EndoClean: A Hybrid Deep Learning Framework for Automated Full-Video Boston Bowel Preparation Scale Assessment
Year: 2026
Type: article
Abstract: Background and Aims: Adequate bowel preparation is the cornerstone of high-quality colonoscopy. The Boston Bowel Preparation Scale (BBPS) is the gold standard for assessment, yet its application suffers from inter-observer variability and lacks a fully automated solution for entire video analysis. This study proposes EndoClean, a novel, fully automated deep learning framework designed to compute the full-segment BBPS score from colonoscopy videos, aiming to provide a standardized, objective, and near expert-level assessment. Methods: EndoClean integrates three distinct models: frame selection, anatomical segmentation, and BBPS scoring. Its performance was rigorously evaluated against a reference standard established by senior experts and compared with junior endoscopists. We assessed assessment precision, inter-rater agreement (quadratic weighted Kappa), and consistency across all colonic segments. Results: The EndoClean system demonstrated superior reliability, achieving a global accuracy of 97.8% for the total BBPS score, with satisfying agreement with senior experts (κ = 0.984; 95% CI: 0.976–0.989). Notably, EndoClean performed significantly better than junior endoscopists in overall BBPS agreements (κ: 0.984 vs. 0.949, p < 0.001) and overall accuracy (97.8% vs. 94.6%, p = 0.037). In segment-specific analysis, the EndoClean surpassed junior doctors particularly in the transverse colon (Accuracy: 97.5% vs. 90.4%, p < 0.001) and effectively reduced misclassifications in clinically ambiguous intermediate scores. For binary adequacy classification, the system achieved a sensitivity of 98.2% and a specificity of 97.3%. Conclusions: EndoClean represents a robust solution in automated quality control, demonstrating performance comparable to senior experts in bowel preparation assessment. By significantly reducing the variability seen in junior endoscopists and providing objective, full-video BBPS scoring, this framework offers a viable, standardized, and efficient solution for clinical practice and large-scale quality monitoring.
Source: Bioengineering
Authors: Yan Zhu, Si-Yuan Li, Pei-Yao Fu, Zhen Zhang, Shuo Wang, Quan-Lin Li, Ping-Hong Zhou
Language: English
Subfield: Oncology
Field: Medicine
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $3,258
Awards: G1860018917, G5224098412
OpenAlex ID: W7133202063
