Deep Learning Based Approach for Fully Automated Detection and Segmentation of Hard Exudate from Retinal Images

摘要

Diabetic retinopathy (DR), which is a major cause of blindness in the world is characterized by hard exudate lesions inthe eyes as these lesions are one of the most prevalent and earliest symptoms of DR. In this paper, a fully automatedmethod for hard exudate delineation is described that could assist ophthalmologists for timely diagnosis of DR beforedisease progress to a level beyond treatment. We used a dataset consist of 107 images to develop a U-Net-based methodfor hard exudate detection and segmentation. This network consists of shrinking and expansive streams in whichshrinking path has the same structure as conventional convolutional networks. In expansive path, obtained features aremerged with those from shrinking path with the proper resolution to generate multi-scale features and accomplishdistinction between hard exudate and normal tissue in retinal images. The training images were augmented artificially toincrease the number of samples in the dataset and avoid overfitting issues. Experimental results showed that ourproposed method reported sensitivity, specificity, accuracy, and Dice similarity coefficient of 96.15%, 80.77%, 88.46%,and 67.23 ± 13.60% on 52 test images, respectively.