Computer-based classification of small colorectal polyps by using narrow-band imaging with optical magnification

Abstract

Background: Recent studies have shown that narrow-band imaging (NBI) is a powerful diagnostic tool for the differentiation between neoplastic and non-neoplastic colorectal polyps.

Objective: To develop a computer-based method for classification of colorectal polyps.

Design: A prospective study.

Setting: University hospital.

Patients: A total of 214 patients with colorectal polyps who underwent a zoom NBI colonoscopy.

Interventions: A total of 434 detected polyps 10 mm or smaller were imaged and subsequently removed for histological analysis.

Main outcome measurements: Diagnostic performance in polyp classification by 2 experts, 2 nonexperts, and a computer-based algorithm.

Results: The expert group and the computer-based algorithm achieved a comparable diagnostic performance (expert group: 93.4% sensitivity, 91.8% specificity, and 92.7% accuracy; computer-based algorithm: 95.0% sensitivity, 90.3% specificity, and 93.1% accuracy) and were both significantly superior to the nonexpert group (86.0% sensitivity, 87.8% specificity, and 86.8% accuracy) in terms of sensitivity, negative predictive value, and accuracy. Subgroup analysis of 255 polyps 5 mm or smaller revealed comparable results without significant differences in the overall analysis of all polyps.

Limitations: No fully automatic classification system.

Conclusions: The study demonstrates that computer-based classification of colon polyps can be achieved with high diagnostic performance.