Non-Invasive to Invasive: Enhancing FFA Synthesis from CFP with a Benchmark Dataset and a Novel Network

TL;DR

This research bridges the gap between non-invasive imaging and FFA, thereby offering promising prospects to enhance ophthalmic diagnosis and patient care, with a focus on reducing harm to patients through non-invasive procedures.

Abstract

Fundus imaging is a pivotal tool in ophthalmology, and different imaging modalities are characterized by their specific advantages. For example, Fundus Fluorescein Angiography (FFA) uniquely provides detailed insights into retinal vascular dynamics and pathology, surpassing Color Fundus Photographs (CFP) in detecting microvascular abnormalities and perfusion status. However, the conventional invasive FFA involves discomfort and risks due to fluorescein dye injection, and it is meaningful but challenging to synthesize FFA images from non-invasive CFP. Previous studies primarily focused on FFA synthesis in a single disease category. In this work, we explore FFA synthesis in multiple diseases by devising a Diffusion-guided generative adversarial network, which introduces an adaptive and dynamic diffusion forward process into the discriminator and adds a category-aware representation enhancer. Moreover, to facilitate this research, we collect the first multi-disease CFP and FFA paired dataset, named the Multi-disease Paired Ocular Synthesis (MPOS) dataset, with four different fundus diseases. Experimental results show that our FFA synthesis network can generate better FFA images compared to state-of-the-art methods. Furthermore, we introduce a paired-modal diagnostic network to validate the effectiveness of synthetic FFA images in the diagnosis of multiple fundus diseases, and the results show that our synthesized FFA images with the real CFP images have higher diagnosis accuracy than that of the compared FFA synthesizing methods. Our research bridges the gap between non-invasive imaging and FFA, thereby offering promising prospects to enhance ophthalmic diagnosis and patient care, with a focus on reducing harm to patients through non-invasive procedures. Our dataset and code will be released to support further research in this field (https://github.com/whq-xxh/FFA-Synthesis).

Figures (4)

• Figure 1: Demonstrating a series of fundus diseases, along with the healthy example. Each category is represented by two distinct paired images, arranged from the first row to the second row, including two different imaging modalities: CFP, and FFA.
• Figure 2: Overview of our novel network: Dynamic Diffusion-Guided GAN. The Diffusion-Guided GAN, features a dynamic diffusion-guided discriminator and a category-aware representation enhancer. This network takes a CFP image $x$ as input and synthesizes the corresponding FFA image $y_g$. First, the input CFP image is fed into an initial convolution layer to obtain the latent representation. Meanwhile, the corresponding category information is embedded by an embedding layer and fused with the latent representation using a summation manner. Then, the latent feature with category information is input into a generator and a registration network to obtain the registered synthesized FFA image. For the discriminator, we design a Diffusion-guided discriminator with an adaptive noising process to add noise of varying degrees, aiming to achieve a more stable training process and improved performance.
• Figure 3: Visualizations between original images and the synthesized images produced by our method and other state-of-the-art methods. Our Dynamic Diffusion-guided GAN model can generate clearer blood vessel regions with more detailed information.
• Figure 4: Illustration of our proposed network for developing a diagnostic model for common fundus diseases utilizing paired multi-modal imaging.