Deepfakes in Ophthalmology: Applications and Realism of Synthetic Retinal Images from Generative Adversarial Networks.
Deepfakes in Ophthalmology: Applications and Realism of Synthetic Retinal Images from Generative Adversarial Networks.
复制标题
DOI：
10.1016/j.xops.2021.100079
复制
发表时间：
2021-12
影响因子：
--
通讯作者：
Campbell, J. Peter
中科院分区：
文献类型：
--
作者：
Chen, Jimmy S.;Coyner, Aaron S.;Chan, R. V. Paul;Hartnett, M. Elizabeth;Moshfeghi, Darius M.;Owen, Leah A.;Kalpathy-Cramer, Jayashree;Chiang, Michael F.;Campbell, J. Peter
生成对抗网络 (GAN) 是深度学习 (DL) 模型，可以根据真实图像创建和修改逼真的合成图像或深度伪造图像。我们研究的目的是评估专家从真实眼底图像中辨别合成视网膜眼底图像的能力，并回顾 GAN 在眼科领域的当前用途和局限性。 GAN 的开发和专家评估以及文献的非正式回顾。从多中心 ROP 筛查项目中获取了总共 4282 个眼底图像和视网膜血管图图像对。 Pix2Pix HD 是一种高分辨率 GAN，首先在眼底和血管图图像对上进行训练和验证，随后用于从保留的测试集中生成 880 张图像。该测试集中的 50 个合成图像和 50 个不同的真实图像被呈现给 4 位 ROP 眼科医生专家，使用定制的在线系统来评估图像是真实的还是合成的。使用眼科、GAN、生成对抗网络、眼科、图像、深度伪造和合成等术语的组合对 PubMed 和 Google Scholars 上的文献进行了回顾。进行祖先搜索以扩大结果。使用准确度百分比来评估专家辨别真实图像与合成图像的能力。使用 Fisher 精确检验评估统计显着性，P 值≤ 0.05 为显着性阈值。大多数专家正确地将 59% 的图像识别为真实图像或合成图像 (P = 0.1)。专家 1 至 4 正确识别了 54%、58%、49% 和 61% 的图像（P 分别 = 0.505、0.158、1.000 和 0.043）。这些结果表明，大多数专家无法区分真实图像和合成图像。此外，我们在眼科文献中发现了 20 种 GAN 的实现，可应用于各种成像模式和眼科疾病。生成对抗网络可以创建与 ROP 眼科医生专家的真实眼底图像无法区分的合成眼底图像。合成图像可以改善深度学习的数据集增强，可用于学员教育，并可能对患者隐私产生影响。
Generative adversarial networks (GANs) are deep learning (DL) models that can create and modify realistic-appearing synthetic images, or deepfakes, from real images. The purpose of our study was to evaluate the ability of experts to discern synthesized retinal fundus images from real fundus images and to review the current uses and limitations of GANs in ophthalmology. Development and expert evaluation of a GAN and an informal review of the literature. A total of 4282 image pairs of fundus images and retinal vessel maps acquired from a multicenter ROP screening program. Pix2Pix HD, a high-resolution GAN, was first trained and validated on fundus and vessel map image pairs and subsequently used to generate 880 images from a held-out test set. Fifty synthetic images from this test set and 50 different real images were presented to 4 expert ROP ophthalmologists using a custom online system for evaluation of whether the images were real or synthetic. Literature was reviewed on PubMed and Google Scholars using combinations of the terms ophthalmology, GANs, generative adversarial networks, ophthalmology, images, deepfakes, and synthetic. Ancestor search was performed to broaden results. Expert ability to discern real versus synthetic images was evaluated using percent accuracy. Statistical significance was evaluated using a Fisher exact test, with P values ≤ 0.05 thresholded for significance. The expert majority correctly identified 59% of images as being real or synthetic (P = 0.1). Experts 1 to 4 correctly identified 54%, 58%, 49%, and 61% of images (P = 0.505, 0.158, 1.000, and 0.043, respectively). These results suggest that the majority of experts could not discern between real and synthetic images. Additionally, we identified 20 implementations of GANs in the ophthalmology literature, with applications in a variety of imaging modalities and ophthalmic diseases. Generative adversarial networks can create synthetic fundus images that are indiscernible from real fundus images by expert ROP ophthalmologists. Synthetic images may improve dataset augmentation for DL, may be used in trainee education, and may have implications for patient privacy.
登录
查看更多内容
影响因子：
1.9
作者：
Farzin, Hadi;Abrishami-Moghaddam, Hamid;Moin, Mohammad-Shahram
通讯作者：
Moin, Mohammad-Shahram
影响因子：
3.6
作者：
Crystal, Dustin T.;Cuccolo, Nicholas G.;Lin, Samuel J.
通讯作者：
Lin, Samuel J.
DOI：
10.1126/science.aaw4399
发表时间：
2019-03-22
期刊：
Science (New York, N.Y.)
影响因子：
--
作者：
Finlayson SG;Bowers JD;Ito J;Zittrain JL;Beam AL;Kohane IS
通讯作者：
Kohane IS
DOI：
10.1093/jamia/ocy017
发表时间：
2018-08-01
期刊：
Journal of the American Medical Informatics Association : JAMIA
影响因子：
--
作者：
Chang K;Balachandar N;Lam C;Yi D;Brown J;Beers A;Rosen B;Rubin DL;Kalpathy-Cramer J
通讯作者：
Kalpathy-Cramer J
影响因子：
13.7
作者：
Christopher, Mark;Bowd, Christopher;Zangwill, Linda M.
通讯作者：
Zangwill, Linda M.