Abstract
Retinal pigment epithelium (RPE) melanin thickness maps, derived from multi-contrast images-including the degree of polarization uniformity (DOPU), optical coherence tomography (OCT) angiography, and the attenuation coefficient-are obtained using multi-contrast polarization-sensitive OCT (PS-OCT). These maps have demonstrated utility for three-dimensional assessment of changes in melanin within the RPE (RPE-melanin). While both OCT angiography and the attenuation coefficient can be derived from conventional OCT, measuring the DOPU requires PS-OCT, which is not available on standard commercial OCT systems. To overcome this limitation, we utilized a convolutional neural network to generate DOPU-like images from standard OCT images and used these to calculate a synthesized RPE-melanin thickness map. We evaluated 22 eyes from 20 patients with serous pigment epithelial detachment (PED) secondary to age-related macular degeneration. Both original and synthesized RPE-melanin thickness maps were calculated from multi-contrast PS-OCT datasets. Active RPE lesions were defined as areas with RPE-melanin thickness of ≥ 70 μm (originalRPE(70) and synRPE(70) for the original and synthesized maps, respectively). Both synthesized and original RPE-melanin thickness maps closely resembled near-infrared autofluorescence imaging. Furthermore, both the originalRPE(70) area and synRPE(70) area were significantly positively correlated with the PED volume. Synthesized RPE-melanin thickness maps may be useful for clinical quantification of RPE-melanin.