Feasibility of Depth-in-Color En Face Optical Coherence Tomography for Colorectal Polyp Classification Using Ensemble Learning and Score-Level Fusion

Abstract

In colorectal cancer screening, accurate detection of precursor lesions is challenging due to their ambiguous surface appearance. Depth-sensitive optical coherence tomography (OCT) with deep learning may improve accuracy. OCT imaging was performed on polyps from 300 subjects. Depth was encoded (surface, mid, deep) in color to generate en face OCT projections. En face projections were then annotated. The projections were then used to train an ensemble network based on the malignant potential of polyps. The area under the curve (AUC) for detecting malignant potential of all polyps was 0.90, and for diminutive polyps (≤ 5 mm), it was 0.88. These results demonstrate a high degree of accuracy in classifying malignant potential ex vivo. Should these results hold in vivo, this algorithm would meet the ASGE's PIVI criteria for NPV, supporting clinical use of OCT for either a lower colon 'diagnose and leave' strategy and/or 'resect and discard' strategy for diminutive polyps.

Keywords: colon polyp detection; optical coherence tomography (OCT); preservation and incorporation of valuable endoscopic innovations (PIVI).