【医学+深度论文：F18】2019 Robust optic disc and cup segmentation with deep learning for glaucoma detection

18

2019 Computerized Medical Imaging and Graphics

Robust optic disc and cup segmentation with deep learning for glaucoma detection

Method ： 分割 OD/OC 计算 CDR
Dataset ：train RIGA ；test DRISHTI-GS1 , RIM-ONE
Architecture : Unet （encoder ：pre-trained ResNet34）
Results ：
DRISHTI-GS1：Dice 视盘 96.44% 视杯 87.39%
     Jacc 视盘 93.19% 视杯 78.08%
RIM-ONE ：Dice 视盘 94.91% 视杯 79.32%
     Jacc 视盘 90.65% 视杯 68.28%

避免了从头开始训练网络，实现了快速网络训练

Methods

视盘和杯的自动分割是以两个阶段的方式实现的

• Pre-Processing

• Data Augumentation
  512×512

  <