2021/07/20 by Duy M. H. Nguyen, Truong T. N. Mai, Nguyen, Duy M. H. +7
Computer Science · Medicine · #Computer Vision and Pattern Recognition (cs.CV) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Neonatal and fetal brain pathology #Retinal Imaging and Analysis
paper · pdf · doi:10.48550/arxiv.2107.09372
openalex publication_date 2021/07/20 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
This paper investigates the problem of domain adaptation for diabetic\nretinopathy (DR) grading. We learn invariant target-domain features by defining\na novel self-supervised task based on retinal vessel image reconstructions,\ninspired by medical domain knowledge. Then, a benchmark of current\nstate-of-the-art unsupervised domain adaptation methods on the DR problem is\nprovided. It can be shown that our approach outperforms existing domain\nadaption strategies. Furthermore, when utilizing entire training data in the\ntarget domain, we are able to compete with several state-of-the-art approaches\nin final classification accuracy just by applying standard network\narchitectures and using image-level labels.\n