A foundation model for generalizable disease detection from retinal images

Zhou, Yukun and Chia, Mark A. and Wagner, Siegfried K. and Ayhan, Murat S. and Williamson, Dominic J. and Struyven, Robbert R. and Liu, Timing and Xu, Moucheng and Lozano, Mateo G. and Woodward-Court, Peter and Kihara, Yuka and Allen, Naomi and Gallacher, John E. J. and Littlejohns, Thomas and Aslam, Tariq and Bishop, Paul and Black, Graeme and Sergouniotis, Panagiotis and Atan, Denize and Dick, Andrew D. and Williams, Cathy and Barman, Sarah and Barrett, Jenny H. and Mackie, Sarah and Braithwaite, Tasanee and Carare, Roxana O. and Ennis, Sarah and Gibson, Jane and Lotery, Andrew J. and Self, Jay and Chakravarthy, Usha and Hogg, Ruth E. and Paterson, Euan and Woodside, Jayne and Peto, Tunde and Mckay, Gareth and Mcguinness, Bernadette and Foster, Paul J. and Balaskas, Konstantinos and Khawaja, Anthony P. and Pontikos, Nikolas and Rahi, Jugnoo S. and Lascaratos, Gerassimos and Patel, Praveen J. and Chan, Michelle and Chua, Sharon Y. L. and Day, Alexander and Desai, Parul and Egan, Cathy and Fruttiger, Marcus and Garway-Heath, David F. and Hardcastle, Alison and Khaw, Sir Peng T. and Moore, Tony and Sivaprasad, Sobha and Strouthidis, Nicholas and Thomas, Dhanes and Tufail, Adnan and Viswanathan, Ananth C. and Dhillon, Bal and Macgillivray, Tom and Sudlow, Cathie and Vitart, Veronique and Doney, Alexander and Trucco, Emanuele and Guggeinheim, Jeremy A. and Morgan, James E. and Hammond, Chris J. and Williams, Katie and Hysi, Pirro and Harding, Simon P. and Zheng, Yalin and Luben, Robert and Luthert, Phil and Sun, Zihan and McKibbin, Martin and O’Sullivan, Eoin and Oram, Richard and Weedon, Mike and Owen, Chris G. and Rudnicka, Alicja R. and Sattar, Naveed and Steel, David and Stratton, Irene and Tapp, Robyn and Yates, Max M. and Petzold, Axel and Madhusudhan, Savita and Altmann, Andre and Lee, Aaron Y. and Topol, Eric J. and Denniston, Alastair K. and Alexander, Daniel C. and Keane, Pearse A. (2023) A foundation model for generalizable disease detection from retinal images. Nature, 622 (7981). pp. 156-163. ISSN 0028-0836

[thumbnail of s41586-023-06555-x.pdf] Text
s41586-023-06555-x.pdf - Published Version
Download (22MB)

Abstract

Medical artificial intelligence (AI) offers great potential for recognizing signs of health conditions in retinal images and expediting the diagnosis of eye diseases and systemic disorders1. However, the development of AI models requires substantial annotation and models are usually task-specific with limited generalizability to different clinical applications2. Here, we present RETFound, a foundation model for retinal images that learns generalizable representations from unlabelled retinal images and provides a basis for label-efficient model adaptation in several applications. Specifically, RETFound is trained on 1.6 million unlabelled retinal images by means of self-supervised learning and then adapted to disease detection tasks with explicit labels. We show that adapted RETFound consistently outperforms several comparison models in the diagnosis and prognosis of sight-threatening eye diseases, as well as incident prediction of complex systemic disorders such as heart failure and myocardial infarction with fewer labelled data. RETFound provides a generalizable solution to improve model performance and alleviate the annotation workload of experts to enable broad clinical AI applications from retinal imaging.

Item Type: Article
Subjects: Science Global Plos > Multidisciplinary
Depositing User: Unnamed user with email support@science.globalplos.com
Date Deposited: 14 Nov 2023 06:58
Last Modified: 14 Nov 2023 06:58
URI: http://ebooks.manu2sent.com/id/eprint/2134

Actions (login required)

View Item
View Item