Artificial intelligence-based measurement outperforms current methods for colorectal polyp size measurement

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate colon polyp size estimation is vital for determining surveillance intervals and predicting malignant progression risk.
• Subjectivity in visual polyp size estimation among endoscopists is a significant clinical challenge.

Purpose of the Study:

• To assess the efficacy of a novel artificial intelligence (AI)-based method for measuring colon polyp size.
• To compare the AI method's accuracy and reliability against current clinical approaches, including visual estimation by endoscopists.

Main Methods:

• Development of a bifurcation-to-bifurcation (BtoB) distance measuring method using the W-Net model, applied to colonoscopy images.
• Comparison of AI-derived measurements with those from eight endoscopists (four experts, four trainees).
• Evaluation of diagnostic ability and reliability using Lin's concordance correlation coefficients (CCCs) and Bland-Altman analyses.

Main Results:

• Visual polyp size estimations showed significant inconsistency among endoscopists, varying with camera view (P < 0.001).
• A trend toward underestimation of polyp sizes, particularly for those >10mm, was observed in both expert and trainee groups.
• The AI technique demonstrated high accuracy and reliability (CCC, 0.961), significantly outperforming visual estimation and biopsy forceps methods.

Conclusions:

• The novel AI measurement method significantly enhances the accuracy and reliability of colon polyp size assessment in colonoscopy images.
• Integration of AI tools is particularly beneficial for improving the efficiency and accuracy of polyp size estimation among trainees.