Detection of flat colorectal neoplasia by artificial intelligence: A systematic review

Abstract

Objectives: This study review focuses on a deep learning method for the detection of colorectal lesions in colonoscopy and AI support for detecting colorectal neoplasia, especially in flat lesions.

Data sources: We performed a systematic electric search with PubMed by using "colonoscopy", "artificial intelligence", and "detection". Finally, nine articles about development and validation study and eight clinical trials met the review criteria.

Results: Development and validation studies showed that trained AI models had high accuracy-approximately 90% or more for detecting lesions. Performance was better in elevated lesions than in superficial lesions in the two studies. Among the eight clinical trials, all but one trial showed a significantly high adenoma detection rate in the CADe group than in the control group. Interestingly, the CADe group detected significantly high flat lesions than the control group in the seven studies.

Conclusion: Flat colorectal neoplasia can be detected by endoscopists who use AI.

Keywords: Artificial interagency; Colorectal cancer; Deep learning; Endoscopic submucosal dissection; Endoscopy.