A Pilot Study of Automated Predictive Models for Retinal Diseases

Diabetic retinopathy, glaucoma, Central Serous Retinopathy (CSR), age-related macular degeneration (AMD), and retinitis are primary causes of visual diseases worldwide. As such, several types of retinal disease predictive or diagnostic models are designed to prevent vision loss or impairment. Since correct prediction is crucial for treatment, a survey of existing retinal disease predictive or diagnostic models was conducted, and algorithms used to predict retinal disease were analyzed. The survey showed that despite improvements with the incorporation of machine learning, many automated retinal disease diagnosis systems still rely heavily on traditional models for classification tasks. Thus, limiting the retinal disease SVM models’ performance in handling complex, high-dimensional retinal images. Therefore, this study incorporates a Convolutional Neural Network-based framework to directly learn discriminative features from raw retinal images without manual intervention to predict kinds of retinal diseases. In the future, the efficiency of this approach will be demonstrated by developing and implementing a CNN-based retinal disease predictive system for diabetic retinopathy, glaucoma, CSR, AMD, and retinitis, and evaluating it for real-world clinical use.