UWAFA-GAN: Ultra-Wide-Angle Fluorescein Angiography Transformation via Multi-scale Generation and Registration Enhancement

TL;DR

This work tackles the invasive and risk-prone practice of ultra-wide-angle fluorescein angiography by proposing UWAFA-GAN, a cross-modality framework that translates UWF-SLO images into UWF-FA images. The model combines dual generators for global and local feature extraction, a registration module to mitigate misalignment, an attention-based skip mechanism, and multi-scale discriminators, optimized with GAN-based, perceptual (VGG), and feature-mapping losses. Empirical results on proprietary intra- and combined datasets show superior quantitative metrics (FID, IS, MS-SSIM, PSNR) and qualitative fidelity, with clinical expert validation indicating realistic synthetic images that support downstream diagnostic tasks. The approach demonstrates potential to reduce invasive procedures while enabling accurate vascular and lesion visualization, though rare lesions remain challenging and future work will integrate pathology priors and broaden datasets.

Abstract

Fundus photography, in combination with the ultra-wide-angle fundus (UWF) techniques, becomes an indispensable diagnostic tool in clinical settings by offering a more comprehensive view of the retina. Nonetheless, UWF fluorescein angiography (UWF-FA) necessitates the administration of a fluorescent dye via injection into the patient's hand or elbow unlike UWF scanning laser ophthalmoscopy (UWF-SLO). To mitigate potential adverse effects associated with injections, researchers have proposed the development of cross-modality medical image generation algorithms capable of converting UWF-SLO images into their UWF-FA counterparts. Current image generation techniques applied to fundus photography encounter difficulties in producing high-resolution retinal images, particularly in capturing minute vascular lesions. To address these issues, we introduce a novel conditional generative adversarial network (UWAFA-GAN) to synthesize UWF-FA from UWF-SLO. This approach employs multi-scale generators and an attention transmit module to efficiently extract both global structures and local lesions. Additionally, to counteract the image blurriness issue that arises from training with misaligned data, a registration module is integrated within this framework. Our method performs non-trivially on inception scores and details generation. Clinical user studies further indicate that the UWF-FA images generated by UWAFA-GAN are clinically comparable to authentic images in terms of diagnostic reliability. Empirical evaluations on our proprietary UWF image datasets elucidate that UWAFA-GAN outperforms extant methodologies. The code is accessible at https://github.com/Tinysqua/UWAFA-GAN.

Figures (6)

• Figure 1: The comprehensive architecture of our proposed UWAFA-GAN model.
• Figure 2: Generators' architectures of the UWAFA-GAN; The fine-level generator (left) and the coarse-level generator (right). The CNN Attention Block and Initial Block are marked with A and B.
• Figure 3: Visual comparison of comparative methods generating UWF-FA from UWF-SLO. The ground truth images are depicted in the sixth column, and the bottom row offers magnified views of the corresponding images from the top row.
• Figure 4: Comparison of the judgment results across clinicians of varying expertise levels. The red digits represent the proportion of generated images considered as authentic one.
• Figure 5: Four pairs of UWF images that elicited false positives in our clinical user study.