Accuracy of artificial intelligence‐assisted detection of esophageal cancer and neoplasms on endoscopic images: A systematic review and meta‐analysis

Abstract

To investigate systematically previous studies on the accuracy of artificial intelligence (AI)-assisted diagnostic models in detecting esophageal neoplasms on endoscopic images so as to provide scientific evidence for the effectiveness of these models.Methods: A literature search was conducted on the PubMed, EMBASE and Cochrane Library databases for studies on the AI-assisted detection of esophageal neoplasms on endoscopic images published up to December 2020. A bivariate mixed-effects regression model was used to calculate the pooled diagnostic efficacy of AI-assisted system. Subgroup analyses and meta-regression analyses were performed to explore the sources of heterogeneity. The effectiveness of AI-assisted models was also compared with that of the endoscopists.Results: Sixteen studies were included in the systematic review and meta-analysis.The pooled sensitivity, specificity, positive and negative likelihood ratios, diagnostic odds ratio and area under the summary receiver operating characteristic curve regarding AI-assisted detection of esophageal neoplasms were 94% (95% confidence interval [CI] 92%-96%), 85% (95% CI 73%-92%), 6.40 (95% CI 3.38-12.11), 0.06 (95% CI 0.04-0.10), 98.88 (95% CI 39.45-247.87) and 0.97 (95% CI 0.95-0.98), respectively. AI-based models performed better than endoscopists in terms of the pooled sensitivity (94% [95% CI 84%-98%] vs 82% [95% CI 77%-86%, P < 0.01). Conclusions:The use of AI results in increased accuracy in detecting early esophageal cancer. However, most of the included studies have a retrospective study design, thus further validation with prospective trials is required.

Systematic review with meta‐analysis: artificial intelligence in the diagnosis of oesophageal diseases

“…On the other hand, in this study, the performance of AI applied to real‐time videos was not statistically different from that on still images, and the performance of AI was similar to that of endoscopists. Additionally, a recent meta‐analysis reported that the inclusion of video clips in the training and validation data sets of AI models could achieve even higher performance than those including images alone 66 …”

Efficacy of Digestive Endoscope Based on Artificial Intelligence System in Diagnosing Early Esophageal Carcinoma

“…Zhang S.M. et al showed that the sensitivity of NBI for the diagnosis of esophageal squamous high-grade intraepithelial neoplasia varied greatly, reaching up to 100.0% by experienced endoscopists and only 69.0% by inexperienced physicians [ 20 ], indicating that the examination results of NBI are still influenced by the subjective factors of endoscopists. Because clinical upper DEN needs to be done by endoscopists, and diagnosis relies entirely on the endoscopist's visual interpretation and pathological biopsy, the essence is continuously accumulating experience to enhance accuracy.…”

“…Zhang S.M. et al adopted a great number of samples to construct an AI model to learn 8,428 endoscopic images of patients with esophageal carcinoma, and the results showed that the sensitivity of AI diagnosis was 98.0% and the PPV was 40.0% [ 20 ]. Based on the features of AI learning, the PPV will continuously increase with the number of samples, and with the building of IOT data platform, the accuracy of AI also elevates.…”

Machine Learning for Future Subtyping of the Tumor Microenvironment of Gastro-Esophageal Adenocarcinomas

“…Several studies trained CNNs to aid early detection in GEA, recently summarized in a meta-analysis that found superiority of applying DL for detection of Barrett’s esophagus [ 132 , 133 , 134 , 135 , 136 , 137 ]. Currently, clinical trials are already investigating its sensitivity and specificity, if applied in a clinical setting, with several studies showing DL models to identify early GEA [ 138 , 139 ].…”