Joint segmentation and classification of retinal arteries/veins from fundus images

Abstract

Objective: Automatic artery/vein (A/V) segmentation from fundus images is required to track blood vessel changes occurring with many pathologies including retinopathy and cardiovascular pathologies. One of the clinical measures that quantifies vessel changes is the arterio-venous ratio (AVR) which represents the ratio between artery and vein diameters. This measure significantly depends on the accuracy of vessel segmentation and classification into arteries and veins. This paper proposes a fast, novel method for semantic A/V segmentation combining deep learning and graph propagation. Methods: A convolutional neural network (CNN) is proposed to jointly segment and classify vessels into arteries and veins. The initial CNN labeling is propagated through a graph representation of the retinal vasculature, whose nodes are defined as the vessel branches and edges are weighted by the cost of linking pairs of branches. To efficiently propagate the labels, the graph is simplified into its minimum spanning tree.Results: The method achieves an accuracy of 94.8% for vessels segmentation. The A/V classification achieves a specificity of 92.9% with a sensitivity of 93.7% on the CT-DRIVE database compared to the state-of-the-art-specificity and sensitivity, both of 91.7%. Conclusion: The results show that our method outperforms the leading previous works on a public dataset for A/V classification and is by far the fastest. Significance: The proposed global AVR calculated on the whole fundus image using our automatic A/V segmentation method can better track vessel changes associated to diabetic retinopathy than the standard local AVR calculated only around the optic disc.

Aiding the Diagnosis of Diabetic and Hypertensive Retinopathy Using Artificial Intelligence-Based Semantic Segmentation

“…Dehazing was used as a pre-processing technique to avoid haze and shadow noise, and classification was performed by a CNN that was trained on 800,000 patches with a dimension of 27 × 27 (the center pixel was considered the decision pixel) [42]. Girard et al proposed a fast deep learning method to segment vessels using a U-Net-inspired CNN for semantic segmentation, where the encoder and decoder provide the down-sampling and up-sampling of the image, respectively [43]. Hu et al proposed a method for retinal vessel segmentation based on a CNN and conditional random fields (CRFs).…”

SPECHT: Self-tuning Plausibility Based Object Detection Enables Quantification of Conflict in Heterogeneous Multi-scale Microscopy

“…(i) Joint or hybrid segmentation and classification decomposes the image into segments and classifies each segment, and have been applied successfully to microscopy [11], chromosome microscopy [12], breast biopsy [13], fundus images [14], and histopathology [15] to name a few. However, typically they require annotations, which we do not have, do not provide a theoretical upper bound to the label assigned to each segment, nor is it clear on how to apply the same method across heterogeneous datasets or quantify conflict.…”

Semi-supervised Multi-task Learning with Chest X-Ray Images

“…Rezaei et al [9] proposed a GAN model combining a set of auto-encoders with an LSTM unit and an FCN as discriminator for semantic segmentation and disease prediction. Girard et al [10] used a U-Net-like architecture coupled with graph propagation to jointly segment and classify retinal vessels. Mehat et al [11] proposed a Y-Net, with parallel discriminative and convolutional modularity, for the joint segmentation and classification of breast biopsy images.…”