2020 年 140 巻 5 号 p. 549-550
Automatic classification of fundus blood vessels into artery and vein is one of the important topics in fundus image analysis. The conventional method, which employs pixel-wise training data, takes high costs for creating the training data. In this paper, we propose a very simple method to create as many training data as possible enough for deep learning, which only needs clipping an image of small size from the whole blood vessel image without any experts knowledge. The effectiveness of the proposed method has been confirmed.
J-STAGEがリニューアルされました! https://www.jstage.jst.go.jp/browse/-char/ja/