Deep-learning-based Segmentation of Fundus Photographs to Detect Central Serous Chorioretinopathy

Published: 23 August 2021| Version 2 | DOI: 10.17632/4k64fwnp4k.2
TaeKeun Yoo


We developed a pix2pix deep learning model for segmentation of subretinal fluid area in fundus photographs to detect central serous chorioretinopathy (CSC). The dataset include fundus photographs and segmentation images from 105 eyes with CSC and 40 healthy eyes. We retrospectively reviewed the medical records and multimodal images of a total of 115 images of patients with had CSC at Aerospace Medical Center and from publicly accessible databases. Finally, the total dataset includes fundus photographs and segmentation images from 115 eyes with CSC and 40 healthy eyes from the medical center and publicly accessible datasets. The reference segmentation for subretinal fluid area was performed manually by an expert ophthalmologist. First, the user should upload "pix2pix_csc_segmentation.ipynb" file in the Google drive. And open the file in the Google drive page. Second, please link the datasets to this colab notebook using Google drive. For example, we save the training dataset at "csc/segmentation/seg_pix/" (in the file) and the test dataset at "csc/segmentation/seg_pix_test/" (in the file, too). Third, run the codes in Google Colab by clicking buttons.