Diagnosis of Diabetic Retinopathy and Glaucoma from Retinal Images Using Deep Convolution Neural Network

Diabetic retinopathy and Glaucoma are major retinal diseases that are the primary sources of sightlessness in working-age people all over the universe. The diagnosis of Diabetic Retinopathy (DR) and Glaucoma through color retinal images necessitates trained professionals identifying the existence and importance of numerous small abnormalities, which is a challenging and time-consuming task due to a complex grading scale. Here we suggest a CNN approach for diagnosing Diabetic Retinopathy and Glaucoma and if provided any High - resolution fundus Image of the Retina and appropriately defining its severity, not having the same. Developed a network with Convolutional Neural Network architecture and done data augmenting that can determine the complex aspects that are implicated in the classification, such as micro aneurysms, retinal haemorrhages, and exudates. For DR, the Kaggle dataset is used, and for Glaucoma, the RIGA dataset is used. Kaggle dataset is used as an input for DR which consists of 3658 images that are classified into 5 different classes and RIGA dataset is used as an input for Glaucoma containing 2-class labels of Glaucoma with 1103 images are analyzed. Obtained good predictive accuracy of 98%, the precision of 97.6%, recall of 96%, and an F-measure of 97% an automated system, for instance, can quickly distinguish between healthy and infected retinal images, minimizing the number of clinician reviews.