Abstract

Optical coherence tomography (OCT) is an advanced imaging modality to detect Barrett’s esophagus (BE) dysplasia, providing widefield, cross-sectional imaging and microscopic resolution. BE dysplasia is characterized under OCT by the presence and number of glandular structures with atypical morphology. Accurate detection and interpretation of BE glands under OCT is essential to detect dysplastic lesions. Object Detection using deep learning has the potential to identify glands from OCT images. We developed a YOLO model to identify the presence of glands in BE tissue. The YOLOv4 object detector was trained on a custom BE dataset of 30 patients with confirmed BE who underwent OCT imaging, of which 222 OCT images included at least one gland. Our model identified glands with a high average precision of 88.79% on the test dataset. We showed that the developed model is robust to rotation, brightness, and blur in images. We have implemented an object detection model to identify glands from OCT images with promising results accurately. This model has the potential to improve the diagnosis and surveillance of BE by eliminating human error and missed dysplastic lesions adaptable for capsule endoscopy applications.

This content is only available via PDF.