Optical coherence tomography-based deep-learning model for detecting central serous chorioretinopathy

Abstract

Central serous chorioretinopathy (CSC) is a common condition characterized by serous detachment of the neurosensory retina at the posterior pole. We built a deep learning system model to diagnose CSC, and distinguish chronic from acute CSC using spectral domain optical coherence tomography (SD-OCT) images. Data from SD-OCT images of patients with CSC and a control group were analyzed with a convolutional neural network. Sensitivity, specificity, accuracy, and area under the receiver operating characteristic curve (AUROC) were used to evaluate the model. For CSC diagnosis, our model showed an accuracy, sensitivity, and specificity of 93.8%, 90.0%, and 99.1%, respectively; AUROC was 98.9% (95% CI, 0.983-0.995); and its diagnostic performance was comparable with VGG-16, Resnet-50, and the diagnoses of five different ophthalmologists. For distinguishing chronic from acute cases, the accuracy, sensitivity, and specificity were 97.6%, 100.0%, and 92.6%, respectively; AUROC was 99.4% (95% CI, 0.985-1.000); performance was better than VGG-16 and Resnet-50, and was as good as the ophthalmologists. Our model performed well when diagnosing CSC and yielded highly accurate results when distinguishing between acute and chronic cases. Thus, automated deep learning system algorithms could play a role independent of human experts in the diagnosis of CSC.

Figure 1

Performances of the model in classifying CSC and normal eyes. (a) ROC curve comparison with two popular CNN-based architectures: VGG-16 and Resnet-50. The AUROC of our model was 0.989%, which is slightly lower than VGG-16 and higher than Resnet-50. (b) Performance comparison between our model and ophthalmologists. The blue line (ROC curve) is created by sweeping a threshold over the predicted probability for a specific clinical diagnosis. The asterisk denotes our model’s performance with the optimal threshold. (c) An expanded version of (b). (d) Sensitivity, specificity, and accuracy of our model and five ophthalmologists including two human experts. The accuracy is the number of true positives and the number of true negatives divided by the total number of test images. AUROC, area under the receiver operating characteristic curve; CNN, convolutional neural network; CSC, central serous chorioretinopathy; ROC, receiver operating characteristic.

Figure 2

Performances for acute CSC vs chronic CSC classification. (a) ROC curve comparison with two popular CNN-based architectures: VGG-16 and Resnet-50. The AUROC of our model was 0.994%, which outperforms both VGG-16 and Resnet-50. (b) Performance comparison between our model and ophthalmologists. The blue line (ROC curve) is created by sweeping a threshold over the predicted probability for a specific clinical diagnosis. The asterisk denotes our model’s performance with the optimal threshold. (c) An expanded version of (b). (d) Sensitivity, specificity, and accuracy of our model and five ophthalmologists. Our model’s performance is better than most of the ophthalmologists. AUROC, area under the receiver operating characteristic curve; CNN, convolutional neural network; CSC, central serous chorioretinopathy; ROC, receiver operating characteristic.

Figure 3

Heat maps for the classification models by Grad-CAM. (a) A heat map for the model classifying normal and CSC OCT images; (b), (c) Heat maps for the model classifying acute and chronic CSC. The Grad-CAM was able to identify pathologic regions on the OCT, which are presented as a heat map. CSC, central serous chorioretinopathy; Grad-CAM, gradient weighted class activation mapping; OCT, optical coherence tomography.

Figure 4

An illustration of the proposed model used in classifying CSC and normal eyes. Our model comprises an input layer, 13 CNN layers with ReLU activation functions, 4 max pooling layers, 2 dropout layers, and 4 FC layers. The last FC layer was used for binary classification. The heat map was generated from the final CNN layer. CNN, convolutional neural network; CSC, central serous chorioretinopathy; FC, fully connected; Grad-CAM, gradient weighted class activation mapping ReLU, rectified linear unit; SD-OCT, spectral domain optical coherence tomography.