Artificial Intelligence for Leaving in Situ Colorectal Polyps.

This two parallel-arms, randomized, multicenter trial is aimed at investigating safety and effectiveness of a Computer-Aided-Diagnosis (CADx)-assisted leave-in-situ strategy (Leave-In-Situ Arm) as opposed to a resect-all strategy (Standard Arm) as implemented by endoscopists in a real-world setting. With this study it will be possible to understand the impact of CADx in patient treatment and management both in terms of clinical outcome and costs.

Real-time diminutive (≤5 mm) colonic polyp characterization by virtual blue-light (i.e. Narrow Banding Imaging \[NBI\], Blue Light Imaging \[BLI\], etc.) has been identified as a key goal for novel endoscopic advanced imaging techniques. The real-time prediction of polyps histology, especially of those located in the rectosigmoid, is clinically relevant as diminutive polyps represent the majority of polyps detected during colonoscopy and have a very low risk of harboring advanced histology (0.3%) and a negligible risk of invasive carcinoma (0%-0.08%). Thus, as recommended by current guidelines, an optical diagnosis would allow diminutive polyps to be resected and discarded without pathological assessment (i.e. resect-and-discard strategy) or left in place without resection in the case of diminutive rectosigmoid hyperplastic polyps (i.e. leave-in-situ strategy), with an enormous cost-saving potential. In addition, the current policy for managing such hyperplastic-appearing polyps is poorly defined and variable from endoscopist to endoscopist. For instance, the European Society of Gastrointestinal Endoscopy (ESGE) recommends to leave-in-situ only diminutive polyps characterized as hyperplastic with high-confidence. However, it is currently unknown how many endoscopists actually do switch to the blue-light advanced imaging (when available) that is required for a high-confident diagnosis when assessing ≤5 mm hyperplastic-appearing lesions. More importantly, the Leave-in-situ strategy, poorly defined in a pre-AI era, has never been clinically validated, leaving uncertainty on the estimate of the possible false negatives generated by an inaccurate diagnosis6. The availability of Computer-Aided-Diagnosis (CADx) tools, which may help endoscopists distinguish neoplastic from non-neoplastic polyps making the characterization process quicker and more objective, would be highly desirable and captivating. A new system for CADx during white light endoscopy has been developed and integrated alongside a previously available Computer-Aided-Detection (CADe) tool (GI-Genius, Medtronic), making real-time characterization straightforward after polyp detection, and has been recently validated in a real time clinical trial, showing how this system exceeds the American Society of Gastrointestinal Endoscopy (ASGE) Criteria for clinical application of cost saving strategies (i.e. leave in situ and resect and discard), and equals the performance of expert endoscopists in optical diagnosis. However, this was a first study, monocentric and without randomization, with a small, albeit adequately powered, sample size. The need for larger trials is thus urgent to speed up the possible implementation of CADx in clinical practice. Aim of this study is to clinically validate the new CADx tool for the implementation of a leave-in-situ strategy in a multicenter, randomized controlled trial.