Clinically applicable deep learning for diagnosis of diabetic retinopathy

Diabetic retinopathy (DR) is the kind of diabetes complication that affects eyes and can damage the blood vessels inside the retina. To diagnose the strength of DR disease based on examination of the retina. Nowadays, the common diagnosis process asks for experienced ophthalmologists to inspect both fundus image and OCT (optical coherence tomography) images, which is time-consuming and not very convenient for remote rural inhabitants. The research purpose in this paper is to propose a new paradigm of automatic DR diagnosis by using artificial intelligence and cloud computing. Inside the DCNN, we changed max-pooling layers with factional max-pooling. We trained using support vector machine (SVM) to learn the underlying boundary of distribution of each category. Using that proposed method, we achieved the results of the recognition up to 86.17%. We also develop an iPhone APP. It called 'Deep Retina' that equipped with a handheld ophthalmoscope, a layman can take fundus images and perform the diagnosis automatically without intervention from ophthalmologists. It is a practically applicable telemedicine system which benefits the home care, remote medical care, and self-examination.