Computer-aided diagnosis improves characterization of Barrett's neoplasia by general endoscopists (with video).

BACKGROUND AND AIMS: Characterization of visible abnormalities in patients with Barrett's esophagus (BE) can be challenging, especially for inexperienced endoscopists. This results in suboptimal diagnostic accuracy and poor interobserver agreement. Computer-aided diagnosis (CADx) systems may assist endoscopists. We aimed to develop, validate, and benchmark a CADx system for BE neoplasia. METHODS: The CADx system received pretraining with ImageNet and then consecutive domain-specific pretraining with GastroNet, which includes 5 million endoscopic images. It was subsequently trained and internally validated using 1758 narrow-band imaging (NBI) images of early BE neoplasia (352 patients) and 1838 NBI images of nondysplastic BE (173 patients) from 8 international centers. CADx was tested prospectively on corresponding image and video test sets with 30 cases (20 patients) of BE neoplasia and 60 cases (31 patients) of nondysplastic BE. The test set was benchmarked by 44 general endoscopists in 2 phases (phase 1, no CADx assistance; phase 2, with CADx assistance). Ten international BE experts provided additional benchmark performance. RESULTS: Stand-alone sensitivity and specificity of the CADx system were 100% and 98% for images and 93% and 96% for videos, respectively. CADx outperformed general endoscopists without CADx assistance in terms of sensitivity (P = .04). Sensitivity and specificity of general endoscopists increased from 84% to 96% and 90% to 98% with CAD assistance (P < .001). CADx assistance increased endoscopists' confidence in characterization (P < .001). CADx performance was similar to that of the BE experts. CONCLUSIONS: CADx assistance significantly increased characterization performance of BE neoplasia by general endoscopists to the level of expert endoscopists. The use of this CADx system may thereby improve daily Barrett surveillance.

Vertical tumor-positive resection margins and the risk of residual neoplasia after endoscopic resection of Barrett's neoplasia: a nationwide cohort with pathology reassessment.

BACKGROUND: This study evaluated the proportion of patients with residual neoplasia after endoscopic resection (ER) for Barrett's neoplasia with confirmed tumor-positive vertical resection margin (R1v). METHODS: This retrospective cohort study included patients undergoing ER for Barrett's neoplasia with histologically documented R1v since 2008 in the Dutch Barrett Expert Centers. We defined R1v as cancer cells touching vertical resection margins and Rx as nonassessable margins. Reassessment of R1v specimens was performed by experienced pathologists until consensus was reached regarding vertical margins. RESULTS: 101/110 included patients had macroscopically complete resections (17 T1a, 84 T1b), and 99/101 (98%) ER specimens were histologically reassessed, with R1v confirmed in 74 patients (75%), Rx in 16%, and R0 in 9%. Presence/absence of residual neoplasia could be assessed in 66/74 patients during endoscopic reassessment (52) and/or in the surgical resection specimen (14), and 33/66 (50%) had residual neoplasia. Residual neoplasia detected during endoscopy was always endoscopically visible and biopsies from a normal-appearing ER scar did not detect additional neoplasia. Of 25 patients who underwent endoscopic follow-up (median 37 months [interquartile range 12-50]), 4 developed local recurrence (16.0%), all detected as visible abnormalities. CONCLUSIONS: After ER with R1v, 50% of patients had no residual neoplasia. Histological evaluation of ER margins appears challenging, as in this study 75% of documented R1v cases were confirmed during reassessment. Endoscopic reassessment 8-12 weeks after ER seems to accurately detect residual neoplasia and can help to determine the most appropriate strategy for patients with R1v.

Dysplastic Recurrence After Successful Treatment for Early Barrett's Neoplasia: Development and Validation of a Prediction Model.

BACKGROUND & AIMS: The combination of endoscopic resection and radiofrequency ablation is the treatment of choice for eradication of Barrett's esophagus (BE) with dysplasia and/or early cancer. Currently, there are no evidence-based recommendations on how to survey patients after successful treatment, and most patients undergo frequent follow-up endoscopies. We aimed to develop and externally validate a prediction model for visible dysplastic recurrence, which can be used to personalize surveillance after treatment. METHODS: We collected data from the Dutch Barrett Expert Center Registry, a nationwide registry that captures outcomes from all patients with BE undergoing endoscopic treatment in the Netherlands in a centralized care setting. We used predictors related to demographics, severity of reflux, histologic status at baseline, and treatment characteristics. We built a Fine and Gray survival model with least absolute shrinkage and selection operator penalization to predict the incidence of visible dysplastic recurrence after initial successful treatment. The model was validated externally in patients with BE treated in Switzerland and Belgium. RESULTS: A total of 1154 patients with complete BE eradication were included for model building. During a mean endoscopic follow-up of 4 years, 38 patients developed recurrent disease (1.0%/person-year). The following characteristics were independently associated with recurrence (strongest to weakest predictor): a new visible lesion during treatment phase, higher number of endoscopic resection treatments, male sex, increasing BE length, high-grade dysplasia or cancer at baseline, and younger age. External validation showed a C-statistic of 0.91 (95% confidence interval, 0.86-0.94) with good calibration. CONCLUSIONS: This is the first externally validated model to predict visible dysplastic recurrence after successful endoscopic eradication treatment of BE with dysplasia or early cancer. On external validation, our model has good discrimination and calibration. This model can help clinicians and patients to determine a personalized follow-up strategy.

Development and External Validation of a Model to Predict Complex Treatment After Radiofrequency Ablation for Barrett's Esophagus With Early Neoplasia.

BACKGROUND & AIMS: Endoscopic eradication therapy for Barrett's esophagus (BE)-related neoplasia is safe and leads to complete eradication in the majority of patients. However, a subgroup will experience a more complex treatment course with a risk for failure or disease progression. Early identification of these patients may improve patient counseling and treatment outcomes. We aimed to develop a prognostic model for a complex treatment course. METHODS: We collected data from a nationwide registry that captures outcomes for all patients undergoing endoscopic eradication therapy for early BE neoplasia. A complex treatment course was defined as neoplastic progression, treatment failure, or the need for endoscopic resection during the radiofrequency ablation treatment phase. We developed a prognostic model using logistic regression. We externally validated our model in an independent registry. RESULTS: A total of 1386 patients were included, of whom 78 (6%) had a complex treatment course. Our model identified patients with a BE length of 9 cm or longer with a visible lesion containing high-grade dysplasia/cancer, and patients with less than 50% squamous conversion after radiofrequency ablation were identified as high risk for a complex treatment. This applied to 8% of the study population and included 93% of all treatment failures and 76% of all patients with advanced neoplastic progression. The model appeared robust in multiple sensitivity analyses and performed well in external validation (area under the curve, 0.84). CONCLUSIONS: We developed a prognostic model that identified patients with a BE length of 9 cm or longer and high-grade dysplasia/esophageal adenocarcinoma and those with poor squamous regeneration as high risk for a complex treatment course. The good performance in external validation suggests that it may be used in clinical management (Netherlands Trial Register: NL7039).

Pedunculated Morphology of T1 Colorectal Tumors Associates With Reduced Risk of Adverse Outcome.

BACKGROUND & AIMS: Risk stratification for adverse events, such as metastasis to lymph nodes, is based only on histologic features of tumors. We aimed to compare adverse outcomes of pedunculated vs nonpedunculated T1 colorectal cancers (CRC). METHODS: We performed a retrospective study of 1656 patients diagnosed with T1CRC from 2000 through 2014 at 14 hospitals in The Netherlands. The median follow-up time of patients was 42.5 months (interquartile range, 18.5-77.5 mo). We evaluated the association between tumor morphology and the primary composite end point, adverse outcome, adjusted for clinical variables, histologic variables, resection margins, and treatment approach. Adverse outcome was defined as metastasis to lymph nodes, distant metastases, local recurrence, or residual tissue. Secondary end points were tumor metastasis, recurrence, and incomplete resection. RESULTS: Adverse outcome occurred in 67 of 723 patients (9.3%) with pedunculated T1CRCs vs 155 of 933 patients (16.6%) with nonpedunculated T1CRCs. Pedunculated morphology was independently associated with decreased risk of adverse outcome (adjusted odds ratio [OR], 0.59; 95% CI, 0.42-0.83; P = .003). Metastasis, incomplete resection, and recurrence were observed in 5.8%, 4.6%, and 3.9% of pedunculated T1CRCs vs 10.6%, 8.0%, and 6.6% of nonpedunculated T1CRCs, respectively. Pedunculated morphology was independently associated with a reduced risk of metastasis (adjusted OR, 0.62; 95% CI, 0.41-0.94; P = .03), incomplete resection (adjusted OR, 0.57; 95% CI, 0.36-0.91; P = .02), and recurrence (adjusted hazard ratio, 0.52; 95% CI, 0.32-0.85; P = .009). Metastasis, incomplete resection, and recurrence did not differ significantly between low-risk pedunculated vs nonpedunculated T1CRCs (0.8% vs 2.9%, P = .38; 1.5% vs 0%, P = .99; 1.5% vs 0%; P = .99). However, incomplete resection and recurrence were significantly lower for high-risk pedunculated vs nonpedunculated T1CRCs (6.5% vs 12.5%; P = .007; 4.4% vs 8.6%; P = .03). CONCLUSIONS: In a retrospective study of patients with T1CRC, we found pedunculated morphology to be associated independently with a decreased risk of adverse outcome in a T1CRC population at high risk of adverse outcome. Incorporating morphologic features of tumors in risk assessment could help predict outcomes of patients with T1CRC and help identify the best candidates for surgery.