Automatic grading of Diabetic macular edema based on end-to-end network

作者：

Highlights：

• An architecture is designed to grade DME with high accuracy and no extra information.

• The method combines channel attention with disease attention to learn more feature.

• Iterative robust homomorphic surface fitting is used to supply data in preprocessing.

• Loss function with class weights is used to solve the class imbalance.

摘要

•An architecture is designed to grade DME with high accuracy and no extra information.•The method combines channel attention with disease attention to learn more feature.•Iterative robust homomorphic surface fitting is used to supply data in preprocessing.•Loss function with class weights is used to solve the class imbalance.

论文关键词：DME,ResNet50,Attention,End-to-end network,Preprocessing

论文评审过程：Received 7 June 2022, Revised 9 August 2022, Accepted 11 September 2022, Available online 16 September 2022, Version of Record 26 September 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.118835