Self-supervised multimodal reconstruction of retinal images over paired datasets
作者：
Highlights：
• Self-supervised multimodal reconstruction tasks are enabled through image registration.
• Non-invasive deep learning pseudo-angiographies are generated from retinographies.
• The generated pseudo-angiographies resemble the original angiographies.
• Generating angiography images requires the recognition of high-level retinal patterns.
• Multimodal reconstruction provides relevant domain information without human labels.
摘要
•Self-supervised multimodal reconstruction tasks are enabled through image registration.•Non-invasive deep learning pseudo-angiographies are generated from retinographies.•The generated pseudo-angiographies resemble the original angiographies.•Generating angiography images requires the recognition of high-level retinal patterns.•Multimodal reconstruction provides relevant domain information without human labels.
论文关键词：Self-supervised learning,Eye fundus,Deep learning,Multimodal,Retinography,Angiography
论文评审过程：Received 13 January 2020, Revised 27 April 2020, Accepted 16 June 2020, Available online 26 June 2020, Version of Record 8 July 2020.
论文官网地址：https://doi.org/10.1016/j.eswa.2020.113674
