Endoscopy-Based Deep Learning Algorithms vs Endoscopists in Early Esophageal Squamous Cell Carcinoma Detection: A...

Area of Science:

• Gastroenterology and Hepatology
• Artificial Intelligence in Medicine
• Oncology Diagnostics

Background:

• Early detection of esophageal squamous cell carcinoma (ESCC) is crucial for improving patient outcomes.
• Endoscopy is a primary tool for ESCC screening, but diagnostic accuracy varies with endoscopist experience.
• Deep learning (DL) algorithms are emerging as potential aids in endoscopic diagnostics.

Purpose of the Study:

• To compare the diagnostic performance of endoscopy-based deep learning (DL) algorithms against endoscopists of varying experience levels for early ESCC detection.
• To evaluate the impact of DL assistance on the diagnostic performance of junior and senior endoscopists.

Main Methods:

• A systematic literature search was conducted across PubMed, Embase, and Web of Science up to March 2026.
• Eligible studies included those evaluating endoscopy-based DL models for early ESCC detection compared to pathological biopsy.
• Pooled sensitivity, specificity, area under the curve (AUC), and diagnostic odds ratio (DOR) were calculated using a bivariate random-effects model.

Main Results:

• DL algorithms demonstrated superior overall performance compared to all endoscopists in sensitivity (0.94 vs. 0.82), specificity (0.88 vs. 0.78), and AUC (0.96 vs. 0.87).
• DL significantly outperformed junior endoscopists and improved their diagnostic accuracy (AUC 0.94 vs. 0.85).
• DL assistance also enhanced the performance of senior endoscopists, though not always significantly, and showed higher accuracy with white-light imaging compared to image-enhanced endoscopy.

Conclusions:

• Endoscopy-based DL algorithms exhibit high accuracy for early ESCC detection and outperform endoscopists, particularly less experienced ones.
• DL assistance can enhance diagnostic performance for both junior and senior endoscopists.
• Further prospective, multicenter studies are needed to validate these findings across diverse populations.