Abstract
We demonstrate an adaptation of deep learning for label-free imaging of the micro-scale lymphatic vessels and aqueous veins in the eye using optical coherence tomography (OCT). The proposed deep learning-based OCT lymphangiography (DL-OCTL) method was trained, validated and tested, using OCT scans (23 volumetric scans comprising 19,736 B-scans) from 11 fresh ex vivo porcine eyes with the corresponding vessel labels generated by a conventional OCT lymphangiography (OCTL) method based on thresholding with attenuation compensation. Compared to conventional OCTL, the DL-OCTL method demonstrates comparable results for imaging lymphatics and aqueous veins in the eye, with an Intersection over Union value of 0.79 ± 0.071 (mean ± standard deviation). In addition, DL-OCTL mitigates the imaging artifacts in conventional OCTL where the OCT signal modelling was corrupted by the tissue heterogeneity, provides ~ 10 times faster processing based on a rough comparison and does not require OCT-related knowledge for correct implementation as in conventional OCTL. With these favorable features, DL-OCTL promises to improve the practicality of OCTL for label-free imaging of lymphatics and aqueous veins for preclinical and clinical imaging applications.