Characterization of Colorectal Lesions Using a Computer-Aided Diagnostic System for Narrow-Band Imaging Endocytoscopy

Abstract

Scan the quick response (QR) code to the left with your mobile device to watch this article's video abstract and others. Don't have a QR code reader? Get one by searching 'QR Scanner' in your mobile device's app store. R ecently, the American Society of GastrointestinalEndoscopy established the Preservation and Incorporation of Valuable Endoscopic Innovations 1 for diminutive colorectal polyps. Preservation and Incorporation of Valuable Endoscopic Innovations suggests that, if an endoscopist diagnoses an agreement of >90% in determining postpolypectomy surveillance intervals and a negative predictive value of >90% with adenomatous histology, pathologic diagnosis might not be necessary. Although magnifying chromoendoscopy, 2 narrow-band imaging (NBI), 3 endocytoscopy (EC), 4 and confocal laser endomicroscopy 5 are highly accurate, interpretation of these modalities is difficult for novices. Furthermore, achieving a negative predictive value of >90% for adenoma is not easy using these modalities 3 and requires comprehensive experiments. 6 To achieve a breakthrough on this issue, we developed a computer-aided diagnosis (CAD) system for EC. This system automatically provides highly accurate diagnosis as expert endoscopists concurrently take EC images (Video Clip 1). 7 Our previous system, based on glandular structural and cellular atypia, required endoscopists to use dye for staining. In contrast, the endocytoscopic vascular pattern can effectively evaluate microvessel findings using EC with NBI (EC-NBI) without using any dye. We reported that EC-NBI has a highly accurate diagnostic ability, similar to other modalities. 8 Because dye staining complicates the procedure, our CAD system for EC-NBI represents a powerful tool for novices and experts who do not use dyes on a routine basis. Therefore, we developed a tentative CAD system model for EC-NBI image. Description of TechnologyWe developed custom software (EndoBRAIN, Cybernet Systems Co., Ltd., Tokyo, Japan) to analyze EC images. We collected a consecutive series of 1079 EC-NBI images (431 nonneoplasms and 648 neoplasms) from 85 lesions to form an image database. To validate the CAD system, we randomly extracted 100 images (50 nonneoplasms, 50 neoplasms) from the database. The remaining 979 images (381 nonneoplasms, 598 neoplasms) were used for machine learning. The inclusion criteria were colorectal lesions that had been detected during colonoscopy using EC and subsequently resected between December 2014 and April 2015. The exclusion criteria were inflammatory bowel disease; lesions for which no clear EC-NBI were available; sessile serrated adenomas/polyps (SSA/Ps); and nonepithelial Figure 1. Output image. (1) Computer diagnosis. (2) Input endocytoscopy with narrow band imaging. (3) Extracted vessel image, in which the green area denotes the extracted vessels. The light-green vessel has the maximum diameter. (4) Probability of computer diagnosis is calculated by the support vector machine classifier.Abbreviations used in this paper: CAD, computer-aided diagnosis; ...

Computer-Aided Diagnosis Based on Convolutional Neural Network System for Colorectal Polyp Classification: Preliminary Experience

“…Thus, a wide variety of CAD systems are available at present. Importantly, the diagnostic accuracy of these novel methods is higher than that achieved by endoscopic diagnosis by less-experienced endoscopists [9][10][11] .…”

“…These real-time CAD systems are operated during the actual endoscopic observation. Moreover, Japanese researchers have successfully developed an automated CAD using endocytoscopy based on automated extraction of ultramagnified nuclear features followed by machine-learning analysis from the same group [9][10][11] . Thus, a wide variety of CAD systems are available at present.…”

“…Previous studies have reported the usefulness of CAD for colon polyps. CAD for colon polyps has been suggested as a particularly useful tool for trainee colonoscopists, as the use of a CAD system helps avoid the complications associated with endoscopic resections [9][10][11] . Moreover, the higher rate of correct diagnoses using a CAD system also suggests the potential for this diagnostic tool to improve the quality of colonoscopic examinations.…”

Evaluation of microvascular findings of deeply invasive colorectal cancer by endocytoscopy with narrow-band imaging

“…However, EC-NBI does not require staining, and testing with EC-NBI therefore can be performed safely and quickly. We recently reported on use of a computer-aided diagnosis (CAD) system for EC-V 18. The CAD system provided sufficient diagnostic ability for lesion characterization, but it could not predict the invasion depth of colorectal cancer.…”

Potential of artificial intelligence‐assisted colonoscopy using an endocytoscope (with video)

“…The predicted histology of the target polyps is output as either neoplastic or non‐neoplastic with a probability (Video S1). The algorithm of this AI system is based on automatic image feature extraction followed by machine learning.…”