Artificial intelligence for detecting superficial esophageal squamous cell carcinoma under multiple endoscopic...

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Diagnosis of esophageal squamous cell carcinoma (ESCC) is complex and requires significant expertise.
• Superficial ESCC detection can be challenging under various endoscopic imaging modalities.

Purpose of the Study:

• To develop an artificial intelligence (AI) system for detecting superficial ESCC.
• To evaluate the AI system's performance across multiple endoscopic imaging modalities.

Main Methods:

• Retrospective collection of 53,933 endoscopic images and 142 videos from multiple hospitals for training and validation.
• Comparison of the AI system's performance against 11 experienced endoscopists.
• Validation of the AI system on internal and external datasets, including video analysis.

Main Results:

• The AI system demonstrated high sensitivity (92.5-99.7%) and specificity (78.5-89.0%) across validation datasets.
• AI performance was comparable to experienced endoscopists, with superior sensitivity for epithelial-confined ESCC on white-light imaging (90.8% vs 82.5%).
• The AI system performed well on video analysis, showing sensitivity of 89.5-100% and specificity of 73.7-89.5%.

Conclusions:

• An AI system was successfully developed for detecting superficial ESCC.
• The AI system demonstrates comparable efficacy to experienced endoscopists across multiple imaging types.
• This AI tool has the potential to support endoscopists, particularly those with less experience, in ESCC detection.