Segmentation of lesions in eye fundus images (EFI) is a di cult problem, due to small sizes, varying morphologies,similarities and lack of contrast. Today, deep learning segmentation architectures are state-of-the-art in mostsegmentation tasks. But metrics need to be interpreted adequately to avoid wrong conclusions, e.g. we showthat 90% global accuracy of the Fully Convolutional Network (FCN) does not mean it segments lesions verywell. In this work we test and compare deep segmentation networks applied to nd lesions in the Eye FundusImages, focusing on comparison and how metrics really should be interpreted to avoid mistakes and why. In thelight of this analysis, we nalize by discussing further challenges that lie ahead.
展开▼