Self-relabeling for noise-tolerant retina vessel segmentation through label reliability estimation

Background Retinal vessel segmentation benefits significantly from deep learning. Its performance relies on sufficient training images with accurate ground-truth segmentation, which are usually manually annotated in the form of binary pixel-wise label maps. Manually annotated ground-truth label maps, more or less, contain errors for part of the pixels. Due to the thin structure of retina vessels, such errors are more frequent and serious in manual annotations, which negatively affect deep learning performance. Methods In this paper, we develop a new method to automatically and iteratively identify and correct such noisy segmentation labels in the process of network training. We consider historical predicted label maps of network-in-training from different epochs and jointly use them to self-supervise the predicted labels during training and dynamically correct the supervised labels with noises. Results We conducted experiments on the three datasets of DRIVE, STARE and CHASE-DB1 with synthetic noises, pseudo-labeled noises, and manually labeled noises. For synthetic noise, the proposed method corrects the original noisy label maps to a more accurate label map by 4.0–$$9.8\%$$ 9.8 % on $$F_1$$ F 1 and 10.7–$$16.8\%$$ 16.8 % on PR on three testing datasets. For the other two types of noise, the method could also improve the label map quality. Conclusions Experiment results verified that the proposed method could achieve better retinal image segmentation performance than many existing methods by simultaneously correcting the noise in the initial label map.

Multiscale U-Net with Spatial Positional Attention for Retinal Vessel Segmentation

Retinal vessel segmentation is essential for the detection and diagnosis of eye diseases. However, it is difficult to accurately identify the vessel boundary due to the large variations of scale in the retinal vessels and the low contrast between the vessel and the background. Deep learning has a good effect on retinal vessel segmentation since it can capture representative and distinguishing features for retinal vessels. An improved U-Net algorithm for retinal vessel segmentation is proposed in this paper. To better identify vessel boundaries, the traditional convolutional operation CNN is replaced by a global convolutional network and boundary refinement in the coding part. To better divide the blood vessel and background, the improved position attention module and channel attention module are introduced in the jumping connection part. Multiscale input and multiscale dense feature pyramid cascade modules are used to better obtain feature information. In the decoding part, convolutional long and short memory networks and deep dilated convolution are used to extract features. In public datasets, DRIVE and CHASE_DB1, the accuracy reached 96.99% and 97.51%. The average performance of the proposed algorithm is better than that of existing algorithms.

MSC-Net: Multitask Learning Network for Retinal Vessel Segmentation and Centerline Extraction

Automatic segmentation and centerline extraction of blood vessels from retinal fundus images is an essential step to measure the state of retinal blood vessels and achieve the goal of auxiliary diagnosis. Combining the information of blood vessel segments and centerline can help improve the continuity of results and performance. However, previous studies have usually treated these two tasks as separate research topics. Therefore, we propose a novel multitask learning network (MSC-Net) for retinal vessel segmentation and centerline extraction. The network uses a multibranch design to combine information between two tasks. Channel and atrous spatial fusion block (CAS-FB) is designed to fuse and correct the features of different branches and different scales. The clDice loss function is also used to constrain the topological continuity of blood vessel segments and centerline. Experimental results on different fundus blood vessel datasets (DRIVE, STARE, and CHASE) show that our method can obtain better segmentation and centerline extraction results at different scales and has better topological continuity than state-of-the-art methods.

Azzopardi Phenomenon in a Non-neoplastic Phthisical Eye: A Case Report

Objective: The Azzopardi phenomenon, known as the deoxyribonucleic acid deposition on various structures due to cellular necrosis, has never been reported in non-neoplastic eyes. Methods: We report a case of a 48-year-old man who had congenital nystagmus with poor vision in both eyes, presented with decreased vision and photophobia in his left eye. An exudative retinal detachment was found, which did not respond to systemic steroid treatment. Glaucoma due to occlusio pupillae was later developed. Laser iridotomy and anti-glaucoma medications decreased intraocular pressure to an acceptable level. Vision in the left eye gradually deteriorated during the 10-year clinical course. Evisceration was finally performed due to persistent dull aching ocular pain along with signs of ocular hypotony Results: Histopathological examination showed phthisis bulbi and focal nodular retinal gliosis. The Azzopardi phenomenon was found at the retinal vessel walls, within the retinal layers and along the internal limiting membrane. There was neither evidence of intraocular tumors nor foreign bodies. Conclusion: This case demonstrated that the Azzopardi phenomenon could be present in a non-neoplastic eye with a longstanding disease that proceeds to phthisis bulbi.

A Two-Stage GAN for High-Resolution Retinal Image Generation and Segmentation

In this paper, we use Generative Adversarial Networks (GANs) to synthesize high-quality retinal images along with the corresponding semantic label-maps, instead of real images during training of a segmentation network. Different from other previous proposals, we employ a two-step approach: first, a progressively growing GAN is trained to generate the semantic label-maps, which describes the blood vessel structure (i.e., the vasculature); second, an image-to-image translation approach is used to obtain realistic retinal images from the generated vasculature. The adoption of a two-stage process simplifies the generation task, so that the network training requires fewer images with consequent lower memory usage. Moreover, learning is effective, and with only a handful of training samples, our approach generates realistic high-resolution images, which can be successfully used to enlarge small available datasets. Comparable results were obtained by employing only synthetic images in place of real data during training. The practical viability of the proposed approach was demonstrated on two well-established benchmark sets for retinal vessel segmentation—both containing a very small number of training samples—obtaining better performance with respect to state-of-the-art techniques.