26
2018 CVPR
Performance assessment of the deep learning technologies in grading glaucoma severity
Method : 分类 (0,1,2,3) 无,轻,中,重
Dataset : Privary 5978 (北京同仁医院 Beijing Tongren Hospital)
Architecture : 比较 8 种经典 net
Results : Best AC 75.5%
中国
to validate and compare the performace of 8 DL architectures
Method
Pre-processing
-
global ROIs
将图像调整为较小的图像,分辨率降低
-
local ROIs
将原始图片的 local OD crop ,分辨率高
Train
- VGG16, VGG19
- ResNet,DenseNet
- InceptionV3, InceptionResNet, Xception
- NASNetMobile(神经架构搜索)
Result
-
the highest classification accuracy 75.5%
DenseNet
global ROIs
pre-trained weights
Discussion
比较和验证了几种经典的深度学习技术在青光眼视神经病变严重程度分级中的性能,特别是研究了全局视场图像和局部视盘图像对分类性能的影响。 研究的重点不是开发新的深度学习架构。
- The experiments demonstrated the feasibility of the deep learning technology in grading glaucoma severity
-
global field-of-view images contain relatively richer information that may be critical for glaucoma assessment , suggesting that we should use the entire field-of-view of a fundus image for training a deep learning network
尽管图像分辨率显著降低,但 global roi在分类精度上优于local roi