MultiGAN: Multi-domain Image Translation from OCT to OCTA

Optical coherence tomography (OCT) and optical coherence tomography angiography (OCTA) are important imaging techniques for assessing and managing retinal diseases. OCTA can display more blood vessel information than OCT, which, however, requires software and hardware modifications on OCT devices. A large number of OCT data does not have corresponding OCTA data, which greatly limits doctors' diagnosis. Considering the inconvenience of acquiring OCTA images and inevitable mechanical artifacts, we introduce image-to-image translation to generate OCTA from OCT. In this paper, we propose a novel method, MultiGAN, which uses one input image to get three target domain outputs without relying on domain code. We utilize the resnet block in skip connections to preserve details. A domain dependent loss is proposed to impose the restrictions among OCTA projection maps. The dataset containing paired OCT and OCTA images from 500 eyes diagnosed with various retinal diseases is used to evaluate the performance of the proposed network. The results based on cross validation experiments demonstrate the stability and superior performances of the proposed model comparing with state-of-the-art models.