A Machine-Learning Approach to Predicting Adenoma Detection and Surveillance Impact of Deeper Sections in Colorectal
Archives of Pathology & Laboratory Medicine
|May 6, 2026
Summary
Routine pathology practice of examining deeper levels of colorectal polyps lacks standardization. Machine learning models can predict adenoma detection and optimize surveillance intervals, improving diagnostic accuracy and resource use.
Area of Science:
- Gastroenterology and Pathology
- Machine Learning in Medicine
- Colorectal Cancer Screening
Background:
- Deeper level sectioning of colorectal polypectomy specimens is common but lacks standardized protocols.
- Limited data exist on the predictability of adenoma detection and the impact of deeper levels on surveillance and resource utilization.
Purpose of the Study:
- To develop machine learning models for predicting clinically significant lesions in deeper polyp levels.
- To identify factors influencing patient surveillance intervals based on deeper level examination outcomes.
Main Methods:
- Retrospective analysis of 94,888 patients and 145,405 colonoscopies (2007-2024).
- Development of machine learning models to predict adenoma detection in deeper levels and impact on surveillance intervals.
- Analysis of polyp characteristics, deeper-level requests, and clinical/endoscopic features.
Main Results:
- Deeper levels were requested in 11% of jars, revealing adenomas in 51% of examined polyps.
- Machine learning models demonstrated high performance (AUC 0.88-0.90, accuracy 81-83%) in predicting outcomes.
- Key predictors included polyp size, location, total polyps, and AI-assisted colonoscopy.
Conclusions:
- Deeper-level sectioning significantly aids adenoma detection and impacts surveillance intervals.
- Machine learning offers a practical approach to optimize deeper-level requests.
- This enhances diagnostic precision and resource allocation in colorectal polyp management.
