Endoscopy 2021; 53(S 01): S9-S10
DOI: 10.1055/s-0041-1724272
Abstracts | ESGE Days
ESGE Days 2021 Oral presentations
Thursday, 25 March 2021 11:00 – 11:45 AI in the esophagus: A clinical challenge Room 6

A Systematic Review and Meta-Analysis on Endoscopists’ Accuracy for Detecting Upper Gastrointestinal Neoplasia in Artificial Intelligence Studies: What is the Gold-Standard?

L Frazzoni
1   University of Bologna, Department of Medical and Surgical Sciences, Bologna, Italy
,
J Arribas
2   University of Porto, CIDES/CINTESIS, Faculty of Medicine, Porto, Portugal
,
G Antonelli
3   Nuovo Regina Margherita Hospital, Digestive Endoscopy Unit, Rome, Italy
4   Sapienza University of Rome, Department of Translational and Precision Medicine, Rome, Italy
,
C Hassan
3   Nuovo Regina Margherita Hospital, Digestive Endoscopy Unit, Rome, Italy
,
L Fuccio
1   University of Bologna, Department of Medical and Surgical Sciences, Bologna, Italy
,
M Dinis-Ribeiro
2   University of Porto, CIDES/CINTESIS, Faculty of Medicine, Porto, Portugal
› Author Affiliations
› Further Information
Also available at
 

Aims Estimates on miss rates for upper gastrointestinal neoplasia (UGIN), i.e. approximately 10 % mainly relies on registry data or old studies. Significant technological breakthroughs and quality parameters have been issued thereafter. We aimed at assessing endoscopists’ accuracy for detecting UGIN within artificial intelligence (AI) studies.

Methods Broad literature search among databases (PubMed/MEDLINE, EMBASE, Scopus) up to July 2020 was performed to identify full articles evaluating endoscopists’ diagnostic accuracy for UGIN compared to AI systems, based on “ex-vivo” images with confirmed histology. Main outcomes were endoscopists’ pooled sensitivity, specificity, positive (PPV) and negative (NPV) predictive values for UGIN. We computed pooled proportion rates, summary receiving operating characteristic (SROC) curves with area under the curves (AUCs), and subgroup analyses.

Results 8 studies published from 2016 to 2020, encompassing 148 endoscopists and 5,439 images were included for quantitative synthesis; 3 studies assessed oesophageal squamous cell neoplasia (ESCN), 5 BERN and 2 GAC. Endoscopists’ diagnostic accuracy was significantly higher for GAC detection (AUC 0.95, CI 0.93-0.98) than ESCN detection (AUC 0.90, CI 0.88-0.92) than BERN detection (AUC 0.86, CI0.84-0.88). Overall, the false negative rate was 18 % with sensitivity 82 % (CI 80-84 %) and NPV 85 % (CI 83-87 %) for UGIN detection, at the pooled 44 % (CI 42-45 %) prevalence of UGIN. Sensitivity was significantly higher for Asian vs. European endoscopists (87 %, CI 84-89 % vs. 75 %, CI 72-78 %), and for experienced vs. inexpert endoscopists (85 %, CI 83-87 % vs. 71 %, CI 67-75 %). Study quality was high with some risk of selection and spectrum bias. Endoscopists’ performance might have been hampered by the use of still images. No significant publication bias was found.

Conclusions We confirm that endoscopists’ accuracy for detecting UGIN is suboptimal. Technological aids, such as AI, might be warranted for inexpert and European endoscopists.

Citation: Frazzoni L, Arribas J, Antonelli G et al. OP12 A SYSTEMATIC REVIEW AND META-ANALYSIS ON ENDOSCOPISTS’ ACCURACY FOR DETECTING UPPER GASTROINTESTINAL NEOPLASIA IN ARTIFICIAL INTELLIGENCE STUDIES: WHAT IS THE GOLD-STANDARD?. Endoscopy 2021; 53: S9.



Publication History

Article published online:
19 March 2021

© 2021. European Society of Gastrointestinal Endoscopy. All rights reserved.

Georg Thieme Verlag KG
Rüdigerstraße 14, 70469 Stuttgart, Germany