Classification of Retinal Cysts on SD-OCT Images Using Stacked Auto-Encoder

Studies have shown that diabetes costs over $770 million in USA. Diabetic retinopathy (DR) and its complication diabetic macular edema (DME) and age-related macular degeneration (AMD) are crucial diseases that might affect the retina and lead to blindness. Optical Coherence Tomography screening is one of the most effective screening methods to diagnose the retinal cysts lesions and to view the retina in 3-D volumes. This paper presents a methodology for automated detection of retinal cysts on Spectral Domain OCT (SD-OCT) volumes. The proposed method considers a generic classification pipeline that extracts the stable regions and then compares the stable regions with the optimal ground truth in order to label the potential regions. After that, the potential regions were resized and sent to the autoencoder to extract the features in an unsupervised fashion. Finally, the trained data was classified using the softmax layer in a supervised fashion, and the test data is passed through the network to validate the results. The results obtained from our pipeline are 93.0% and 82.0% for sensitivity (SE) and specificity (SP) respectively.