Bowel preparation assessment using artificial intelligence: Systematic review.

Abstract

Background and study aims: Insufficient bowel preparation is the leading cause of missed adenomas in colonoscopy. The Boston Bowel Preparation Scale (BBPS) is the most thoroughly validated and widely used scale to estimate risk of missed adenomas. Artificial intelligence (AI) could automatically quantify bowel preparation, thus reducing bias and limitations inherent in human rating. This systematic review aimed to identify, describe, and evaluate all AI-BPS systems for colonoscopy.

Methods: A systematic literature review was conducted using MEDLINE, EMBASE, and SCOPUS based on three sets of terms aligned with the inclusion criteria: colonoscopy, BPS, and AI. Two reviewers independently evaluated and completed data extraction from the articles.

Results: A total of 1,449 studies were identified, with eight meeting the eligibility criteria. Six AI-BPS systems were trained on expert BBPS ratings, and two studies used a fecal-mucosal ratio. All studies compared their AI-BPS with expert BBPS ratings; two showed that their AI-BPS outperformed expert BBPS ratings, and six showed comparable performances. Three studies also demonstrated correlations with adenoma detection rates (ADRs), adenoma miss rates (AMRs), or polyp detection rates (PDRs). Only one prospective study implemented its AI-BPS, finding lower AMR in adequately prepared compared with inadequately prepared bowels.

Conclusions: AI-BPS can standardize and outperform human bowel preparation evaluation by better correlating with expert BBPS ratings, AMR, ADR, and PDR. Further research following recommended reporting guidelines is needed to allow for cross-study comparisons and meta-analysis, which was not possible in this study due to heterogonous study design and reporting metrics.