Location-aware optical coherent tomography (OCT) tethered capsule endomicroscopy (TCE) of the small intestine

For OCT-tethered capsule endomicroscopy (TCE) to be a useful minimally invasive tool for evaluating Crohn’s disease, the capsule must be able to be localized within the terminal ileum where the disease often manifests. Here, we developed a machine learning algorithm to assign OCT images of the small intestine into their corresponding anatomical regions. We selected a convolutional neural network and trained it on a set of 2108 cross-sectional images obtained from four swine ex vivo imaging studies to classify images into duodenum, jejunum, or terminal ileum. The model achieved 93±1.72% (95% confidence interval) accuracy on a separate test set of 846 images. These results suggest machine learning may be used to automatically determine when the capsule is in the terminal ileum, enabling microscopic evaluation of this anatomical segment that exhibits pathology in Crohn’s disease.