Identification of Barrett's neoplasia: Beyond Seattle protocol

Barrett's esophagus (BE), a complication of gastroesophageal reflux disease (GERD), is derived from prolonged gastric acid or bile exposure to the esophagus.^{1, 2} This refluxate may lead to erosions and chronic inflammatory cells infiltration in the esophageal mucosa. Prolonged damage in healthy squamous epithelium in esophagus would promote its replacement with intestinal metaplasia containing goblet cells, that is, BE. BE is considered as a premalignant lesion for esophageal adenocarcinoma (EAC).² EAC is the predominant form of esophageal cancer in Western countries with progressively increased incidence.³ BE can have neoplastic transformation from low-grade dysplasia (LGD), high-grade dysplasia (HGD), to EAC.² The prognosis of advanced esophageal cancer is poor, while the survival is excellent if detected at early stages for early intervention.^{3, 4} Therefore, it is extremely important to detect dysplasia or early EAC during surveillance endoscopy in BE patients.

Seattle protocol is recommended for endoscopic surveillance in BE patients. It is suggested to perform random biopsies at four quadrants every 1 to 2 cm of the Barrett's segment for detecting subtle dysplasia.⁵ Modern enhanced imaging technologies have been developed to improve dysplasia detection beyond the traditional high definition—white light endoscopy (HD-WLE). Update guidelines from the American Society for Gastrointestinal Endoscopy (ASGE) suggest using dye-based or virtual chromoendoscopy to detect target lesion for biopsies identified in Barrett's segment.⁵ In a systemic review involving 14 studies with over 800 patients, chromoendoscopy would show a 34% increase in yield in detecting dysplasia or cancer compared with WLE, irrespective of dye-based or virtual.⁶ Nevertheless, ASGE did not recommend chromoendoscopy as a replacement for the Seattle protocol but rather as an adjunct technique.⁵

Several dyes, including acetic acid, methylene blue, and indigo carmine, are the dyes commonly used to detect Barrett's dysplasia in surveillance. Acetic acid is the only dye-based chromoendoscopy that fulfill the ASGE preservation and incorporation of valuable innovations (PIVI) thresholds (sensitivity 96.6%, negative predictive value 98.3%, specificity 84.6%).^{5, 7} However, the dye application in BE surveillance is hampered by increased cost for special dye spraying equipment, dye preparation, increased procedure time, potential risk of including DNA damage, and difficulty in adequate dye application evenly.⁵ Because of these limitations, virtual chromoendoscopy may be the preferred, advanced imaging technique for BE surveillance.

Virtual chromoendoscopy applied light filters, emitting light with a short wavelength or postprocessing techniques to enhance the detection of Barrett's neoplasia. There were three major endoscopic platforms of virtual chromoendoscopy, that is, narrow band imaging (NBI, Olympus), blue light imaging (BLI, Fujinon), and i-Scan (Pentax Medical). Virtual chromoendoscopy bears the benefits of being risk-free for patients, ease to perform by button-pressing, and no extra cost because of pre-equipment of this technique in most endoscopes. Compared with the WLE with light wavelengths of 400 to 700 nm, NBI uses shorter wavelengths (400-540 nm) by filtering to enhance the surface mucosa and vascular pattern.⁸ BLI applies two different lasers as light source (410 and 450 nm) to provide brighter and high-resolution endoscopic images of gastrointestinal lesions.⁹ A recent study showed that experts using BLI were able to improve their performance in delineating neoplastic lesions compared with white light endoscopy.⁹ And i-Scan use proprietary post-image acquisition processing technology to modify the white light image, enhancing the superficial mucosal and vascular patterns.⁵ Application of NBI is demonstrated to increase the detection of dysplasia and reduce the number of biopsies.¹⁰ Recently, the Barrett's International NBI Group (BING) has developed and validated an NBI classification system in patients with BE.¹¹ By using NBI imaging to evaluate the mucosal pattern and the vascular pattern as either regular or irregular, the system shows an accuracy more than 90% and substantial level of interobserver agreement (κ = 0.681). In the issue of Adv Dig Med, Chen et al. used BLI, instead of NBI, to validate the BING classification in five medical centers in Taiwan.¹² A total of 12 endoscopists (six more experienced and six less experienced) participated in the evaluation program composed of pretest, educational, and post-test. The test sets contained 80 endoscopic images from non-dysplastic, LGD to HGD Barrett lesions. The overall accuracy is 0.73 before and after training (more experienced: 0.74 to 0.77; less experienced: 0.72 to 0.69). The accuracies in both groups did not change significantly after training. The overall interobserver agreement (κ value) improved from 0.4419 to 0.5573 after training (P < .0001), with more prominent in the less experienced group (more experienced: 0.5471 to 0.5837; less experienced: 0.3625 to 0.5499, P < .0001 in both groups). The authors concluded that the diagnosis of Barrett's dysplasia by using BING classification with BLI assessment showed good accuracy and moderate interobserver agreement. Actually, such accuracy and κ value were inferior to those results from the original reports showing 85% of accuracy and 0.68 of κ value. The original BING classification was mainly used to detect HGD and EAC, while the current study included mainly LGD and few HGD. And it is reported that the surface patterns are similar in LGD and BE without dysplasia.¹¹ It can be difficult to detect the surface changes of LGD by NBI or BLI, leading to the low accuracy and κ value. Furthermore, the enrollment of less experienced gastroenterologists may also contribute to the discrepancy. Since Taiwan is a low prevalence of EAC,¹³ practicing endoscopists may not be familiar with the identification of Barrett's dysplasia, which may be partly responsible for the results. Another interesting result in this study was the failure to show improvement in accuracy after education module training. Unfamiliarity of identifying Barrett's dysplasia may also contribute to this result and including more dysplasia photos for training may reverse the unfavorable effects. Actually, an international group has recently generated a new BLI for Barrett's neoplasia classification (BLINC) based on the color, pit, and vessel pattern by using the BLI technique. It revealed good results with both high accuracy (95.2%) and κ value (0.83) in 10 experts.¹⁴ However, among 15 non-expert endoscopists after a web-based training of BLINC classification, it showed an insignificant improvement of accuracy (86.8%-88.3%, P = .42) and κ value (0.60-0.67, P = .20). In comparison with BING classification, BLINC added “color” or “focal darkness” as a parameter to distinguish neoplasia in BE. BLI uses light-emitting diodes to directly emit a blue light without involving a narrow-band filter or digital postprocessing technology, which facilitates the contrast-enhancing properties and brightness of the blue light image. This character of BLI may help to recognize the dysplastic area in BE and make it possible to include “color” as a category in BLINC classification. However, “darkness” is a subjective description and it remains questionable regarding “How dark is dark?.” Furthermore, BLINC classification added subclassification of distribution and density in the mucosal pit and vessel pattern, which is also subjective, to analyze BE to distinguish between neoplastic and non-neoplastic mucosa. This may make BLINC difficult to make improvement after training. Future applications of artificial intelligence (AI) can be promising for solving the issues to help endoscopists identify the neoplastic lesion in BE upon chromoendoscopy.¹⁵

In conclusion, given the poor prognosis of advanced esophageal cancer, it is important to improve our ability to detect early EAC and BE dysplasia. The current ASGE guideline suggests high-quality endoscopic examination with Seattle protocol as the standard of care for BE surveillance. Chen's study further demonstrated that a simplified BING classification to detect Barrett's dysplasia by application of BLI can be useful in a low EAC incidence region, such as Taiwan. Future studies are mandatory to evaluate the role of AI in optimizing the detection of dysplasia in BE patients.

The authors declare no conflict of interest.