscopists.⁵⁹ The methodological rigor of recent studies has shown that CNN can be an effective tool in improving endoscopic detection of BE neoplasia in real time, with a synergistic effect when used with endoscopists. These promising results have highlighted a performance benefit over conventional diagnostic endoscopy, calling into question the current gold standard. Application of CNN raises the diagnostic performance of BE detection during screening endoscopy, closing the diagnostic performance gap between community hospitals and specialized centers.

AI for depth of invasion (Table 2)

As stated earlier, preoperative differentiation between mucosal (T1a) and submucosal (T1b) cancers has relevant therapeutic and prognostic implications. A pivotal study by Knabe et al. demonstrated an overall accuracy rate of 73% in the classification of T stages in Barrett’s cancer.⁶⁰ Another study by Ebigbo et al. demonstrated sensitivity and specificity of 77% and 64% for differentiating between T1a and T1b lesions.⁶¹ Van der Laan et al. demonstrated that AI-assisted gastroenterologists performed better than their unassisted counterparts in the identification of dysplasia in optical biopsies of BE generated by endocystoscopy.⁴⁴ Although its diagnostic performance currently remains comparable to current gold standard measures, the CADx system can improve with more robust training regimens and expansion of labeled datasets, with real potential to serve as a timely and accurate biopsy predictor during real-time endoscopy.

Table 2.

Performance of Studies Applying AI for Predicting the Depth of Invasion

Assessing depth of invasion	Ebigbo et al.⁶¹	2021	Developing a CNN to differentiate between T1a and T1b Barrett’s cancer	Image based	Retrospective	108 T1a, 122 T1b images	77% sensitivity, 64% specificity, 71% accuracy.
Assessing depth of invasion	Knabe et al.⁶⁰	2022	Determine the accuracy of CNN in T staging Barrett’s carcinoma	Image based	Retrospective	18 images of non-dysplastic BE, 119 images of T1a, 11 images of T2, 44 images of T3, 7 images of T4.	Early Barrett’s neoplasia: 57% sensitivity, 77% specificity, 67% accuracyAdvanced stage (T3 / T4): 71% sensitivity, 73% specificity, 73% accuracy

BE, Barrett’s esophagus; CNN, convoluted neural network.

AI for Barrett’s related histology (Table 3)

As stated earlier, the clinical care of patients with esophageal diseases relies on the histological evaluation of biopsies and resection specimens. In this regard, the application of AI to digital endoscopy will play a key role in improving and standardizing clinical care. Gehrung et al. applied DL to the Cytosponge-TFF3 test, a noninvasive diagnostic test for BE.⁶² DL was implemented to create a human-in-the-loop triage process, with AI flagging difficult cases for expert pathology review with similar sensitivity and specificity to fully manual reviews.⁶² DL can interpret voice biomarkers, particularly voice signal periodicity, to differentiate patients with normal esophageal mucosa, gastroesophageal reflux disease (GERD), and BE.⁶³ TissueCypher implements DL to interpret biomarker expression and tissue structures of esophageal biopsies to generate scores denoting low, mid, and high risk of progression from NDBE to HGD or EAC.⁶⁴ AI has also been applied to augment primary prevention through the advent of electronic health record-based ML models. These models can identify high-risk patients from EMR documentation by generating a predictive risk score, improving the efficacy of primary prevention screening.⁶⁵ CNN has also been applied in the analysis of histopathology.^66–69 Tomita et al. developed a CNN that differentiated normal tissue, NDBE, BE with dysplasia, and adenocarcinoma with a mean accuracy of 83%.⁷⁰ Faghani et al. demonstrated CNN could categorize histopathology slides into NDBE, LGD, or HGD with an F1 score between 0.91 and 1.0.⁷¹ When applied with wide angle transepithelial sampling, computerized microscopy can improve diagnostic accuracy of BE.^72,73 This technology creates a three-dimensional rendering of the tissue biopsy, utilizing neural networks to compare it with thousands of images of BE dysplasia. In turn, this CNN can identify pathological biopsies with much greater accuracy and improve interobserver bias of pathologists compared with forceps biopsy.⁷⁴

Table 3.

Performance of Studies Applying AI for Predicting Histology

Application in histology	Sali et al.⁶⁶	2020	Implement deep learning techniques to detect BE and dysplasia in H&E biopsies	Image-based	Retrospective	120 patients with dysplastic BE, non-dysplastic BE, or normal squamous tissue.	95% accuracy, 99% AUROC, 94% precision
	Guleria et al.⁶⁷	2021	Applying deep learning methods to histopathology images and videos to identify dysplasia in BE.	Image and video-based	Retrospective	387 images of normal squamous tissue, non-dysplastic BE, and dysplasia.	Images: 90% sensitivity, 94% accuracy.
	Codipilly et al.⁶⁸	2021	Applying artificial intelligence to diagnose BE on histology slides.	Image-based	Retrospective	168 non-dysplastic BE slides, 266 LGD slides, 153 HGD slides.	Non-dysplastic BE: 93% sensitivity, 100% specificityHigh-grade dysplasia: 100% sensitivity, 99% specificity
	Faghani et al.⁷¹	2022	Developing a CNN to predict dysplasia grade on histology slides.	Image-based	Retrospective	164 non-dysplastic BE slides, 226 LGD slides, 152 HGD slides.	81% sensitivity, 100% specificity, F1 score 0.91 (NDBE), 0.9 (LGD), 1.0 (HGD)
	Nguyen Wenker et al.⁶⁹	2023	Developing a NLP to identify dysplasia in BE patients on histopathology reports	Document-based (pathology reports)	Retrospective	400 reports	99% accuracy, 92% recall, 100% precision, F-measure of 96%.

AUROC, area under the receiver operator curve; BE, Barrett’s esophagus; CNN, convoluted neural network; H&E, hematoxylin and eosin; HGD, high-grade dysplasia; LGD, low-grade dysplasia; NDBE, nondysplastic Barrett’s esophagus; NLP, natural language processing.

AI for quality control of upper gastrointestinal endoscopy (Table 4)

AI has the potential to be integrated during endoscopy to augment efficiency and assure quality. Quality improving systems such as WISENSE can monitor blind spots, automatically generate photo documentation, and time the procedure, improving not just Adenoma Detection Rate (ADR) but also endoscopy efficiency.⁷⁵ In addition, AI has been implemented to reduce interobserver variance, with algorithms designed to automatically detect the GEJ and interpret BE extension using the Prague classification (Table 4).^61,62 One study developed an AI system that automatically identifies the squamous-columnar junction and GEJ on images.⁵⁷ Another study worked on an AI system that automatically determined the extension of the BE according to the Prague classification.⁷⁶ With the extension of BE as a relevant factor for risk stratification, AI-assisted standardized and automated reporting has the potential to significantly improve patient care.

Table 4.

Performance of Studies Applying AI for Quality Improvement

Application in quality improvement	Ali et al.⁷⁶	2021	Applying CNN to quantify the area of Barrett’s epithelium and islands	Video-based	Retrospective	194 videos	97% accuracy for Prague classification and island measurements. 98% accuracy of quantifying Barrett’s epithelium area.
Application in quality improvement	Pan et al.⁵⁷	2021	Efficacy of a deep learning method to identify BE.	Image-based	Retrospective	443 images (187 patients)	Intersection over union was 0.56 for gastroesophageal junction and 0.82 for squamocolumnar junction

BE, Barrett’s esophagus; CNN, convoluted neural network.

AI can integrate the results of the endoscopy to produce guideline-based recommendations on further surveillance measures, while also completing billing documentation. With AI, an endoscopy “report card” can be calculated, analyzing the endoscopist’s ability to avoid blind spots, effectively irrigate, resect polyps, and biopsy suspicious lesions. For training programs, CNN can be used as a standardized measure to evaluate the competency of trainees as they transition to independent practice. These advantages serve to improve not just the efficiency but also the performance of endoscopists (Table 5).

Table 5.

Applications of AI as a Quality Improvement Tool

Domain	Current gap	Artificial intelligence application
General endoscopic practices	Quality of endoscopy varies widely between endoscopists. Manual capturing of photos and timing of procedure.	AI applications can automatically document photos, assess pre-endoscopic preparation, time withdrawal, and decrease the rate of blind spots.^75,86–88
Determination of Prague Classification for risk stratification of BE.	Significant under-utilization in community settings, with uptake of 22–69%.	Advanced deep learning can generate C&M values with 97.2% accuracy and determine the area of Barrett epithelium with 98.4% accuracy.⁷⁶
Assessment and acquisition of endoscopic skills	Subjective assessment of trainee by experienced endoscopist	AI can objectively assess mucosal visualization and automatically track statistics including cecal intubation, leading to personalized learning plans and objective standards of competency.⁸⁹
Surveillance screening	Random biopsies were taken with historically low sensitivity and sampling error.⁹⁰ Low adherence to Seattle biopsy protocol.	Advanced imaging techniques bolstered by CNN can offer targeted biopsies with sensitivity and specificity of 90% and 88%, respectively.³⁶
Post-endoscopic deliverables	Manual generation of endoscopic report	AI can automatically generate a post-endoscopic report incorporating evidence-based guidelines on further surveillance.

AI, artificial intelligence; BE, Barrett’s esophagus; CNN, convoluted neural network.

AI systems approved for clinical use

Several AI systems have been developed, demonstrating impressive ability of the CADx systems to identify dysplasia and early EAC. WISE VISION (NEC Corp, Tokyo, Japan) can differentiate between NDBE and dysplastic BE and offers a visual representation of the area that has been classified as HGD or neoplastic.⁷⁷ CADU (Odin Medical Ltd, London, UK) can differentiate between NDBE and dysplastic BE and offers a visual representation when the observed image is deemed dysplastic.⁷⁸

Barriers to implementation

Several barriers have slowed the integration of AI into the endoscopy suite. From a development standpoint, more diverse, well-annotated images are needed to create validated, robust neural networks. Preliminary studies highlight the limitations of ML exclusively on high-quality image datasets, creating neural networks without dexterity for the real world.^33,42,79

From a financial standpoint, governing bodies must purchase AI products designed by consortiums and industry companies, evaluating the expected life of the machines and the need for upgrading software. To allow for equitable access, medical centers across the authority’s jurisdiction would need to be accounted for. This results in both an impressive one-time cost as well as recurring costs anticipated for upgrades and repair.

For experienced endoscopists, AI is a novel, and at times, frighteningly powerful technology. Machine algorithms operate decisively and more accurately than humans; however, users still feel entitled to transparency and accountability in decision-making, coined the “black box” criticism.^38,80 Thankfully, systems are being trained to integrate explanations into decision-making, allowing the user to make an informed decision and overrule when needed.⁸¹

Procedurally, the AI-generated alert boxes during endoscopy can be viewed as “distracting” or “redundant” by users.⁸² Thankfully, the user experience will only improve with continual feedback, with ongoing studies aimed to minimize distractions.⁸³ Ultimately, CNN should be viewed as a support tool with immense growth potential, aimed to improve patient care and physician efficiency.

A proposed value chain

The promise of AI is to improve patients’ outcomes, physician performance, and workload. To summarize the various benefits of integrating AI into Barrett’s clinical practice, we have developed a value chain, encompassing the entire continuum of care.

Once AI becomes a clinical reality available to physicians, the proposed use during upper endoscopy will be with live video images that will be sent to the AI application and analyzed in real time. The application will be a full workflow solution starting with a quality algorithm capable of monitoring blind spots and mucosal cleanness. It will then help reduce interobserver bias by automatically measuring the size of Barrett’s segment and documenting the Prague classification. Throughout the procedure, blind spots would be identified, and photos automatically captured and stored (Figure 1).

Fig. 1.

A potential value chain depicting artificial intelligence (AI) integration in Barrett’s esophagus (BE) surveillance and management.

As CADx is integrated into surveillance endoscopy, detection rates of HGD and early EAC can improve. Optical biopsy would also save patients from further diagnostic procedures for tissue sampling and offer unparalleled efficiency in therapeutic management. These measures can reduce health care costs through more effective malignancy screening, reducing costs associated with chemotherapy and surgical interventions. Moreover, they will improve patient safety, protecting them from the risks of surgery or further, unnecessary investigations. Endoscopists benefit dramatically from the integration of CADx systems. At the end of the procedure, an impression would be generated, summarizing the salient findings and providing recommendations on management and screening interval based on guideline recommendations. AI tool can then proceed with automated clinical and procedural documentation that can be automatically incorporated into daily clinical practice. AI tools will likely also be available to help optimize patient scheduling and procedural workflow.

Conclusions

EAC is a morbid disease, in part due to poor performance of identification of dysplasia and early EAC during surveillance screening.⁸⁴ AI offers unparalleled convenience and growth opportunities, creating an “arms race” between consortiums to build CNN that is accurate, reliable, and scalable. CADx system can differentiate NDBE from LGD, HGD, and early EAC with impressive accuracy, sensitivity, and specificity. Its ability to predict the depth of infiltration offers immense financial and safety benefits. In addition, CNN can improve endoscopic efficiency, offer comparative performance metrics, and serve as a new gold standard for trainees.⁸⁵ Its multiple applications in quality improvement and assurance serve as a worthwhile investment for private medical centers and governing health authorities. For the practicing gastroenterologist, this technology still requires further validation before routine incorporation, but nonetheless has the exciting potential to advance the field of endoscopy.

Footnotes

Authors’ Contributions

Conception and design: J.B., H.J.K., M.F.B., and N.P. Analysis and interpretation of data: J.B., M.F.B., and N.P. Drafting of article: J.B. and H.J.K. Critical revision of the article for important intellectual content: M.F.B. and N.P. Final approval of the article: M.F.B. and N.P.

Author Disclosure Statement

N.P.: VP of Medical Affairs at Satisfai Health, Speaker for Phathom Pharmaceuticals. M.F.B.: Founder, Chairman, and CMO at Satisfai Health. J.B.: No conflicts to disclose. H.J.K.: No conflicts to disclose.

Funding Information

No funding was received for this article.

References

Shaheen

, Falk

, Iyer

, et al. Diagnosis and management of Barrett’s esophagus: An updated ACG guideline. Am J Gastroenterol, 2022; 117(4):559–587; doi: 10.1038/ajg.2015.322

Thrift

. Barrett’s esophagus and esophageal adenocarcinoma: How common are they really? Dig Dis Sci, 2018; 63(8):1988–1996; doi: 10.1007/s10620-018-5068-6

Qumseya

, Yang

, Guo

. Trends in prevalence of esophageal adenocarcinoma: Findings from a statewide database of over 6 million patients. Endosc Int Open, 2024; 12(2):E218–E226; doi: 10.1055/a-2221-7974

Sheikh

, Roshandel

, McCormack

, et al. Current status and future prospects for esophageal cancer. Cancers (Basel), 2023; 15(3):765; doi: 10.3390/cancers15030765

Weusten

, Bisschops

, Coron

, et al. Endoscopic management of Barrett’s esophagus: European Society of Gastrointestinal Endoscopy (ESGE) position statement. Endoscopy, 2017; 49(2):191–198; doi: 10.1055/a-2176-2440

Fitzgerald

, Di Pietro

, Ragunath

, et al.; British Society of Gastroenterology. British society of gastroenterology guidelines on the diagnosis and management of Barrett’s oesophagus. Gut, 2014; 63(1):7–42; doi: gutjnl-2013-305372

Qumseya

, Sultan

, Bain

, et al.; ASGE Standards of Practice Committee Chair. ASGE guideline on screening and surveillance of Barrett’s esophagus. Gastrointest Endosc, 2019; 90(3):335–359.e2; doi: 10.1016/j.gie.2019.05.012

Wani

, Yadlapati

, Singh

, et al.; Post-Endoscopy Esophageal Neoplasia Expert Consensus Panel. Post-endoscopy esophageal neoplasia in Barrett’s esophagus: Consensus statements from an international expert panel. Gastroenterology, 2022; 162(2):366–372; doi: 10.1053/j.gastro.2021.09.067

Gupta

, Gaddam

, Wani

, et al. Longer inspection time is associated with increased detection of high-grade dysplasia and esophageal adenocarcinoma in Barrett’s esophagus. Gastrointest Endosc, 2012; 76(3):531–538; doi: 10.1016/j.gie.2012.04.470

10.

Gill

, Singh

. Endoscopic imaging in Barrett’s esophagus: Current practice and future applications. Ann Gastroenterol, 2012; 25(2):89.

11.

Peleg

, Ollech

, Shamah

, et al. Seattle protocol is more effective in detection of dysplasia compared to technology-assisted targeted biopsies in patients with Barrett’s esophagus. J Clin Med, 2023; 12(7):2544; doi: 10.3390/jcm12072544

12.

Waterhouse

, Januszewicz

, Ali

, et al. Spectral endoscopy enhances contrast for neoplasia in surveillance of Barrett’s esophagus. Cancer Res, 2021; 81(12):3415–3425; doi: 10.1158/0008-5472.CAN-21-0474

13.

Moriarty

, Shah

, Rubenstein

, et al. Costs associated with Barrett’s esophagus screening in the community: An economic analysis of a prospective randomized controlled trial of sedated versus hospital unsedated versus mobile community unsedated endoscopy. Gastrointest Endosc, 2018; 87(1):88–94.e2; doi: 10.1016/j.gie.2017.04.019

14.

F van der

, Zinger

, Curvers

, et al. Computer-aided detection of early neoplastic lesions in Barrett’s esophagus. Endoscopy, 2016; 48(7):617–624; doi: 10.1055/s-0042-105284

15.

Visrodia

, Singh

, Krishnamoorthi

, et al. Magnitude of missed esophageal adenocarcinoma after Barrett’s esophagus diagnosis: A systematic review and meta-analysis. Gastroenterology, 2016; 150(3):599–607.e7; doi: 10.1053/j.gastro.2015.11.040

16.

Sawas

, Majzoub

, Haddad

, et al.; Post-Endoscopy Esophageal Adenocarcinoma Consensus Panel. Magnitude and time-trend analysis of postendoscopy esophageal adenocarcinoma: A systematic review and meta-analysis. Clin Gastroenterol Hepatol, 2022; 20(2):e31–50–e50; doi: 10.1016/j.cgh.2021.04.032

17.

Pech

, Gossner

, Manner

, et al. Prospective evaluation of the macroscopic types and location of early Barrett’s neoplasia in 380 lesions. Endoscopy, 2007; 39(7):588–593; doi: 10.1055/s-2007-966363

18.

Abela

, Going

, Mackenzie

, et al. Systematic four-quadrant biopsy detects Barrett’s dysplasia in more patients than nonsystematic biopsy. Am J Gastroenterol, 2008; 103(4):850–855; doi: 10.1111/j.1572-0241.2007.01746.x

19.

Abrams

, Kapel

, Lindberg

, et al. Adherence to biopsy guidelines for Barrett’s esophagus surveillance in the community setting in the United States. Clin Gastroenterol Hepatol, 2009; 7(7):736–742; quiz 710; doi: 10.1016/j.cgh.2008.12.027

20.

Curvers

, ten Kate

, Krishnadath

, et al. Low-grade dysplasia in Barrett’s esophagus: Overdiagnosed and underestimated. Am J Gastroenterol, 2010; 105(7):1523–1530; doi: 10.1038/ajg.2010.171

21.

Wani

, Qumseya

, Sultan

, et al.; Standards of Practice Committee. Endoscopic eradication therapy for patients with Barrett’s esophagus–associated dysplasia and intramucosal cancer. Gastrointest Endosc, 2018; 87(4):907–931.e9; doi: 10.1016/j.gie.2017.10.011

22.

Pech

, May

, Manner

, et al. Long-term efficacy and safety of endoscopic resection for patients with mucosal adenocarcinoma of the esophagus. Gastroenterology, 2014; 146(3):652–660.e1; doi: 10.1053/j.gastro.2013.11.006

23.

Lee

, Ronellenfitsch

, Hofstetter

, et al. Predicting lymph node metastases in early esophageal adenocarcinoma using a simple scoring system. J Am Coll Surg, 2013; 217(2):191–199; doi: 10.1016/j.jamcollsurg.2013.03.015

24.

Manner

, Pech

, Heldmann

, et al. Efficacy, safety, and long-term results of endoscopic treatment for early stage adenocarcinoma of the esophagus with low-risk sm1 invasion. Clin Gastroenterol Hepatol, 2013; 11(6):630–635; quiz e45; doi: 10.1016/j.cgh.2012.12.040

25.

Pimentel-Nunes

, Dinis-Ribeiro

, Ponchon

, et al. Endoscopic submucosal dissection: European Society of Gastrointestinal Endoscopy (ESGE) Guideline. Endoscopy, 2015; 47(9):829–854; doi: 10.1055/s-0034-1392882

26.

Draganov

, Wang

, Othman

, et al. AGA institute clinical practice update: Endoscopic submucosal dissection in the United States. Clin Gastroenterol Hepatol, 2019; 17(1):16–25.e1; doi: 10.1016/j.cgh.2018.07.041

27.

Malik

, Pathania

, Rathaur

, et al. Overview of artificial intelligence in medicine. J Family Med Prim Care, 2019; 8(7):2328–2331; doi: 10.4103/jfmpc.jfmpc_440_19

28.

What is Machine Learning? | IBM [Internet]. Available from: https://www.ibm.com/topics/machine-learning [Last accessed: December 7, 2023].

29.

Juarez

, Marcos

, Stiglic

, et al. Artificial intelligence in medicine: 21st international conference on artificial intelligence in medicine, AIME 2023, Portorož, Slovenia, June 12–15, 2023, Proceedings [Internet]. Cham: Springer Nature Switzerland; 2023 [December 7, 2023]. (Lecture Notes in Computer Science; vol. 13897). https://link-springer-com-443.web.bisu.edu.cn/10.1007/978-3-031-34344-5

30.

Tischendorf

JJW

, Gross

, Winograd

, et al. Computer-aided classification of colorectal polyps based on vascular patterns: A pilot study. Endoscopy, 2010; 42(3):203–207; doi: 10.1055/s-0029-1243861

31.

Kominami

, Yoshida

, Tanaka

, et al. Computer-aided diagnosis of colorectal polyp histology by using a real-time image recognition system and narrow-band imaging magnifying colonoscopy. Gastrointest Endosc, 2016; 83(3):643–649; doi: 10.1016/j.gie.2015.08.004

32.

de Groof

, van der Sommen

, van der Putten

, et al. The Argos project: The development of a computer-aided detection system to improve detection of Barrett’s neoplasia on white light endoscopy. United European Gastroenterol J, 2019; 7(4):538–547; doi: 10.1177/2050640619837443

33.

Ebigbo

, Mendel

, Probst

, et al. Computer-aided diagnosis using deep learning in the evaluation of early oesophageal adenocarcinoma. Gut, 2019; 68(7):1143–1145; doi: 10.1136/gutjnl-2018-317573

34.

Ghatwary

, Zolgharni

, Ye

. Early esophageal adenocarcinoma detection using deep learning methods. Int J Comput Assist Radiol Surg, 2019; 14(4):611–621; doi: 10.1007/s11548-019-01914-4

35.

Liu

, Hua

, Wu

, et al. Automatic classification of esophageal lesions in endoscopic images using a convolutional neural network. Ann Transl Med, 2020; 8(7):486; doi: 10.21037/atm.2020.03.24

36.

de Groof

, Struyvenberg

, van der Putten

, et al. Deep-Learning system detects neoplasia in patients with Barrett’s Esophagus with higher accuracy than endoscopists in a multistep training and validation study with benchmarking. Gastroenterology, 2020; 158(4):915–929.e4; doi: 10.1016/j.gie.2019.12.048

37.

van der Putten

, de Groof

, Struyvenberg

, et al. Multi-stage domain-specific pretraining for improved detection and localization of Barrett’s neoplasia: A comprehensive clinically validated study. Artif Intell Med, 2020; 107:101914; doi: 10.1016/j.artmed.2020.101914

38.

de Souza

, Mendel

, Strasser

, et al. Convolutional neural networks for the evaluation of cancer in Barrett’s esophagus: Explainable AI to lighten up the black-box. Comput Biol Med, 2021; 135:104578; doi: 10.1016/j.compbiomed.2021.104578

39.

Hashimoto

, Requa

, Dao

, 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(6):1264–1271.e1; doi: 10.1016/j.gie.2019.12.049

40.

Lui

TKL

, Tsui

VWM

, Leung

. Accuracy of artificial intelligence-assisted detection of upper GI lesions: A systematic review and meta-analysis. Gastrointest Endosc, 2020; 92(4):821–830.e9; doi: 10.1016/j.gie.2020.06.034

41.

Takeda

, Asaoka

, Ueyama

, et al. Development of an artificial intelligence diagnostic system using linked color imaging for Barrett’s esophagus. J Clin Med, 2024; 13(7):1990; doi: 10.3390/jcm13071990

42.

Fockens

, Jukema

, Boers

, et al. Towards a robust and compact deep learning system for primary detection of early Barrett’s neoplasia: Initial image‐based results of training on a multi‐center retrospectively collected data set. United European Gastroenterol J, 2023; 11(4):324–336; doi: 10.1002/ueg2.12363

43.

Tsai

, Yen

, Tsai

, et al. Artificial intelligence system for the detection of Barrett’s esophagus. World J Gastroenterol, 2023; 29(48):6198–6207; doi: 10.3748/wjg.v29.i48.6198

44.

van der Laan

JJH

, van der Putten

, Zhao

, et al. Optical biopsy of Dysplasia in Barrett’s oesophagus assisted by artificial intelligence. Cancers (Basel), 2023; 15(7):1950; doi: 10.3390/cancers15071950

45.

Iwagami

, Ishihara

, Aoyama

, et al. Artificial intelligence for the detection of esophageal and esophagogastric junctional adenocarcinoma. J Gastroenterol Hepatol, 2021; 36(1):131–136; doi: 10.1111/jgh.15136

46.

Ebigbo

, Mendel

, Probst

, et al. Real-time use of artificial intelligence in the evaluation of cancer in Barrett’s oesophagus. Gut, 2020; 69(4):615–616; doi: 10.1136/gutjnl-2019-319460

47.

J van der

, Struyvenberg

, de Groof

, et al. Endoscopy-Driven pretraining for classification of dysplasia in Barrett’s esophagus with endoscopic narrow-band imaging zoom videos. Appl Sci, 2020; 10(10):3407; doi: 10.3390/app10103407

48.

de Groof

, Struyvenberg

, Fockens

, et al. Deep learning algorithm detection of Barrett’s neoplasia with high accuracy during live endoscopic procedures: A pilot study (with video). Gastrointest Endosc, 2020; 91(6):1242–1250; doi: 10.1053/j.gastro.2019.11.030

49.

Struyvenberg

, De Groof

, Van Der Putten

, et al. A computer-assisted algorithm for narrow-band imaging-based tissue characterization in Barrett’s esophagus. Gastrointest Endosc, 2021; 93(1):89–98; doi: 10.1016/j.gie.2020.05.050

50.

Hussein

, González-Bueno Puyal

, Lines

, et al. A new artificial intelligence system successfully detects and localises early neoplasia in Barrett’s esophagus by using convolutional neural networks. United European Gastroenterol J, 2022; 10(6):528–537; doi: 10.1002/ueg2.12233

51.

Fockens

, Jong

, Jukema

, et al. A deep learning system for detection of early Barrett’s neoplasia: A model development and validation study. Lancet Digit Health, 2023; 5(12):e905–16; doi: 10.1016/S2589-7500(23)00199-1

52.

Kusters

CHJ

, Boers

TGW

, Jukema

, et al. A CAD System for Real-Time Characterization of Neoplasia in Barrett’s Esophagus NBI Videos. In: Cancer Prevention Through Early Detection - 1st International Workshop, CaPTion 2022, Held in Conjunction with MICCAI 2022, Proceedings [Internet]. Springer Science and Business Media Deutschland GmbH; 2022. pp. 89–98. https://researchinformation.amsterdamumc.org/en/publications/a-cad-system-for-real-time-characterization-of-neoplasia-in-barre [May 27, 2024].

53.

Hussein

, Lines

, González-Bueno Puyal

, et al. Computer-aided characterization of early cancer in Barrett’s esophagus on i-scan magnification imaging: A multicenter international study. Gastrointest Endosc, 2023; 97(4):646–654; doi: 10.1016/j.gie.2022.11.020

54.

Ebigbo

, Mendel

, Probst

, et al. Multimodal imaging for detection and segmentation of Barrett’s esophagus-related neoplasia using artificial intelligence. Endoscopy, 2022; 54(10):e587; doi: 10.1055/a-1704-7885

55.

Abdelrahim

, Saiko

, Maeda

, et al. Development and validation of artificial neural networks model for detection of Barrett’s neoplasia: A multicenter pragmatic nonrandomized trial (with video). Gastrointest Endosc, 2023; 97(3):422–434; doi: 10.1016/j.gie.2022.10.031

56.

Meinikheim

, Mendel

, Palm

, et al. Influence of artificial intelligence on the diagnostic performance of endoscopists in the assessment of Barrett’s esophagus: A tandem randomized and video trial. Endoscopy [Internet], 2024; 56(9):641–649; doi: 10.1055/a-2296-5696

57.

Pan

, Li

, Wang

, et al. Identification of Barrett’s esophagus in endoscopic images using deep learning. BMC Gastroenterol, 2021; 21(1):479; doi: 10.1186/s12876-021-02055-2

58.

Fockens

, Jukema

, Jong

, et al. Video-Based computer aided detection system detects Barrett’s neoplasia with high accuracy during live endoscopy procedures: A multi-center pilot and feasibility study. Gastrointest Endosc, 2023; 97(6, Supplement); doi: 10.1016/j.gie.2023.04.1657

59.

Jukema

, Kusters

CHJ

, Jong

, et al. Computer-Aided diagnosis improves characterization of Barrett’s neoplasia by general endoscopists. Gastrointest Endosc, 2024; S0016-5107(24)00233-5); doi: 10.1016/j.gie.2024.04.013

60.

Knabe

, Welsch

, Blasberg

, et al. Artificial intelligence-assisted staging in Barrett’s carcinoma. Endoscopy, 2022; 54(12):1191–1197; doi: 10.1055/a-1811-9407

61.

Ebigbo

, Mendel

, Rückert

, et al. Endoscopic prediction of submucosal invasion in Barrett’s cancer with the use of artificial intelligence: A pilot study. Endoscopy, 2021; 53(9):878–883; doi: 10.1055/a-1311-8570

62.

Gehrung

, Crispin-Ortuzar

, Berman

, et al. Triage-driven diagnosis of Barrett’s esophagus for early detection of esophageal adenocarcinoma using deep learning. Nat Med, 2021; 27(5):833–841; doi: 10.1038/s41591-021-01287-9

63.

Kamboj

, Yarlagadda

, Pietrowicz

, et al. S402 voice enabled artificial intelligence for detection of pathologic gastroesophageal reflux disease and Barrett’s esophagus. Am J Gastroenterol, 2022; 117(10S):e281–e282; doi: 10.14309/01.ajg.0000858248.86803.d7

64.

Frei

, Khoshiwal

, Konte

, et al. Tissue systems pathology test objectively risk stratifies Barrett’s esophagus patients with low-grade dysplasia. Am J Gastroenterol, 2021; 116(4):675–682; doi: 10.14309/ajg.0000000000001037

65.

Iyer

, Sachdeva

, Leggett

, et al. Development of electronic health record-based machine learning models to predict Barrett’s esophagus and esophageal adenocarcinoma risk. Clin Transl Gastroenterol, 2023; 14(10):e00637; doi: 10.14309/ctg.0000000000000637

66.

Sali

, Moradinasab

, Guleria

, et al. Deep learning for whole-slide tissue histopathology classification: A comparative study in the identification of dysplastic and non-dysplastic Barrett’s esophagus. J Pers Med, 2020; 10(4):141; doi: 10.3390/jpm10040141

67.

Guleria

, Shah

, Pulido

, et al. Deep learning systems detect dysplasia with human-like accuracy using histopathology and probe-based confocal laser endomicroscopy. Sci Rep, 2021; 11(1):5086; doi: 10.1038/s41598-021-84510-4

68.

Codipilly

, Vogelsang

, Agarwal

, et al. Utilization of a deep learning artificial intelligence model in the histologic diagnosis of dysplasia in Barrett’s esophagus. Gastrointest Endosc, 2021; 93(6):AB288–AB289; doi: 10.1016/j.gie.2021.03.604

69.

Nguyen Wenker

, Natarajan

, Caskey

, et al. Using natural language processing to automatically identify dysplasia in pathology reports for patients with Barrett’s esophagus. Clin Gastroenterol Hepatol, 2023; 21(5):1198–1204; doi: 10.1016/j.cgh.2022.09.005

70.

Tomita

, Abdollahi

, Wei

, et al. Attention-Based deep neural networks for detection of cancerous and precancerous esophagus tissue on histopathological slides. JAMA Netw Open, 2019; 2(11):e1914645; doi: 10.1001/jamanetworkopen.2019.14645

71.

Faghani

, Codipilly

, Vogelsang

, et al. Development of a deep learning model for the histologic diagnosis of dysplasia in Barrett’s esophagus. Gastrointest Endosc, 2022; 96(6):918–925.e3; doi: 10.1016/j.gie.2022.06.013

72.

Agha

, Srinivasan

, Hyder

, et al. WATS3D versus forceps biopsy in screening for Barrett’s esophagus: Experience in community endoscopy centers. Ann Gastroenterol, 2021; 34(2):164–168; doi: 10.20524/aog.2020.0564

73.

Raphael

, Stewart

, Sejpal

, et al. Adjunctive yield of wide-area transepithelial sampling for dysplasia detection after advanced imaging and random biopsies in Barrett’s Esophagus. Clin Transl Gastroenterol, 2019; 10(12):e00107; doi: 10.14309/ctg.0000000000000107

74.

Vennalaganti

, Naag Kanakadandi

, Gross

, et al. Inter-Observer agreement among pathologists using wide-area transepithelial sampling with computer-assisted analysis in patients with Barrett’s esophagus. Am J Gastroenterol, 2015; 110(9):1257–1260; doi: 10.1038/ajg.2015.116

75.

, Zhang

, Zhou

, et al. Randomised controlled trial of WISENSE, a real-time quality improving system for monitoring blind spots during esophagogastroduodenoscopy. Gut, 2019; 68(12):2161–2169; doi: 10.1136/gutjnl-2018-317366

76.

Ali

, Bailey

, Ash

, et al.; TGU Investigators. A pilot study on automatic three-dimensional quantification of Barrett’s esophagus for risk stratification and therapy monitoring. Gastroenterology, 2021; 161(3):865–878.e8; doi: 10.1053/j.gastro.2021.05.059

77.

Hossain

, Abdelrahim

, Tanasescu

, et al. Performance of a novel computer‐aided diagnosis system in the characterization of colorectal polyps, and its role in meeting Preservation and Incorporation of Valuable Endoscopic Innovations standards set by the American society of gastrointestinal endoscopy. DEN Open, 2023; 3(1):e178; doi: 10.1002/deo2.178

78.

Hairdy

. Introducing CADU from Odin Vision. 2022.

79.

Patel

, Arora

, Roknsharifi

, et al. Artificial intelligence in the detection of Barrett’s Esophagus: A systematic review. Cureus, 2023; 15(10):e47755; doi: 10.7759/cureus.47755

80.

Hamade

, Sharma

. Artificial intelligence in Barrett’s Esophagus. Ther Adv Gastrointest Endosc, 2021; doi: 10.1177/26317745211049964

81.

Amann

, Blasimme

, Vayena

, et al.; Precise4Q consortium. Explainability for artificial intelligence in healthcare: A multidisciplinary perspective. BMC Med Inform Decis Mak, 2020; 20(1):310; doi: 10.1186/s12911-020-01332-6

82.

Siau

, Berzin

. Fine‐tuning the artificial intelligence experience in endoscopy. United European Gastroenterol J, 2022; 10(5):449–450; doi: 10.1002/ueg2.12253

83.

Brand

, Troya

, Krenzer

, et al. Development and evaluation of a deep learning model to improve the usability of polyp detection systems during interventions. United European Gastroenterol J, 2022; 10(5):477–484; doi: 10.1002/ueg2.12235

84.

Delpisheh

, Veisani

, Sayehmiri

, et al. Esophageal carcinoma: Long-term survival in consecutive series of patients through a retrospective cohort study. Gastroenterol Hepatol Bed Bench, 2014; 7(2):101–107.

85.

Popovic

, Glisic

, Milosavljevic

, et al. The importance of artificial intelligence in upper gastrointestinal endoscopy. Diagnostics (Basel), 2023; 13(18):2862; doi: 10.3390/diagnostics13182862

86.

Chen

, Wu

, Li

, et al. Comparing blind spots of unsedated ultrafine, sedated, and unsedated conventional gastroscopy with and without artificial intelligence: A prospective, single-blind, 3-parallel-group, randomized, single-center trial. Gastrointest Endosc, 2020; 91(2):332–339.e3; doi: 10.1016/j.gie.2019.09.016

87.

Liu

, Wu

, Yuan

, et al. Artificial intelligence-based assessments of colonoscopic withdrawal technique: A new method for measuring and enhancing the quality of fold examination. Endoscopy, 2022; 54(10):972–979; doi: 10.1055/a-1799-8297

88.

Zhou

, Yao

, Wu

, et al. Multi-step validation of a deep learning-based system for the quantification of bowel preparation: A prospective, observational study. Lancet Digit Health, 2021; 3(11):e697–706–e706; doi: 10.1016/S2589-7500(21)00109-6

89.

Dhaliwal

, Walsh

. Artificial intelligence in pediatric endoscopy: Current status and future applications. Gastrointest Endosc Clin N Am, 2023; 33(2):291–308; doi: 10.1016/j.giec.2022.12.001

90.

Reid

, Blount

, Feng

, et al. Optimizing endoscopic biopsy detection of early cancers in Barrett’s high-grade dysplasia. Am J Gastroenterol, 2000; 95(11):3089–3096; doi: 10.1016/S0002-9270(00)01975-4