A Framework for Generating an ICGA from a Fundus Image using GAN

Fundus image is an initial method for examining eye diseases. Due to its visualization limitation, diseases such as wet macular degeneration (wet amd) and polypoidal choroidal vasculopathy (PCV) looks similar. An indocyanine green angiography (ICGA) was introduced to cope with this limitation, but the obtaining process is invasive and complicated. To avoid unnecessary painful processes for non-PCV patients, this work presents a framework to generate an ICGA from an input fundus image for a PCV screening. Given pairs of high-resolution fundus and ICGA images, they are pre-processed to be trained using a generative adversarial network (GAN). The generated patches are combined and post-processed to obtained the generated ICGA image. Several experiments were conducted to acquire suitable pre-processing, augmentation, post-processing function of the proposed framework. The image quality assessment (IQA) called PSNR was employed to assess the generated images. The average PSNR of max and mean function of the generated ICGA when comparing with the original colored fundus is 16.35, while the one with the original ICGA is 16.25. Although the PSNR score is less than the acceptable range, the visual output shows the sign of the polyp that can be applied in further classification tasks.