Hard Exudates Segmentation in Fundus Image via Combining Automatic and CNN-based Interactive Methods

Hard exudates (EX) are one of the early signs of the diabetic retinopathy (DR), a leading cause of irreversible blindness. Computer
assisted segmentation method makes screening faster and provide quantitative measurements of EX lesions. In this paper, we propose a method
which takes advantage of both an automatic segmentation model and an interactive one which enables click-based corrections in the case of supoptimal EX segmentation. We evaluated the proposed method on three fundus image datasets including two public datasets (IDRiD and DDR)
and a private dataset (E-Hos) using the area under precision-recall curve (AUPR) metric. The results show that the proposed method achieved
AUPR scores of 0.851, 0.693 and 0.761 on the IDRiD, DDR and E-Hos datasets, respectively. The experiments also demonstrates that the proposed method outperforms several state-of-the-art CNNs in EX lesion segmentation.