Endoscopy serves as the principal technique for identifying asymptomatic esophageal squamous cell carcinoma (ESCC) and precancerous lesions. Detecting early-stage esophageal cancers, which respond better to treatment, remains a significant challenge due to their subtle presentation. Enhancing the detection rates of such early stages is crucial. Now, a new study has demonstrated that integrating a deep learning system into routine endoscopy can significantly improve the detection of early-stage esophageal cancers.

The large-scale randomized controlled trial (RCT), conducted by researchers at Taizhou Hospital (Zhejiang, China), evaluated the effectiveness of a deep learning–based system named ENDOANGEL-ELD for detecting esophageal cancer. The results published in Science Translational Medicine reveal that this AI system nearly doubled the detection capability of clinicians in identifying high-risk esophageal lesions, including both cancerous and precancerous conditions, compared to traditional unassisted endoscopy.

In the trial, 3,117 patients were randomly assigned to undergo either AI-assisted or standard endoscopy. The findings indicated a significant improvement in detection rates of high-risk esophageal lesions when using the AI system, with detection rates of 1.8% compared to 0.9% in the unassisted group. The ENDOANGEL-ELD system exhibited high sensitivity (89.7%), specificity (98.5%), and overall accuracy (98.2%), and was noted for its safety with no adverse events reported. These results underscore the potential of AI to enhance the early diagnosis and treatment of esophageal cancer, which could improve patient outcomes significantly. 

http://www.gzjiayumed.com/en/index.asp .