Real-time colorectal cancer diagnosis using PR-OCT with deep learning

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Optical coherence tomography (OCT) shows potential for differentiating normal colonic mucosa from neoplasia.
• Endoscopic biopsy is the current gold-standard for colorectal cancer screening and surveillance.
• Processing large volumes of OCT data poses a challenge for clinical translation.

Purpose of the Study:

• To develop a deep learning-based pattern recognition (PR) OCT system for automated, real-time diagnosis of colonic neoplasia.
• To evaluate the diagnostic accuracy of the PR-OCT system compared to histologic findings.

Main Methods:

• A convolutional neural network was designed to analyze structural patterns in human colon OCT images.
• The network was trained and tested on approximately 26,000 OCT images from various colonic tissue types.
• The PR-OCT system's diagnoses were compared against known histologic findings.

Main Results:

• The PR-OCT system achieved 100% sensitivity and 99.7% specificity in detecting neoplastic colorectal tissue.
• An area under the receiver operating characteristic (ROC) curve (AUC) of 0.998 was obtained.
• The system successfully identified patterns distinguishing normal from neoplastic colonic mucosa.

Conclusions:

• The PR-OCT system provides accurate, real-time computer-aided diagnosis of colonic neoplastic mucosa.
• This technology can serve as an "optical biopsy" tool to aid clinicians in early neoplasm screening.
• Future development aims to integrate this system for real-time screening and treatment evaluation.