A clinically interpretable convolutional neural network for the real-time prediction of early squamous cell cancer of the esophagus: comparing diagnostic performance with a panel of expert European and Asian endoscopists.

BACKGROUND AND AIMS: Intrapapillary capillary loops (IPCLs) are microvascular structures that correlate with the invasion depth of early squamous cell neoplasia and allow accurate prediction of histology. Artificial intelligence may improve human recognition of IPCL patterns and prediction of histology to allow prompt access to endoscopic therapy for early squamous cell neoplasia where appropriate. METHODS: One hundred fifteen patients were recruited at 2 academic Taiwanese hospitals. Magnification endoscopy narrow-band imaging videos of squamous mucosa were labeled as dysplastic or normal according to their histology, and IPCL patterns were classified by consensus of 3 experienced clinicians. A convolutional neural network (CNN) was trained to classify IPCLs, using 67,742 high-quality magnification endoscopy narrow-band images by 5-fold cross validation. Performance measures were calculated to give an average F1 score, accuracy, sensitivity, and specificity. A panel of 5 Asian and 4 European experts predicted the histology of a random selection of 158 images using the Japanese Endoscopic Society IPCL classification; accuracy, sensitivity, specificity, positive and negative predictive values were calculated. RESULTS: Expert European Union (EU) and Asian endoscopists attained F1 scores (a measure of binary classification accuracy) of 97.0% and 98%, respectively. Sensitivity and accuracy of the EU and Asian clinicians were 97%, 98% and 96.9%, 97.1%, respectively. The CNN average F1 score was 94%, sensitivity 93.7%, and accuracy 91.7%. Our CNN operates at video rate and generates class activation maps that can be used to visually validate CNN predictions. CONCLUSIONS: We report a clinically interpretable CNN developed to predict histology based on IPCL patterns, in real time, using the largest reported dataset of images for this purpose. Our CNN achieved diagnostic performance comparable with an expert panel of endoscopists.

Risk factors for serious adverse events associated with multiband mucosectomy in Barrett's esophagus: an international multicenter analysis of 3827 endoscopic resection procedures.

BACKGROUND AND AIMS: Multiband mucosectomy (MBM) is a widely used technique for the treatment of Barrett's esophagus (BE). However, large multicenter studies enabling a generalizable estimation of the risk of serious adverse events, such as perforation and postprocedural bleeding, are lacking. The aim of this study was to estimate the rate of, and risk factors for, serious adverse events associated with MBM. METHODS: In this retrospective analysis, consecutive patients who underwent MBM for treatment of BE in 14 tertiary referral centers in Europe, the United States, Canada, and Australia were included. Primary outcomes were perforation and postprocedural bleeding rate. Potential risk factors were identified by logistic regression. RESULTS: Between 2001 and 2016, a total of 3827 MBM procedures were performed in 2447 patients (84% male, mean age 66 years, median BE length C2M4). Perforation occurred in 17 procedures (0.4%; 95% confidence interval [CI], 0.3-0.7), of which 15 could be treated endoscopically or conservatively. Female gender was an independent risk factor for perforation (odds ratio [OR], 2.77; 95% CI, 1.02-7.57; P = .05). Postprocedural bleeding occurred after 35 procedures (0.9%; 95% CI, 0.6-1.3). The number of resections (OR, 1.15; 95% CI, 1.06-1.25; P < .001) was significantly associated with postprocedural bleeding. CONCLUSION: The results of this study show that MBM for BE is safe with a low risk of serious adverse events. In addition, most of the adverse events could be managed endoscopically or conservatively. The number of resections was an independent risk factor for postprocedural bleeding.

Virtual chromoendoscopy by using optical enhancement improves the detection of Barrett's esophagus-associated neoplasia.

BACKGROUND AND AIMS: The Seattle protocol for endoscopic Barrett's esophagus (BE) surveillance samples a small portion of the mucosal surface area, risking a potentially high miss rate of early neoplastic lesions. We assessed whether the new iScan Optical Enhancement system (OE) improves the detection of early BE-associated neoplasia compared with high-definition white-light endoscopy (HD-WLE) in both expert and trainee endoscopists to target sampling of suspicious areas. Such a system may both improve early neoplasia detection and reduce the need for random biopsies. METHODS: A total of 41 patients undergoing endoscopic BE surveillance from January 2016 to November 2017 were recruited from 3 international referral centers. Matched still images in both HD-WLE (n = 130) and iScan OE (n = 132) were obtained from endoscopic examinations. Two experts, unblinded to the videos and histology, delineated known neoplasia, forming a consensus criterion standard. Seven expert and 7 trainee endoscopists marked 1 position per image where they would expect a target biopsy to identify dysplastic tissue. The same expert panel then reviewed magnification images and, using a previously validated classification system, attempted to classify mucosa as dysplastic or nondysplastic, based on the mucosal and vascular (MV) patterns observed on magnification endoscopy. Diagnostic accuracy, sensitivity, specificity, negative predictive value (NPV), and positive predictive value (PPV) were calculated. Improvements in dysplasia detection in HD-WLE versus OE and interobserver agreement were assessed by multilevel logistic regression analysis and Krippendorff alpha, respectively. Improvements in diagnostic performance were expressed as an odds ratio between the odds of improvement in OE compared with the odds of improvement in HD-WLE. RESULTS: Accuracy of neoplasia detection was significantly higher in all trainees who used OE versus HD-WLE (76% vs 63%) and in 6 experts (84% vs 77%). OE improved sensitivity of dysplasia detection compared with HD-WLE in 6 trainees (81% vs 71%) and 5 experts (77% vs 67%). Specificity improved in 6 trainees who used OE versus HD-WLE (70% vs 55%) and in 5 experts (92% vs 86%). PPV improved in both an expert and trainee cohort, but NPV improved significantly only in trainees. By using the MV classification and OE magnification endoscopy compared with HD-WLE, we demonstrated improvements in accuracy (79.9% vs 66.7%), sensitivity (86.3% vs 83.4%), and specificity (71.2% vs 53.6%) of dysplasia detection. PPV improved (62%-76.6%), as did NPV (67.7%-78.5%). Interobserver agreement also improved by using OE from 0.30 to 0.55. CONCLUSION: iScan OE may improve dysplasia detection on endoscopic imaging of BE as well as the accuracy of histology prediction compared with HD-WLE, when OE magnification endoscopy is used in conjunction with a simple classification system by both expert and non-expert endoscopists.