Minireviews
Copyright ©The Author(s) 2020. Published by Baishideng Publishing Group Inc. All rights reserved.
World J Gastroenterol. Oct 21, 2020; 26(39): 5959-5969
Published online Oct 21, 2020. doi: 10.3748/wjg.v26.i39.5959
Artificial intelligence technique in detection of early esophageal cancer
Lu-Ming Huang, Wen-Juan Yang, Zhi-Yin Huang, Cheng-Wei Tang, Jing Li
Lu-Ming Huang, Wen-Juan Yang, Zhi-Yin Huang, Cheng-Wei Tang, Jing Li, Department of Gastroenterology, West China Hospital Sichuan University, Chengdu 610041, Sichuan Province, China
ORCID number: Lu-Ming Huang (0000-0002-3620-5767); Wen-Juan Yang (0000-0002-6163-1450); Zhi-Yin Huang (0000-0002-0639-4421); Cheng-Wei Tang (0000-0002-2289-8240); Jing Li (0000-0002-6929-409X).
Author contributions: Huang LM wrote the review; Li J and Tang CW designed and revised the manuscript; Huang LM, Yang WJ, and Huang ZY searched and collected the literature; all authors discussed the statement and conclusions and approved the final version to be published.
Supported by Key Research and Development Program of Science and Technology Department of Sichuan Province, No. 2018GZ0088; Science & Technology Bureau of Chengdu, China, No. 2017-CY02-00023-GX.
Conflict-of-interest statement: There is no conflict of interest associated with any of the senior author or other coauthors who contributed their efforts in this manuscript.
Open-Access: This article is an open-access article that was selected by an in-house editor and fully peer-reviewed by external reviewers. It is distributed in accordance with the Creative Commons Attribution NonCommercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/
Corresponding author: Jing Li, MD, PhD, Associate Professor, Department of Gastroenterology, West China Hospital Sichuan University, No. 37 Guoxue Lane, Chengdu 610041, Sichuan Province, China. melody224@163.com
Received: July 16, 2020
Peer-review started: July 16, 2020
First decision: August 8, 2020
Revised: August 22, 2020
Accepted: September 4, 2020
Article in press: September 4, 2020
Published online: October 21, 2020

Abstract

Due to the rapid progression and poor prognosis of esophageal cancer (EC), the early detection and diagnosis of early EC are of great value for the prognosis improvement of patients. However, the endoscopic detection of early EC, especially Barrett's dysplasia or squamous epithelial dysplasia, is difficult. Therefore, the requirement for more efficient methods of detection and characterization of early EC has led to intensive research in the field of artificial intelligence (AI). Deep learning (DL) has brought about breakthroughs in processing images, videos, and other aspects, whereas convolutional neural networks (CNNs) have shone lights on detection of endoscopic images and videos. Many studies on CNNs in endoscopic analysis of early EC demonstrate excellent performance including sensitivity and specificity and progress gradually from in vitro image analysis for classification to real-time detection of early esophageal neoplasia. When AI technique comes to the pathological diagnosis, borderline lesions that are difficult to determine may become easier than before. In gene diagnosis, due to the lack of tissue specificity of gene diagnostic markers, they can only be used as supplementary measures at present. In predicting the risk of cancer, there is still a lack of prospective clinical research to confirm the accuracy of the risk stratification model.

Key Words: Artificial intelligence, Early esophageal cancer, Barrett's esophagus, Esophageal squamous cell carcinoma, Endoscopic diagnosis, Pathological diagnosis

Core Tip: The requirement for more efficient methods of detection and characterization of early esophageal cancer (EC) has led to intensive research in the field of artificial intelligence (AI). Thus, application of AI technique in endoscopic detection of early EC is reviewed intensively. Furthermore, pathological and gene diagnosis for early EC as well as its risk stratification is also commented.



INTRODUCTION

Esophageal cancer (EC) is the eighth most common cancer and the sixth leading cause of cancer death worldwide[1]. EC mainly consists of esophageal adenocarcinoma (EAC) and esophageal squamous cell carcinoma (ESCC). EAC is the most common pathological type in Western countries, more than 40% of patients with EAC are diagnosed after the disease has metastasized, and the 5-year survival rate is less than 20%[2,3]. Although the incidence of EAC has been increasing globally, ESCC remains the most common pathological type (80%) of all ECs with the highest incidence across a ‘cancer belt’ extending from East Africa and across the Middle East to Asia. Only 20% of patients with ESCC survive longer than 3 years, primarily due to late-stage diagnosis[4]. In low-resource settings, the 5-year survival is much lower at about 3.4%[5]. Early diagnosis may be associated with significantly improved outcomes for all ECs.

Barrett's esophagus (BE) is a premalignant condition characterized by the replacement of normal squamous esophageal epithelium by metaplastic intestinal epithelium containing goblet cells. It is a result of chronic inflammation of the esophagus, conferring a significantly increased risk of EAC. Endoscopic surveillance for BE patients to enable early detection of dysplasia or carcinoma was recommended by GI societies of Western countries[6,7]. The endoscopic surveillance in patients with BE is required with random 4-quadrant biopsy specimens obtained every 1 to 2 cm to detect dysplasia[8]. This method is invasive, time-consuming, and difficult to comply with[9].

Gastroscopy remains the major way to detection of early ESCC. However, endoscopic features of these early lesions are subtle and easily missed with conventional white-light endoscopy (WLE)[10]. Intrapapillary capillary loops (IPCLs) are microvessels, which have been considered a marker of ESCC, because their changes in morphology correlate with the invasion depth of ESCC. Advanced endoscopic-imaging modalities, such as narrow band imaging (NBI), in combination with magnification endoscopy, afford improved visualization of subtle microvascular patterns in the esophageal mucosa of patients with ESCC. Although NBI has showed a high sensitivity for the detection of ESCC, its performance in characterizing these lesions is still limited[11].

Therefore, the requirement for more efficient methods of detection and characterization of early EC has led to intensive research in the field of artificial intelligence (AI), which can be defined by an intelligence established by machines in contrast to the natural intelligence displayed by humans and other animals[12]. Machine learning (ML) and deep learning (DL) are important parts of AI. ML can be divided into supervised and unsupervised methods. Unsupervised learning is to identify groups within data according to commonalities, lack of knowledge of the number of groups or their significance. When the training packet contains input-output pairs, a supervised learning model is required to map new input to output. Conventional ML techniques are limited in their ability to process natural data in their raw form. During the early stage of research and development, the model training was mainly with ML by which researchers have to manually extract the possible disease features based on clinical knowledge. The power of this computer-aided diagnosis (CAD) system is weak and is not enough to be applied in clinical real-time diagnosis.

Convolutional neural networks (CNNs) are supervised ML models inspired by the visual cortex of the human brain processing and recognizing images. Each artificial neuron is a computing unit and all of them are connected to each other, forming a network. By the multiple network layers, CNNs may extract the key features from an image with minimal preprocessing and then provide a final classification through the fully connected layers as the output. The competition of increasing performance has led to a progressive complexity of pooling layers resulting in the concept of DL[13]. The key aspect of DL is that these layers of features are not designed by human engineers. They are learned from data using a general-purpose learning procedure. DL has brought about breakthroughs in processing images, videos, and other aspects, whereas recurrent CNNs have shone lights on detection of endoscopic images and videos.

Application of AI technique in early EC detection has been over 15 years. Many studies on CNNs in endoscopic analysis of early EC demonstrate excellent performance including sensitivity and specificity and progress gradually from in vitro image analysis for classification to real-time detection of early esophageal neoplasia. In this manuscript, we will discuss the following: (1) Utility of AI technique in endoscopic detection of early EC; (2) Role of AI in pathological diagnosis of early EC; (3) AI in gene diagnosis of early EC; (4) AI in risk stratification of early EC; and (5) Conclusion and outlook.

AI IN ENDOSCOPIC DETECTION OF EARLY EC
Barrett's dysplasia and early EAC

AI based on WLE and NBI: There are some limitations to recognize BE-related early neoplastic lesions by WLE, a conventional technology. High-definition WLE (HD-WLE) and NBI were ever considered to enhance the accuracy of the diagnosis of BE-related early neoplastic lesions. But the improvement still not satisfied endoscopists. This situation stimulated development of CAD system for early neoplastic lesions in BE based on supervised ML[14,15]. However, it was still difficult for this system to locate BE-related early neoplastic lesions and to select biopsy sites. To solve those problems, Ebigbo et al[16] established a CAD system based on DL. The accuracy of the system using HD-WLE was better than that of general endoscopists. Moreover, the system displayed the ability to locate the lesion and the area coincidence rate between the lesions delineated by the system and by experts was up to 72%. Generally, the specificity of NBI was higher than that of conventional WLE[17]. Compared with the use of HD-WLE, the system showed no obvious advantage when using NBI[16]. However, there was a common problem in the above studies; that is, the same image dataset was used in both the training stage and validation stage. This obviously cannot reflect clinical practice, so it is very important to use different image datasets for training and validation. de Groof et al[18] used different HD-WLE images for training and testing, and the results showed that the sensitivity and specificity of the system were significantly higher than those of general endoscopists. The combinations of AI and HD-WLE/NBI, which are widely used in the clinic, perform well in the diagnosis of BE-related early tumor lesions and are superior to general endoscopists. However, in different studies, the precision of the CAD system in delineating lesions and the criteria for the evaluation of the ability of lesion location are quite inconsistent. More high-quality studies are needed in the future.

AI based on endoscopic optical coherence tomography and confocal laser endomicroscopy: In addition to WLE and NBI, endoscopic optical coherence tomography (EOCT) and confocal laser endomicroscopy (CLE) are also used to diagnose early EAC/BE-associated dysplasia. EOCT can identify BE-related early tumor lesions by analyzing esophageal mucosal and submucosal structures[19]. CLE can observe the mucosal tissue and cellular morphology to achieve optical biopsy[20]. However, the complexity of these two imaging technologies, the time-consuming reading of images, and the need for senior endoscopists limit their clinical use. To solve this problem, Qi et al[19,21] extracted multiple EOCT image features and combined one or multiple features to classify the lesions, but the results were not satisfactory. After that, Swager et al[22] used volumetric laser endomicroscopy (VLE, integrated with second-generation OCT) images for training and testing, and the results showed that the CAD system was superior to VLE experts. Veronese et al[20] used CLE images, and the results showed that the system could accurately distinguish gastric metaplasia (GM), intestinal metaplasia (IM), and neoplasia. Ghatwary et al[23] showed that the sensitivity of the system using CLE images to diagnose IM and neoplasia was significantly higher than that to diagnose GM. Similarly, the CAD system established by Hong failed to identify GM, the sensitivity of diagnosing IM was not significantly different from that of the above study, and the sensitivity of diagnosing neoplasia was slightly decreased[24]. This may be due to the limited number of images of GM and neoplasia. At present, there are few clinical studies in this field, and the available images are limited. In the future, more research is needed to confirm the value of VLE/CLE combined with AI in the diagnosis of early EAC.

Real-time diagnosis by AI: Currently, research in this field is limited. Ebigbo et al[25] continued to optimize the CAD system based on previous research and applied it to clinical real-time detection for the first time. While 14 patients with Barrett's neoplasia were under endoscopic examination, 62 endoscopic images (36 early EAC and 26 BE without dysplasia) were captured using the CAD system for real-time classification, and the results showed that the sensitivity and specificity were 83.7% and 100%, respectively. There was no significant difference between the system and experienced endoscopists. However, there were still some shortcomings in this study. First, the numbers of patients and images were low. Second, the system still used images for diagnosis, not videos for real-time detection. Finally, the ability of AI to assist in the delineation of lesions and biopsy guidance was not verified. In addition, the CAD system built by Hashimoto et al[17] could meet the needs of clinical real-time detection; unfortunately, the researchers did not verify the performance of the system in real-time diagnosis.

Esophageal squamous dysplasia and early ESCC

AI based on WLE and NBI: Lugol's chromoendoscopy is the standard screening method for ESCC; however, in view of its low specificity and longtime consumption, it is necessary to adopt new endoscopic techniques. Although WLE has been proven to be unsuitable for screening early ESCC alone, considering its clinical popularity, some researchers still hope to introduce AI to improve the accuracy of WLE. Cai et al[26] built a CAD system based on DL and tested it with WLE images. The results showed that the accuracy of the system in the diagnosis of early EC was significantly higher than that of junior and mid-level endoscopists, and there was no significant difference between the system and senior endoscopists. The results of Ohmori et al[27] showed that the sensitivity of NBI was higher than that of WLE, but the specificity was lower, and there was no significant difference between the overall performance of the system and endoscopic experts[27]. The results of Horie et al[28] were similar to those of Ohmori et al[27].

In recent years, it has been found that esophageal intrapapillary capillary loop (IPCL) represents an endoscopically visible feature of esophageal squamous cell neoplasia, and its morphological changes are closely related to the depth of tumor invasion[29]. The classification of IPCL based on NBI proposed by the Japanese Endoscopic Society has been widely used in clinical practice because it is easy to understand[30]. Nevertheless, classifying IPCL still requires sufficient experience, and its interpretation is still subjective. Therefore, to classify the IPCL more objectively and help less experienced endoscopists make full use of NBI, combining NBI with AI will be the best solution. Zhao et al[31] analyzed magnifying NBI images and showed that the diagnostic accuracy of the CAD system was better than that of junior and mid-level endoscopists, and there was no significant difference between the system and senior endoscopists. However, this study focused on the classification of IPCL only and did not further verify the accuracy of the system to determine the depth of tumor invasion. For early ESCC, accurately determining the depth of tumor invasion is the premise of choosing the appropriate treatment. Nakagawa et al[32] used non-magnifying endoscopic images to test the CAD system and found that the sensitivity of the system to diagnose epithelial-submucosal cancers invading up to 200 μm (EP-SM1) was higher than that of endoscopic experts, but the specificity was lower. There was no significant difference between the system and endoscopic experts based on magnifying endoscopic images. Further analysis showed that the system performed well in the diagnosis of EP/lamina propria mucosa (LPM) and had no significant difference from endoscopic experts. However, the sensitivity of the system to diagnose muscularis mucosa (MM)/SM1 was poor, as was the performance of endoscopic experts. In the diagnosis of SM2/3, the sensitivity of this system was slightly higher than that of endoscopic experts. Tokai et al[33] also used non-magnifying endoscopic images for analysis and found that the sensitivity of NBI was slightly higher than that of WLE, and the diagnostic performance of the system was better than that of endoscopic experts. Further analysis showed that the accuracy of the CAD system in diagnosing EP/LPM and MM was more than 90%, which was significantly higher than that in diagnosing SM1 and SM2. The reasons may be that the training image set did not contain normal esophageal images, and the system mistook extramural compression and not fully extended esophageal wall for lesion features.

AI based on endocytoscopy and high-resolution microendoscopy: Endocytoscopy is a new technology that combines magnifying endoscopy with vital staining. Because of its excellent magnifying ability, endoscopists can clearly observe the epithelial cells of the esophageal mucosa to achieve a similar effect to pathological diagnosis[34]. However, if endoscopists want to use endocytoscopy to complete real-time detection independently, they need a solid foundation for pathology, which is obviously not practical. Therefore, AI may be the best option to assist endoscopic diagnosis. Kumagai et al[35] showed that the performance of the system using higher magnification images was better than that of lower magnification images. However, there was no stratified analysis of superficial EC and advanced EC in this study. The results could not accurately reflect the ability of endocytoscopy combined with AI in the diagnosis of early EC. High-resolution microendoscopy (HRME) can be used to observe the esophageal mucosal tissue and cellular morphology. Shin et al[36] tested the ability of different image features to distinguish tumors from non-tumor lesions, and the best features selected had an 84% sensitivity and 95% specificity. However, it took a long time to analyze a single image, and the sensitivity was not ideal. In the future, more high-quality studies are needed to confirm its role in the diagnosis of early EC.

Real-time diagnosis by AI: Quang et al[37] used the same HRME image data as Shin et al[36] for training and validation, and the results showed that the sensitivity was 95% and the specificity was 91%. Then, three patients with suspected EC underwent endoscopic examination with a 100% accuracy. However, there were some limitations in this study. First, the number of patients and images in real-time diagnosis was too small. Second, the system still used images for diagnosis, not videos for real-time detection. Everson et al[38] and Guo et al[39] established CAD systems based on DL with good diagnostic performance using NBI videos. Unfortunately, these systems have not been applied in real-time diagnosis. Endoscopic detection of early EC by various endoscopic techniques assisted by AI is summarized in Table 1.

Table 1 Application of artificial intelligence in endoscopic detection of early esophageal cancer.
Ref.ModalityAI techniqueNo. of images/cases in training dataset (positive/negative)No. of images/cases in test dataset (positive/negative)Results
van der Sommen et al[14], 2016HD-WLESVM100 (60 early BE neoplasia/40 BE)100 (60 early BE neoplasia/40 BE)Sensitivity 83%/specificity 83%
de Groof et al[15], 2019HD-WLESVM60 (40 early BE neoplasia/20 BE)60 (40 early BE neoplasia/20 BE)Sensitivity 95%/specificity 85%
Ebigbo et al[16], 2019HD-WLECNN100 (50 early BE neoplasia/50 BE)100 (50 early BE neoplasia/50 BE)Sensitivity 92%/specificity 100%
Ebigbo et al[16], 2019HD-WLE/NBICNN148 (early BE neoplasia/BE)148 (early BE neoplasia/BE)HD-WLE sensitivity 97%/specificity 88%; NBI sensitivity 94%/specificity 80%
Hashimoto et al[17], 2020WLE/NBICNN1374 (early BE neoplasia/BE)253 WLE (146 early BE neoplasia/107 BE) 205 NBI (79 early BE neoplasia/126 BE)WLE sensitivity 98.6%/specificity 88.8%; NBI sensitivity 92.4%/specificity 99.2%
de Groof et al[18], 2020HD-WLEResNet-UNet1247 WLE + 297 HD-WLE (early BE neoplasia/BE)80 (40 early BE neoplasia/40 BE)Sensitivity 90%/specificity 88%
Swager et al[22], 2017VLESVM60 (30 early BE neoplasia/30 BE)60 (30 early BE neoplasia/30 BE)Sensitivity 90%/specificity 93%
Veronese et al[20], 2013CLESVM337 (23 GM/263 IM/51 neoplasia)337 (23 GM/263 IM/51 neoplasia)Sensitivity 96%/95%/100%
Ghatwary et al[23], 2019CLESVM262 (GM/IM/neoplasia)262 (GM/IM/neoplasia)Sensitivity 70%/93%/93%
Hong et al[24], 2017CLECNN236 (26 GM/155 IM/55 neoplasia)26 (4 GM/17 IM/5 neoplasia)Sensitivity 0%/100%/80%
Ebigbo et al[25], 2020HD-WLECNN129 (early BE neoplasia/BE)62 (36 early BE neoplasia/26 BE)Sensitivity 83.7%/specificity 100%
Cai et al[26], 2019WLECNN2428 (1332 early ESCC/1096 healthy control)187 (91 early ESCC/96 healthy control)Sensitivity 97.8%/specificity 85.4%
Ohmori et al[27], 2020WLE/NBISingle Shot MultiBox Detector22562 (17435 superficial ESCC/5127 control)727 (255 WLE/268 non-magnifying NBI/204 magnifying NBI)WLE sensitivity 90%/specificity 76%; non-magnifying NBI sensitivity 100%/specificity 63%; magnifying NBI sensitivity 98%/specificity 56%
Zhao et al[31], 2019Magnifying NBIDouble-labeling fully convolutional network1383 (207 type A IPCL/970 type B1 IPCL/206 type B2 IPCL)1383 (207 type A IPCL/970 type B1 IPCL/206 type B2 IPCL)Sensitivity 71.5%/91.1%/83.0%
Nakagawa et al[32], 2019WLE/NBICNN8660 non-magnifying (7230 EP-SM1/1430 SM2/3); 5678 magnifying (4916 EP-SM1/762 SM2/3)914 (405 non-magnifying/509 magnifying)Non-magnifying sensitivity 95.4%/specificity 79.2%; magnifying sensitivity 91.6%/specificity 79.2%
Tokai et al[33], 2020WLE/NBICNN1751 superficial ESCC291 (201 EP-SM1/90 SM2)Sensitivity 84.1%/specificity 73.3%
Shin et al[36], 2015HRMETwo-class linear discriminant analysis104 (15 early ESCC/89 control)167 (19 early ESCC/148 control)Sensitivity 84%/specificity 95%
Quang et al[37], 2016HRMETwo-class linear discriminant analysis104 (15 early ESCC/89 control)3 (1 early ESCC/2 control)Sensitivity 100%/specificity 100%
Everson et al[38], 2019magnifying NBICNN7046 sequential images (squamous cell neoplasia/healthy control)7046 sequential images (squamous cell neoplasia/healthy control)Sensitivity 89.7%/specificity 96.9%
Guo et al[39], 2020NBISegNet6473 (early ESCC/control)47 (27 non-magnifying videos/20 magnifying videos)Non-magnifying NBI sensitivity 60.8%; magnifying NBI sensitivity 96.1%
AI IN PATHOLOGICAL DIAGNOSIS OF EARLY EC

Although AI combined with endoscopic diagnosis has made progress, endoscopic diagnosis is still unable to replace the gold standard of pathological diagnosis. However, there is a problem in the pathological diagnosis of early esophageal neoplasia; that is, the accuracy of diagnosing dysplasia is not ideal with considerable interobserver variability. To solve this problem, Sabo et al[40] established two models to distinguish no dysplasia (ND) from low grade dysplasia (LGD) and LGD from high grade dysplasia (HGD) by extracting the hematoxylin and eosin (HE) stained pathological section image features of BE patients. The results showed that the two models performed well in the diagnosis of indistinguishable borderline lesions. Baak et al combined HE stained pathological section image characteristics with p53/Ki67 immunohistochemical indicators and used pathological specimens of surgical resection[41] and endoscopic biopsy samples[42] of BE for testing. It was found that the system performed well in distinguishing ND from LGD and LGD from HGD; however, the accuracy of distinguishing HGD from intramucosal carcinoma needs to be improved. The performance of the system was better than that of the general pathologist and only slightly inferior to that of the experienced pathologist. In addition, there were still some studies using immunohistochemical indicators only[43] or the characteristics of nuclear DNA structure and organization combined with DNA ploidy analysis[44] to grade BE dysplasia, but the results were not satisfactory. Therefore, HE stained pathological image features supplemented by relevant immune indicators or other image enhancement techniques may be the main direction of AI-assisted pathological diagnosis in the future.

AI IN GENE DIAGNOSIS OF EARLY EC

As mentioned above, the invasiveness of endoscopic diagnosis and variability of pathological diagnosis, together with recent advances in the pathogenesis of EC and the development of various omics technologies, have made early EC gene diagnosis a hot topic of research. Zhang et al[45] and Yu et al[46] used microRNAs and long noncoding RNAs specifically expressed in patients with ESCC to establish diagnostic models, and the results showed that both of them can be used to effectively distinguish early from advanced tumors. Xing et al[47] selected characteristic secretory proteins based on the RNA transcriptome data of ESCC patients to establish a diagnostic model with a high sensitivity but unsatisfactory specificity for early ESCC. It may be that the selection of characteristic secretory proteins in this study was based on genes up-regulated in tumor tissues and that the down-regulated genes were not fully utilized. In addition, Shen et al[48] and Zhai et al[49] used microRNA and protein markers specifically expressed in the plasma of ESCC patients to build diagnostic models and found that both of them could effectively distinguish esophageal squamous dysplasia (ESD) from healthy controls. However, compared with ESD or early ESCC, the AI-assisted gene diagnosis of BE-related dysplasia and early EAC is less studied. The results of Slaby et al[50] showed that characteristic microRNAs in BE tissues could be used to distinguish BE with or without dysplasia. These markers may not be specific for EC; therefore, gene diagnosis can only be used as an assistant measure for endoscopic and pathological diagnosis until specific markers of EC gene diagnosis are established.

AI IN RISK STRATIFICATION OF EARLY EC

Considering that only 0.12%-0.43% of BE patients may progress to EAC every year[51-53], it is particularly necessary to establish an effective model to predict the risk of EAC in BE patients. Previous risk stratification was mainly based on the presence of dysplasia, but its effect was not ideal. Critchley-Thorne et al[54] established a predictive model based on the characteristic differences of tissue immunofluorescence markers and histopathological images between patients with BE who developed EAC and those who did not. The results were not satisfactory, with more than 30% of BE patients who developed EAC classified as low-risk. Li et al[55] established a predictive model based on the differences in single nucleotide polymorphisms in the biopsy tissues of BE patients with good performance for predicting EC. Unlike EAC, ESCC has no recognized precancerous disease, and endoscopic screening for the general population is obviously impractical. Therefore, it will be the focus of clinical screening to establish an effective model to predict which individuals are more likely to develop ESD. Etemadi et al[56]'s model based on the epidemiological data of patients with ESD and healthy controls showed that the performance of the model was not good. Moghtadaei et al[57]'s predictive model based on epidemiological data was better than Etemadi's. Unfortunately, there is a lack of prospective follow-up studies to verify the true accuracy of these models.

CONCLUSION

In conclusion, AI is trying to be used in the endoscopic detection, pathological diagnosis, gene diagnosis, and cancer risk prediction of early EC. It is helpful for endoscopists and pathologists to improve the accuracy of diagnosis and to assist clinicians for treatment and making follow-up strategies. The borderline lesions of EC pathology that remain difficult to determine may be the main direction of AI-assisted pathological diagnosis in the future. Gene diagnosis can only be used as an assistant measure for endoscopic and pathological diagnosis until specific markers of EC gene diagnosis are established. Higher precision of the CAD system in delineating lesions is necessary for improvement of the accuracy of diagnosis, and enhancement of accuracy for the risk stratification model may be of benefit for the prediction of EC risk.

Footnotes

Manuscript source: Invited manuscript

Specialty type: Gastroenterology and hepatology

Country/Territory of origin: China

Peer-review report’s scientific quality classification

Grade A (Excellent): 0

Grade B (Very good): B

Grade C (Good): C

Grade D (Fair): D

Grade E (Poor): 0

P-Reviewer: Cabezuelo AS, Que J S-Editor: Yan JP L-Editor: Wang TQ P-Editor: Li JH

References
1.  Bray F, Ferlay J, Soerjomataram I, Siegel RL, Torre LA, Jemal A. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2018;68:394-424.  [PubMed]  [DOI]
2.  Hur C, Miller M, Kong CY, Dowling EC, Nattinger KJ, Dunn M, Feuer EJ. Trends in esophageal adenocarcinoma incidence and mortality. Cancer. 2013;119:1149-1158.  [PubMed]  [DOI]
3.  Thrift AP. The epidemic of oesophageal carcinoma: Where are we now? Cancer Epidemiol. 2016;41:88-95.  [PubMed]  [DOI]
4.  Ferlay J, Colombet M, Soerjomataram I, Mathers C, Parkin DM, Piñeros M, Znaor A, Bray F. Estimating the global cancer incidence and mortality in 2018: GLOBOCAN sources and methods. Int J Cancer. 2019;144:1941-1953.  [PubMed]  [DOI]
5.  Aghcheli K, Marjani HA, Nasrollahzadeh D, Islami F, Shakeri R, Sotoudeh M, Abedi-Ardekani B, Ghavamnasiri MR, Razaei E, Khalilipour E, Mohtashami S, Makhdoomi Y, Rajabzadeh R, Merat S, Sotoudehmanesh R, Semnani S, Malekzadeh R. Prognostic factors for esophageal squamous cell carcinoma--a population-based study in Golestan Province, Iran, a high incidence area. PLoS One. 2011;6:e22152.  [PubMed]  [DOI]
6.  Shaheen NJ, Falk GW, Iyer PG, Gerson LB; American College of Gastroenterology. ACG Clinical Guideline: Diagnosis and Management of Barrett's Esophagus. Am J Gastroenterol. 2016;111:30-50; quiz 51.  [PubMed]  [DOI]
7.  Fitzgerald RC, di Pietro M, Ragunath K, Ang Y, Kang JY, Watson P, Trudgill N, Patel P, Kaye PV, Sanders S, O'Donovan M, Bird-Lieberman E, Bhandari P, Jankowski JA, Attwood S, Parsons SL, Loft D, Lagergren J, Moayyedi P, Lyratzopoulos G, de Caestecker J; British Society of Gastroenterology. British Society of Gastroenterology guidelines on the diagnosis and management of Barrett's oesophagus. Gut. 2014;63:7-42.  [PubMed]  [DOI]
8.  Sharma P, Savides TJ, Canto MI, Corley DA, Falk GW, Goldblum JR, Wang KK, Wallace MB, Wolfsen HC; ASGE Technology and Standards of Practice Committee. The American Society for Gastrointestinal Endoscopy PIVI (Preservation and Incorporation of Valuable Endoscopic Innovations) on imaging in Barrett's Esophagus. Gastrointest Endosc. 2012;76:252-254.  [PubMed]  [DOI]
9.  Tavakkoli A, Appelman HD, Beer DG, Madiyal C, Khodadost M, Nofz K, Metko V, Elta G, Wang T, Rubenstein JH. Use of Appropriate Surveillance for Patients With Nondysplastic Barrett's Esophagus. Clin Gastroenterol Hepatol. 2018;16:862-869.e3.  [PubMed]  [DOI]
10.  Menon S, Trudgill N. How commonly is upper gastrointestinal cancer missed at endoscopy? A meta-analysis. Endosc Int Open. 2014;2:E46-E50.  [PubMed]  [DOI]
11.  Nagami Y, Tominaga K, Machida H, Nakatani M, Kameda N, Sugimori S, Okazaki H, Tanigawa T, Yamagami H, Kubo N, Shiba M, Watanabe K, Watanabe T, Iguchi H, Fujiwara Y, Ohira M, Hirakawa K, Arakawa T. Usefulness of non-magnifying narrow-band imaging in screening of early esophageal squamous cell carcinoma: a prospective comparative study using propensity score matching. Am J Gastroenterol. 2014;109:845-854.  [PubMed]  [DOI]
12.  Le Berre C, Sandborn WJ, Aridhi S, Devignes MD, Fournier L, Smaïl-Tabbone M, Danese S, Peyrin-Biroulet L. Application of Artificial Intelligence to Gastroenterology and Hepatology. Gastroenterology. 2020;158:76-94.e2.  [PubMed]  [DOI]
13.  LeCun Y, Bengio Y, Hinton G. Deep learning. Nature. 2015;521:436-444.  [PubMed]  [DOI]
14.  van der Sommen F, Zinger S, Curvers WL, Bisschops R, Pech O, Weusten BL, Bergman JJ, de With PH, Schoon EJ. Computer-aided detection of early neoplastic lesions in Barrett's esophagus. Endoscopy. 2016;48:617-624.  [PubMed]  [DOI]
15.  de Groof J, van der Sommen F, van der Putten J, Struyvenberg MR, Zinger S, Curvers WL, Pech O, Meining A, Neuhaus H, Bisschops R, Schoon EJ, de With PH, Bergman JJ. 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;7:538-547.  [PubMed]  [DOI]
16.  Ebigbo A, Mendel R, Probst A, Manzeneder J, Souza LA, Papa JP, Palm C, Messmann H. Computer-aided diagnosis using deep learning in the evaluation of early oesophageal adenocarcinoma. Gut. 2019;68:1143-1145.  [PubMed]  [DOI]
17.  Hashimoto R, Requa J, Dao T, Ninh A, Tran E, Mai D, Lugo M, El-Hage Chehade N, Chang KJ, Karnes WE, Samarasena JB. 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.e1.  [PubMed]  [DOI]
18.  de Groof AJ, Struyvenberg MR, van der Putten J, van der Sommen F, Fockens KN, Curvers WL, Zinger S, Pouw RE, Coron E, Baldaque-Silva F, Pech O, Weusten B, Meining A, Neuhaus H, Bisschops R, Dent J, Schoon EJ, de With PH, Bergman JJ. 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;158:915-929.e4.  [PubMed]  [DOI]
19.  Qi X, Sivak MV, Isenberg G, Willis JE, Rollins AM. Computer-aided diagnosis of dysplasia in Barrett's esophagus using endoscopic optical coherence tomography. J Biomed Opt. 2006;11:044010.  [PubMed]  [DOI]
20.  Veronese E, Grisan E, Diamantis G, Battaglia G, Crosta C, Trovato C. Hybrid patch-based and image-wide classification of confocal laser endomicroscopy images in Barrett’s esophagus surveillance.  IEEE International Symposium on Biomedical Imaging 2013.  [PubMed]  [DOI]
21.  Qi X, Pan Y, Sivak MV, Willis JE, Isenberg G, Rollins AM. Image analysis for classification of dysplasia in Barrett's esophagus using endoscopic optical coherence tomography. Biomed Opt Express. 2010;1:825-847.  [PubMed]  [DOI]
22.  Swager AF, van der Sommen F, Klomp SR, Zinger S, Meijer SL, Schoon EJ, Bergman JJGHM, de With PH, Curvers WL. Computer-aided detection of early Barrett's neoplasia using volumetric laser endomicroscopy. Gastrointest Endosc. 2017;86:839-846.  [PubMed]  [DOI]
23.  Ghatwary N, Ahmed A, Grisan E, Jalab H, Bidaut L, Ye X. In-vivo Barrett's esophagus digital pathology stage classification through feature enhancement of confocal laser endomicroscopy. J Med Imaging (Bellingham). 2019;6:014502.  [PubMed]  [DOI]
24.  Jisu Hong, Bo-Yong Park, Hyunjin Park. Convolutional neural network classifier for distinguishing Barrett's esophagus and neoplasia endomicroscopy images. Conf Proc IEEE Eng Med Biol Soc. 2017;2017:2892-2895.  [PubMed]  [DOI]
25.  Ebigbo A, Mendel R, Probst A, Manzeneder J, Prinz F, de Souza LA, Papa J, Palm C, Messmann H. Real-time use of artificial intelligence in the evaluation of cancer in Barrett's oesophagus. Gut. 2020;69:615-616.  [PubMed]  [DOI]
26.  Cai SL, Li B, Tan WM, Niu XJ, Yu HH, Yao LQ, Zhou PH, Yan B, Zhong YS. Using a deep learning system in endoscopy for screening of early esophageal squamous cell carcinoma (with video). Gastrointest Endosc. 2019;90:745-753.e2.  [PubMed]  [DOI]
27.  Ohmori M, Ishihara R, Aoyama K, Nakagawa K, Iwagami H, Matsuura N, Shichijo S, Yamamoto K, Nagaike K, Nakahara M, Inoue T, Aoi K, Okada H, Tada T. Endoscopic detection and differentiation of esophageal lesions using a deep neural network. Gastrointest Endosc. 2020;91:301-309.e1.  [PubMed]  [DOI]
28.  Horie Y, Yoshio T, Aoyama K, Yoshimizu S, Horiuchi Y, Ishiyama A, Hirasawa T, Tsuchida T, Ozawa T, Ishihara S, Kumagai Y, Fujishiro M, Maetani I, Fujisaki J, Tada T. Diagnostic outcomes of esophageal cancer by artificial intelligence using convolutional neural networks. Gastrointest Endosc. 2019;89:25-32.  [PubMed]  [DOI]
29.  Sato H, Inoue H, Ikeda H, Sato C, Onimaru M, Hayee B, Phlanusi C, Santi EG, Kobayashi Y, Kudo SE. Utility of intrapapillary capillary loops seen on magnifying narrow-band imaging in estimating invasive depth of esophageal squamous cell carcinoma. Endoscopy. 2015;47:122-128.  [PubMed]  [DOI]
30.  Oyama T, Inoue H, Arima M, Momma K, Omori T, Ishihara R, Hirasawa D, Takeuchi M, Tomori A, Goda K. Prediction of the invasion depth of superficial squamous cell carcinoma based on microvessel morphology: magnifying endoscopic classification of the Japan Esophageal Society. Esophagus. 2017;14:105-112.  [PubMed]  [DOI]
31.  Zhao YY, Xue DX, Wang YL, Zhang R, Sun B, Cai YP, Feng H, Cai Y, Xu JM. Computer-assisted diagnosis of early esophageal squamous cell carcinoma using narrow-band imaging magnifying endoscopy. Endoscopy. 2019;51:333-341.  [PubMed]  [DOI]
32.  Nakagawa K, Ishihara R, Aoyama K, Ohmori M, Nakahira H, Matsuura N, Shichijo S, Nishida T, Yamada T, Yamaguchi S, Ogiyama H, Egawa S, Kishida O, Tada T. Classification for invasion depth of esophageal squamous cell carcinoma using a deep neural network compared with experienced endoscopists. Gastrointest Endosc. 2019;90:407-414.  [PubMed]  [DOI]
33.  Tokai Y, Yoshio T, Aoyama K, Horie Y, Yoshimizu S, Horiuchi Y, Ishiyama A, Tsuchida T, Hirasawa T, Sakakibara Y, Yamada T, Yamaguchi S, Fujisaki J, Tada T. Application of artificial intelligence using convolutional neural networks in determining the invasion depth of esophageal squamous cell carcinoma. Esophagus. 2020;17:250-256.  [PubMed]  [DOI]
34.  Kumagai Y, Monma K, Kawada K. Magnifying chromoendoscopy of the esophagus: in-vivo pathological diagnosis using an endocytoscopy system. Endoscopy. 2004;36:590-594.  [PubMed]  [DOI]
35.  Kumagai Y, Takubo K, Kawada K, Aoyama K, Endo Y, Ozawa T, Hirasawa T, Yoshio T, Ishihara S, Fujishiro M, Tamaru JI, Mochiki E, Ishida H, Tada T. Diagnosis using deep-learning artificial intelligence based on the endocytoscopic observation of the esophagus. Esophagus. 2019;16:180-187.  [PubMed]  [DOI]
36.  Shin D, Protano MA, Polydorides AD, Dawsey SM, Pierce MC, Kim MK, Schwarz RA, Quang T, Parikh N, Bhutani MS, Zhang F, Wang G, Xue L, Wang X, Xu H, Anandasabapathy S, Richards-Kortum RR. Quantitative analysis of high-resolution microendoscopic images for diagnosis of esophageal squamous cell carcinoma. Clin Gastroenterol Hepatol. 2015;13:272-279.e2.  [PubMed]  [DOI]
37.  Quang T, Schwarz RA, Dawsey SM, Tan MC, Patel K, Yu X, Wang G, Zhang F, Xu H, Anandasabapathy S, Richards-Kortum R. A tablet-interfaced high-resolution microendoscope with automated image interpretation for real-time evaluation of esophageal squamous cell neoplasia. Gastrointest Endosc. 2016;84:834-841.  [PubMed]  [DOI]
38.  Everson M, Herrera L, Li W, Luengo IM, Ahmad O, Banks M, Magee C, Alzoubaidi D, Hsu HM, Graham D, Vercauteren T, Lovat L, Ourselin S, Kashin S, Wang HP, Wang WL, Haidry RJ. Artificial intelligence for the real-time classification of intrapapillary capillary loop patterns in the endoscopic diagnosis of early oesophageal squamous cell carcinoma: A proof-of-concept study. United European Gastroenterol J. 2019;7:297-306.  [PubMed]  [DOI]
39.  Guo L, Xiao X, Wu C, Zeng X, Zhang Y, Du J, Bai S, Xie J, Zhang Z, Li Y, Wang X, Cheung O, Sharma M, Liu J, Hu B. Real-time automated diagnosis of precancerous lesions and early esophageal squamous cell carcinoma using a deep learning model (with videos). Gastrointest Endosc. 2020;91:41-51.  [PubMed]  [DOI]
40.  Sabo E, Beck AH, Montgomery EA, Bhattacharya B, Meitner P, Wang JY, Resnick MB. Computerized morphometry as an aid in determining the grade of dysplasia and progression to adenocarcinoma in Barrett's esophagus. Lab Invest. 2006;86:1261-1271.  [PubMed]  [DOI]
41.  Polkowski W, Baak JP, van Lanschot JJ, Meijer GA, Schuurmans LT, Ten Kate FJ, Obertop H, Offerhaus GJ. Clinical decision making in Barrett's oesophagus can be supported by computerized immunoquantitation and morphometry of features associated with proliferation and differentiation. J Pathol. 1998;184:161-168.  [PubMed]  [DOI]
42.  van Sandick JW, Baak JP, van Lanschot JJ, Polkowski W, ten Kate FJ, Obertop H, Offerhaus GJ. Computerized quantitative pathology for the grading of dysplasia in surveillance biopsies of Barrett's oesophagus. J Pathol. 2000;190:177-183.  [PubMed]  [DOI]
43.  Karamchandani DM, Lehman HL, Ohanessian SE, Massé J, Welsh PA, Odze RD, Goldblum JR, Berg AS, Stairs DB. Increasing diagnostic accuracy to grade dysplasia in Barrett's esophagus using an immunohistochemical panel for CDX2, p120ctn, c-Myc and Jagged1. Diagn Pathol. 2016;11:23.  [PubMed]  [DOI]
44.  Dunn JM, Hveem T, Pretorius M, Oukrif D, Nielsen B, Albregtsen F, Lovat LB, Novelli MR, Danielsen HE. Comparison of nuclear texture analysis and image cytometric DNA analysis for the assessment of dysplasia in Barrett's oesophagus. Br J Cancer. 2011;105:1218-1223.  [PubMed]  [DOI]
45.  Zhang C, Wang C, Chen X, Yang C, Li K, Wang J, Dai J, Hu Z, Zhou X, Chen L, Zhang Y, Li Y, Qiu H, Xing J, Liang Z, Ren B, Yang C, Zen K, Zhang CY. Expression profile of microRNAs in serum: a fingerprint for esophageal squamous cell carcinoma. Clin Chem. 2010;56:1871-1879.  [PubMed]  [DOI]
46.  Yu J, Wu X, Huang K, Zhu M, Zhang X, Zhang Y, Chen S, Xu X, Zhang Q. Bioinformatics identification of lncRNA biomarkers associated with the progression of esophageal squamous cell carcinoma. Mol Med Rep. 2019;19:5309-5320.  [PubMed]  [DOI]
47.  Xing S, Zheng X, Wei LQ, Song SJ, Liu D, Xue N, Liu XM, Wu MT, Zhong Q, Huang CM, Zeng MS, Liu WL. Development and Validation of a Serum Biomarker Panel for the Detection of Esophageal Squamous Cell Carcinoma through RNA Transcriptome Sequencing. J Cancer. 2017;8:2346-2355.  [PubMed]  [DOI]
48.  Shen Y, Ding Y, Ma Q, Zhao L, Guo X, Shao Y, Niu C, He Y, Zhang F, Zheng D, Wei W, Liu F. Identification of Novel Circulating miRNA Biomarkers for the Diagnosis of Esophageal Squamous Cell Carcinoma and Squamous Dysplasia. Cancer Epidemiol Biomarkers Prev. 2019;28:1212-1220.  [PubMed]  [DOI]
49.  Zhai XH, Yu JK, Lin C, Wang LD, Zheng S. Combining proteomics, serum biomarkers and bioinformatics to discriminate between esophageal squamous cell carcinoma and pre-cancerous lesion. J Zhejiang Univ Sci B. 2012;13:964-971.  [PubMed]  [DOI]
50.  Slaby O, Srovnal J, Radova L, Gregar J, Juracek J, Luzna P, Svoboda M, Hajduch M, Ehrmann J. Dynamic changes in microRNA expression profiles reflect progression of Barrett's esophagus to esophageal adenocarcinoma. Carcinogenesis. 2015;36:521-527.  [PubMed]  [DOI]
51.  Bhat S, Coleman HG, Yousef F, Johnston BT, McManus DT, Gavin AT, Murray LJ. Risk of malignant progression in Barrett's esophagus patients: results from a large population-based study. J Natl Cancer Inst. 2011;103:1049-1057.  [PubMed]  [DOI]
52.  Hvid-Jensen F, Pedersen L, Drewes AM, Sørensen HT, Funch-Jensen P. Incidence of adenocarcinoma among patients with Barrett's esophagus. N Engl J Med. 2011;365:1375-1383.  [PubMed]  [DOI]
53.  de Jonge PJ, van Blankenstein M, Looman CW, Casparie MK, Meijer GA, Kuipers EJ. Risk of malignant progression in patients with Barrett's oesophagus: a Dutch nationwide cohort study. Gut. 2010;59:1030-1036.  [PubMed]  [DOI]
54.  Critchley-Thorne RJ, Duits LC, Prichard JW, Davison JM, Jobe BA, Campbell BB, Zhang Y, Repa KA, Reese LM, Li J, Diehl DL, Jhala NC, Ginsberg G, DeMarshall M, Foxwell T, Zaidi AH, Lansing Taylor D, Rustgi AK, Bergman JJ, Falk GW. A Tissue Systems Pathology Assay for High-Risk Barrett's Esophagus. Cancer Epidemiol Biomarkers Prev. 2016;25:958-968.  [PubMed]  [DOI]
55.  Li X, Paulson TG, Galipeau PC, Sanchez CA, Liu K, Kuhner MK, Maley CC, Self SG, Vaughan TL, Reid BJ, Blount PL. Assessment of Esophageal Adenocarcinoma Risk Using Somatic Chromosome Alterations in Longitudinal Samples in Barrett's Esophagus. Cancer Prev Res (Phila). 2015;8:845-856.  [PubMed]  [DOI]
56.  Etemadi A, Abnet CC, Golozar A, Malekzadeh R, Dawsey SM. Modeling the risk of esophageal squamous cell carcinoma and squamous dysplasia in a high risk area in Iran. Arch Iran Med. 2012;15:18-21.  [PubMed]  [DOI]
57.  Moghtadaei M, Hashemi Golpayegani MR, Malekzadeh R. A variable structure fuzzy neural network model of squamous dysplasia and esophageal squamous cell carcinoma based on a global chaotic optimization algorithm. J Theor Biol. 2013;318:164-172.  [PubMed]  [DOI]