A two-phase pre-filtering approach to the automatic screening of digital fundus images

In this paper, we present an approach to decrease the computational burden of an automatic screening system designed for diabetic retinopathy. The proposed method consists of two steps. First, a pre-screening algorithm is considered to classify the input digital fundus images based on their abnormality. If an image is found to be abnormal, it will not be analyzed further with robust lesion detector algorithms. As an improvement, we introduce a novel feature extraction approach based on clinical observations. The second step of the proposed method detects regions which contain possible lesions for images that have been passed pre-screening. These regions will serve as inputs to lesion detectors later on, which can achieve better computational performance by operating on specific regions only instead of the entire image. Experimental results show that both two steps of the proposed approach are valid to efficiently exclude a large amount of data from further processing to improve the performance of an automatic screening system.