Abstract:
Automatic retinal image analysis has remained an important topic of research in the last 10 years. Various algorithms and methods have been developed for analyzing retinal images. Most of these methods use public retinal image databases for testing the performance of their methods without checking the retinal image quality before proceeding to performance evaluation making the performance metrics reported by these methods inconsistent. In this paper we propose a deep learning based approach to assess the quality of input retinal images. The method starts with a deep learning based classification which identifies the image quality in terms of sharpness, illumination and homogeneity followed by an unsupervised second stage evaluating the field definition and content in the image. Using the intra-database cross-validation technique, the proposed method achieves overall sensitivity, specificity, positive predictive value, negative predictive value and accuracy of above 90\% when tested on 7007 images collected from seven different public databases, including our own developed database UoA-DR database. Hence the proposed method is generalized and robust making it more suitable to adopt in clinical practice.