Can a Polyp Detection and Characterization System Predict Complete Resection?

Introduction: Artificial Intelligence (AI) is one of the most evolving fields in endoscopy. We aimed to test if a system for polyp detection and polyp characterization can be used to predict complete endoscopic resection of colon adenomas. Methods: We used the CAD-Eye AI system (Fujifilm Europe) in consecutive patients who received polypectomy using a cold snare. After resection, the submucosal space was flushed with water using an irrigation pump. Images were obtained using the CAD Eye system, and the characterization of the system was noted and afterward compared to histology of the removed specimen. Results: In total, 17 polypectomies were observed, and in no case the AI was able to give information about resection status. First, the resection plane itself was classified as being adenomatous in all cases, while, second, all adenomas were resected completely, thus harboring no potential for overlying misinterpretations in the images. Conclusion: An AI system trained to characterize polyps in healthy surrounding colorectal mucosa cannot predict the state of resection after removal of the adenoma. This is explained by the training and programming. Endoscopists using AI from now on should learn about the basics of AI and the pitfalls in interpreting results from AI.

Automated summary

The study showed that an AI system trained to characterize polyps in healthy surrounding colorectal mucosa cannot predict the state of resection after removal of the adenoma. This limitation is due to the training and programming of the system. Endoscopists using AI should learn about the basics of AI and the potential pitfalls in interpreting AI results.