Classification of EYE Diseases Using Multi-Model CNN

This paper proposes a classification method that utilizes Convolutional Neural Networks (CNNs) to detect eye diseases at an early stage. The most common eye diseases such as Glaucoma (7.7 million), Cataract (94million) and Diabetic Retinopathy (3.9million) are responsible for blindness worldwide. Early detection of these diseases can prevent vision loss. Diabetic Retinopathy for instance, is caused by damage to the retinal blood vessels from diabetes. This can lead to vision loss when the blood vessels swell, leak or close, or when new abnormal blood vessels grow. Glaucoma is another condition that damages the optic nerve, causing vision loss. Cataract, on the other hand, results in blurred or hazy vision due to the clouding of the natural lens in the eye. In this paper, the automatic identification of three major eye diseases-cataracts, glaucoma, and diabetic retinopathy-in pictures of the fundus is proposed using a unique deep neural network called ManualNet. The proposed paper utilizes a variety of pre-trained architectures to train with a medical image dataset, to capture the colors and textures of lesions specific to respective diseases. The proposed Classification of eye disease using multi-model CNN resembles human decision-making and the results obtained through experimentation with different eye disease features as input to convolutional neural networks.