Assessment of central serous chorioretinopathy (CSC) depicted on color fundus photographs using deep Learning

PURPOSE To investigate whether and to what extent central serous chorioretinopathy (CSC) depicted on color fundus photographs can be assessed using deep learning technology. METHODS We collected a total of 2,504 fundus images acquired on different subjects. We verified the CSC status of these images using their corresponding optical coherence tomography images. A total of 1,329 images depicted CSC. These images were preprocessed and normalized. This resulting data set was randomly split into three parts in the ratio of 8:1:1, respectively, for training, validation, and testing purposes. We used the deep learning architecture termed Inception-V3 to train the classifier. We performed nonparametric receiver operating characteristic analyses to assess the capability of the developed algorithm to identify CSC. To study the inter-reader variability and compare the performance of the computerized scheme and human experts, we asked two ophthalmologists (i.e., Rater #1 and #2) to independently review the same testing data set in a blind manner. We assessed the performance difference between the computer algorithms and the two experts using the receiver operating characteristic curves and computed their pair-wise agreements using Cohen's Kappa coefficients. RESULTS The areas under the receiver operating characteristic curve for the computer, Rater #1, and Rater #2 were 0.934 (95% confidence interval = 0.905-0.963), 0.859 (95% confidence interval = 0.809-0.908), and 0.725 (95% confidence interval = 0.662-0.788). The Kappa coefficient between the two raters was 0.48 (P < 0.001), while the Kappa coefficients between the computer and the two raters were 0.59 (P < 0.001) and 0.33 (P < 0.05). CONCLUSION Our experiments showed that the computer algorithm based on deep learning can assess CSC depicted on color fundus photographs in a relatively reliable and consistent way.