Feasibility of Depth-in-Color En Face Optical Coherence Tomography for Colorectal Polyp Classification Using Ensemble...

Area of Science:

• Gastroenterology and Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Colorectal cancer screening faces challenges in accurately identifying precursor lesions due to subtle surface appearances.
• Optical coherence tomography (OCT) offers depth-sensitive imaging, potentially enhancing lesion detection.

Purpose of the Study:

• To evaluate the accuracy of deep learning-based analysis of en face OCT projections for classifying the malignant potential of colorectal polyps.
• To assess the performance of the algorithm for both all polyps and diminutive polyps (≤5 mm).

Main Methods:

• OCT imaging was performed on 300 subjects' polyps.
• Depth information was color-encoded to create en face OCT projections.
• An ensemble deep learning network was trained on annotated projections to predict malignant potential.

Main Results:

• The algorithm achieved an area under the curve (AUC) of 0.90 for detecting malignant potential in all polyps.
• For diminutive polyps (≤5 mm), the AUC was 0.88.
• These ex vivo results indicate high accuracy in classifying polyp malignancy.

Conclusions:

• The developed deep learning algorithm demonstrates high accuracy in classifying the malignant potential of colorectal polyps ex vivo.
• If validated in vivo, this OCT-based approach could meet clinical criteria (ASGE PIVI) for negative predictive value.
• The findings support potential clinical applications of OCT in colorectal cancer screening, including 'diagnose and leave' or 'resect and discard' strategies for diminutive polyps.