Automated detection of retinal disorders from OCT images using artificial neural network

The advent of Optical Coherence Tomography (OCT) imaging has engendered a quantum leap in ophthalmological disease diagnosis. Specifically, in relation to various retinal disorders, OCT has facilitated visualization of minute structural changes in retinal and choroid layers. However, due to dearth of ophthalmologists, and time and effort required in manual analysis, a large number of patients fail to enjoy the full benefit of OCT-based diagnosis. Against this backdrop, we propose to automate detection of retinal disorders so as to reduce clinicians' burden per patient, and hence increase access to such eyecare. In this regard, we demonstrated automated diagnosis using an artificial neural network (ANN) classifier. In the process, we demonstrated the importance of choroidal features in addition to the usual age, gender and retinal features in improving detection performance. Specifically, using a dataset of 169 normal and diseased images each, upon Monte Carlo cross validation we obtained sensitivity, specificity and accuracy levels of 99.02 ± 1.57%, 98.29 ± 1.68% and 98.65 ± 1.09%, respectively.