EAD-NET A NOVEL LESION SEGMENTATION METHOD IN DIABETIC RETINOPATHY USING NEURAL NETWORKS
Abstract : Diabetic retinopathy (DR) is a common chronic fundus disease, which has four different kinds of microvessel structure and microvascular lesions: microaneurysms (MAs), hemorrhages (HEs), hard exudates, and soft exudates. Accurate detection and counting of them are a basic but important work. The manual annotation of these lesions is a labor-intensive task in clinical analysis. To solve the problem, we proposed a novel segmentation method for different lesions in DR. Our method is based on a convolutional neural network and can be divided into encoder module, attention module, and decoder module, so we refer it as EAD-Net. After normalization and augmentation, the fundus images were sent to the EAD-Net for automated feature extraction and pixel-wise label prediction. Given the evaluation metrics based on the matching degree between detected candidates and ground truth lesions, our method achieved sensitivity of 92.77%, specificity of 99.98%, and accuracy of 99.97% on the e_ophtha_EX dataset and comparable AUPR (Area under Precision-Recall curve) scores on IDRiD dataset. Moreover, the results on the local dataset also show that our EAD-Net has better performance than original U-net in most metrics, especially in the sensitivity and F1-score, with nearly ten percent improvement. The proposed EAD-Net is a novel method based on clinical DR diagnosis. It has satisfactory results on the segmentation of four different kinds of lesions. These effective segmentations have important clinical significance in the monitoring and diagnosis of DR.
? They differ in the number of studies that built their own CNN structure, and those who preferred to use the existing structures, such as VGG, ResNet, or AlexNet, with transfer learning is slightly small.
? It is notable that the accuracy of the system which built their own CNN structure is higher than those using the existing structures.
? The existence of a reliable DR screening system capable of detecting different lesions types and DR stages leads to an effective follow up system for DR patients, which averted the danger of losing sight.
? It is a global public health problem related to microcirculation disorders which seriously affects human health.
? With the residual structure, the gradient can propagate directly through the skip connection from later layers to the earlier layers, so the vanishing gradient problem can be inhibited.
? There exist many small or fuzzy lesions and too many pooling layers might lead too much semantic information loss, we only use the max pooling layer once to avoid this problem.
• We propose a novel local structure feature called a ring gradient descriptor (RGD) to describe the structural differences between an object and its surrounding area.
• To make up for the lack of traditional features, a novel local structure feature called ring gradient descriptor (RGD) is proposed, which scans the background around the target in an annular way to find the region most similar to the target and calculates the similarity between the region and the target.
• To accomplish this, we propose a novel candidate extraction algorithm based on dual-gray threshold segmentation and morphological processing.
? To a certain degree, the performance of our method on e_ophtha_EX dataset can demonstrate its robustness to normal samples.
? The candidate pixels belonging to red lesions and blood vessels are separated from a reconstructed retinal image with modified coefficients, and then, the full curvelet-based blood vessels are removed, leaving the remaining part as detected red lesions.
? In contrast, our study used a single network structure and only a few changes are needed for the hyper parameter settings.
|