INTERMEDIATE AND DEEP CAPILLARY PLEXUSES IN MACHINE LEARNING SEGMENTATION OF HIGH-RESOLUTION OPTICAL COHERENCE TOMOGRAPHY IMAGING.

Abstract

To describe imaging produced by machine learning-based segmentation of high-resolution optical coherence tomography imaging of the intermediate capillary plexus and deep capillary plexus, layers of vessels not imaged well by dye-based angiography. Three healthy subjects with no ocular problems were imaged with spectral domain optical coherence tomography using an instrument with a scanning speed of 85,000 A-scans per second and 3 µm axial optical resolution. A random forest segmentation strategy was used to segment the intermediate capillary plexus and deep capillary plexus. The depth-resolved imaging data was visualized with the help of volume rendering. The high-resolution optical coherence tomography showed the intermediate capillary plexus and deep capillary plexus at the outer borders of the inner nuclear layer. These vessels could be visualized with unprecedented detail in three dimensions. There were multiple bridging vessels connecting to the whorl-like patterns of capillary mesh of the deep capillary plexus, a feature only previously imaged in histologic evaluation of excised eyes. High-resolution optical coherence tomography, machine learning, and advanced image display techniques have wide relevancy in studying the retina in health and disease. Application of this approach has provided images of the deeper vascular layers of the eye that approximate histologic imaging, but noninvasively.