Cross-Domain Images Generation of Fundus Fluorescence Angiography Based on Generative Adversarial Networks with Self-Attention Mechanism

Fundus Fluorescein Angiography (FFA) is considered the gold standard technique for diagnosing retinal diseases. However, the invasive nature of the examination and the risk of adverse reactions to the contrast agent pose limitations and safety concerns. Therefore, it is important to explore a safe, efficient, and cost-effective alternative to acquire FFA images. Based on the rapid development of deep learning techniques in image synthesis, this paper proposes a supervised cross-domain generative adversarial network with self-attention mechanism to translate Infrared (IR) fundus images into angiography images, both acquired by confocal Scanning Laser Ophthalmoscope (cSLO). Moreover, a preprocessing method suitable for the paired fundus data of different modalities is also proposed. The experimental results prove that the performance of our method is demonstrated to be superior to state-of-the-art cross-domain image translation networks.