Ophthalmology. Retina · 2025
Noninvasive Synthesis of Multiframe Ultra-Widefield Fluorescein Angiography from Color Fundus Photographs.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Chen Ruoyu, Xu Kezheng, Zheng Kangyan, Zhang Weiyi, Lu Yan, He Mingguang, Shi Danli
Аннотация
PURPOSE: To generate dye-free ultra-widefield fluorescein angiography (UWF-FA) images from noninvasive ultra-widefield color fundus photography (UWF-CFP) using generative artificial intelligence (AI) and to evaluate its effectiveness in diabetic retinopathy (DR) screening.
DESIGN: A cross-sectional study involving generative AI.
PARTICIPANTS: This study included 1263 patients with DR (2747 UWF-CFP images and 18 321 UWF-FA images) from the Second People's Hospital of Foshan.
METHODS: Ultra-widefield CFP and UWF-FA image pairs were matched and used to train a pix2pixHD generative adversarial network (GAN)-based model modified with Gradient Variance Loss. The generated UWF-FA images were evaluated using quantitative similarity metrics and qualitative ophthalmologist evaluation. An external data set, DeepDRiD, was used to validate the contribution of the generated UWF-FA images to DR grading.
MAIN OUTCOME MEASURES: The area under the receiver operating characteristic curve for DR grading.
RESULTS: The generated early-, mid-, and late-phase UWF-FA images demonstrated high authenticity, with multiscale similarity scores ranging from 0.70 to 0.91 and qualitative evaluation scores from 1.64 to 1.98 (1 = real UWF-FA quality). In a Turing test with 50 randomly selected images, 56% to 76% of the generated images were indistinguishable from real images. Generated UWF-FA images successfully depict details of DR lesions, such as a nonperfusion area and leakage. The incorporation of these generated UWF-FA images in DR grading significantly improved the area under the receiver operating characteristic curve from 0.869 to 0.904 compared with the baseline model using UWF-CFP images alone (P < 0.001).
CONCLUSIONS: The study suggests that the GAN-based model can successfully generate realistic multiframe UWF-FA images, which could enhance DR grading without the need for IV dye injection, with the potential to improve the safety and accessibility of DR screening.
FINANCIAL DISCLOSURE(S): Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.
DESIGN: A cross-sectional study involving generative AI.
PARTICIPANTS: This study included 1263 patients with DR (2747 UWF-CFP images and 18 321 UWF-FA images) from the Second People's Hospital of Foshan.
METHODS: Ultra-widefield CFP and UWF-FA image pairs were matched and used to train a pix2pixHD generative adversarial network (GAN)-based model modified with Gradient Variance Loss. The generated UWF-FA images were evaluated using quantitative similarity metrics and qualitative ophthalmologist evaluation. An external data set, DeepDRiD, was used to validate the contribution of the generated UWF-FA images to DR grading.
MAIN OUTCOME MEASURES: The area under the receiver operating characteristic curve for DR grading.
RESULTS: The generated early-, mid-, and late-phase UWF-FA images demonstrated high authenticity, with multiscale similarity scores ranging from 0.70 to 0.91 and qualitative evaluation scores from 1.64 to 1.98 (1 = real UWF-FA quality). In a Turing test with 50 randomly selected images, 56% to 76% of the generated images were indistinguishable from real images. Generated UWF-FA images successfully depict details of DR lesions, such as a nonperfusion area and leakage. The incorporation of these generated UWF-FA images in DR grading significantly improved the area under the receiver operating characteristic curve from 0.869 to 0.904 compared with the baseline model using UWF-CFP images alone (P < 0.001).
CONCLUSIONS: The study suggests that the GAN-based model can successfully generate realistic multiframe UWF-FA images, which could enhance DR grading without the need for IV dye injection, with the potential to improve the safety and accessibility of DR screening.
FINANCIAL DISCLOSURE(S): Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.