A retinal vessel detection approach using convolution neural network with reinforcement sample learning strategy

Computer-aided detection (CAD) provides an efficient way to assist doctors to interpret fundus images. In a CAD system, retinal vessel (RV) detection is an important step to identify the retinal disease regions automatically and accurately. However, RV detection is still a challenging problem due to variations in morphology of the vessels on a noisy background. In this paper, we formulate the detection task as a classification problem and solve it using a convolutional neural network (CNN) as a two-class classifier. The proposed model has 2 convolution layers, 2 pooling layers, 1 dropout layer and 1 loss layer. The contributions of the algorithm are two-fold. First, a new model of CNN is designed to automatically extract features and classify the retinal vessel region. Compared to traditional classification procedures, it is fully automatic and does not need preprocessing and manual extraction and description of features. Second, a novel reinforcement sample learning scheme is proposed to train the CNN with fewer iterations of epochs and less training time. The proposed model is trained and tested using the Digital Retinal Images for Vessel Extraction (DRIVE) and Structured Analysis of the Retina (STARE) data sets. The proposed CNN achieves better performance and significantly outperforms the state-of-the-art for automatic retinal vessel segmentation on the DRIVE data set with 91.99% accuracy and 0.9652 AUC score (area under ROC), and on the STARE data set with 92.20% accuracy and 0.9440 AUC value. We further compare our result with several state-of-the-art methods based on AUC values. The comparison shows that our proposal yields the second best AUC value. This demonstrates the efficiency of the proposed method without pre-processing and with high accuracy and training speed.