PATCH-BASED FULLY CONVOLUTIONAL NEURAL NETWORK WITH SKIP CONNECTIONS FOR RETINAL BLOOD VESSEL SEGMENTATION

Automated segmentation of retinal blood vessels plays an important role in the computer aided diagnosis of retinal diseases. The paper presents a new formulation of patch-based fully Convolutional Neural Networks (CNNs) that allows accurate segmentation of the retinal blood vessels. A major modification in this retinal blood vessel segmentation task is to improve and speed-up the patch-based fully CNN training by local entropy sampling and a skip CNN architecture with class-balancing loss. The proposed method is experimented on DRIVE dataset and achieves strong performance and significantly outperforms the-state-of-the-art for retinal blood vessel segmentation with 78.11% sensitivity, 98.39% specificity, 95.60% accuracy, 87.36% precision and 97.92% AUC score respectively.