Skip to content

Diagnostic accuracy of deep learning algorithms based on conventional endoscopic images for screening and discrimination of esophageal lesions

Diagnostic accuracy of deep learning algorithms based on conventional endoscopic images for screening and discrimination of esophageal lesions

Status
Recruiting
Phases
Early Phase 1
Study type
Observational
Source
ChiCTR
Registry ID
ChiCTR1900025883
Enrollment
Unknown
Registered
2019-09-12
Start date
2018-05-01
Completion date
Unknown
Last updated
2019-09-16

For informational purposes only — not medical advice. Sourced from public registries and may not reflect the latest updates. Terms

Conditions

Esophageal lesions

Interventions

Gold Standard:All images of superficial or advanced ESCC were confirmed histologically. Esophageal varices, esophageal scars, reflux esophagitis, normal esophageal mucosa, and parts of the esophageal
Index test:Artificial&#32
intelligence&#32
system

Sponsors

West China Hospital, Sichuan University
Lead Sponsor

Eligibility

Sex/Gender
All
Age
18 Years to 90 Years

Inclusion criteria

Inclusion criteria: From January 2008 to February 2019, we retrospectively obtained esophageal images from West China Hospital of Sichuan University. Pictures including esophagitis, esophageal varices, esophageal submucosal protuberances lesions, superficial esophageal squamous cell carcinoma (ESCC), advanced esophageal squamous cell carcinoma, and esophageal scar. All images of superficial or advanced ESCC were confirmed histologically. Esophageal varices, esophageal scars, reflux esophagitis, normal esophageal mucosa, and parts of the esophageal submucosal protuberances were substantiated by three well-experienced endoscopists with more than 25 years experience; the other parts of the esophageal submucosal protuberances were confirmed by EUS.

Exclusion criteria

Exclusion criteria: We excluded poor quality images resulting from blur, defocus, halation, blood, and mucus. Magnified images, Lugos liquid staining images, and endoscopic ultrasound (EUS) images were also excluded.

Design outcomes

Primary

MeasureTime frame
Sensitivity;Specificity;Diagnostic accuracy;

Secondary

MeasureTime frame
Positive predict value;Negative predict value;Positive likelihood ratio;Negative likelihood ratio;

Countries

China

Contacts

Public ContactYang Jinlin

West China Hospital, Sichuan University

mouse-577@163.com+86 18980602058

Outcome results

None listed

Source: ChiCTR (via WHO ICTRP) · Data processed: Feb 4, 2026