Optic Disc Cupping Research Articles

Real Time Retinal Optic Disc Segmentation via Guided filter and Discrete Wavelet Transform

In the world, glaucoma is the one of the main causes for loosing vision. Optical Coherence Tomography or fundus camera is utilized to capture the optic disc and optic cup images. For the detection of glaucoma and afterwards monitoring the patients, investigation of the head of optic nerve or cup-to-disc ratio (CDR) is an important factor. For computing the CDR value, segmentation of optic disc and optic cup are utilized for the isolation of the relevant parts of the fundus image. Even though ophthalmologists are computing the CDR value physically, however, it limits the identification of glaucoma at the early stage. The accurate value of CDR is hard to find out if the optic cup and optic disc are not well defined properly. Thus, this paper has suggested a combined model of Guided filter and Discrete Wave Transform (DWT) for the enhancement and OD segmentation. Hysteresis thresholding is utilized for the optic cup segmentation. A set of 50 images of 25 patients are obtained from Visakha Eye Hospital, Visakhapatnam, India, is used to verify the performance of the suggested model. The recommended approach is also verified on the glaucoma subset of High-Resolution Fundus (HRF) database. The experimental results of both the databases demonstrate that the CDR value is computed accurately with low computational time.

Comparison of age-related vascular changes in the optic disc of patients with diabetes, with glaucomatous and non-glaucomatous features

ObjectiveTo identify age-related vascular changes in the optic discs of patients with diabetes with and without signs of glaucoma.Methods and analysisA total of 2153 eyes of 1797 patients with diabetes...

Pseudo-reversal of optic disc cupping in a case of pseudo-exfoliation glaucoma following trabeculectomy

Reversal of glaucomatous cupping is rare in adults. A case of pseudo-reversal of total cupping in an elderly gentleman with pseudo-exfoliation glaucoma who underwent combined trabeculectomy with cataract surgery is reported. Early laser suture lysis to control high post-operative intra-ocular pressure (IOP) led to sudden lowering of IOP with subsequent optic disc edema, which manifested as reversal of cupping, albeit temporarily. Eight months later, the visual field was stable, but the macula had developed an epiretinal membrane, probably secondary to hypotony-associated inflammation, thus highlighting the importance of timing the suture lysis and aggressive control of intra-ocular inflammation.

Assessment of Optic Nerve Changes in Patients Receiving Anti-Tuberculosis Drugs at Different Time Interval

Objectives: To assess optic nerve changes in patients receiving anti-tuberculosis drugs and to assess optic nerve changes at different time intervals of therapy. Study design and settings: Descriptive cross-sectional study was conducted from August 2021 to November 2021 in Madinah Teaching Hospital, Faisalabad. Methodology: 200 eyes of 100 patients receiving anti-tuberculosis therapy since 2-month, 4-month, 6 month and 8 months were taken. Patients did not have any systemic disease other than Tuberculosis. Indirect Ophthalmoscopy was performed to assess changes in fundus and optic nerve head. Optical Coherence Tomography was performed for assessment of retinal nerve fiber layer thickness and cup to disc ratio. Data was analysed by using descriptive statistics and chi-square test with SPSS version 20. Results: In 2 months there was 16(66.66%) normal fundus, 8(33.33%) glaucomatous optic disc cupping. In 4 months, there was 12(15.38%) normal fundus, 16(20%) optic atrophy, 16(20%) optic neuropathy, 4.00(5%) optic neuritis and 30(38%) glaucomatous optic disc cupping. In 6 months, there was 7(21%) normal fundus, 4(12.5%) optic atrophy and 21(50%) optic neuropathy. In 8 months, there was 24(36%) optic atrophy, 34(51.5%) optic neuropathy, 4(6%) optic neuritis and 4(6%) Glaucomatous optic dis cupping (P 0.00). Result of this study also shows that retinal nerve fiber layer thickness is normal in 83(41.5%), increase in 4(2.0%) and decrease in 113(56.5%). Conclusions: This study concluded that anti-tuberculosis drugs responsible for optic nerve changes and severity of optic nerve changes increase when duration of therapy increases

Unilateral advanced glaucoma in isolated congenital ectropion uveae with ipsilateral ptosis: A pictorial description of five children.

Aim: To report the cases of five children with unilateral advanced glaucoma in isolated congenital ectropion uveae (CEU) with ipsilateral ptosis and myopia. Methods: This is an ambispective observational case series. After diagnosing one patient with CEU and glaucoma, consecutive patients presenting with unilateral ptosis, congenital iris anomaly, and glaucoma between 2014 to 2020, and had completed a minimum one-year postoperative follow-up, were analyzed. Results: Of the 1421 newly registered pediatric glaucoma patients in the period under review, five children were diagnosed with CEU. All patients presented with gradual painless diminution of vision in the left eye in early adolescence. The left eye of all patients had peculiar clinical features: mild congenital ptosis, high iris insertion, crypt-less smooth iris surface, congenital ectropion uveae, pigments over anterior lens capsule, high myopia, advanced glaucomatous optic disc cupping, and very high intraocular pressure (IOP), which was > 45 mmHg in all cases. The right eye showed signs of angle dysgenesis with mild anterior iris insertion and numerous fine iris processes. Antiglaucoma medications and angle surgery failed to control the IOP, and all children required glaucoma filtration surgery, resulting in reasonable IOP control. Despite the older age, postoperative strict amblyopia treatment resulted in significant improvement in vision.Conclusions: Although ectropion uveae and ptosis have been present since birth, unilaterality, and the asymptomatic nature of the disease led to the late presentation with irreversible damage. Early surgical management and amblyopia therapy are the cornerstones of management.Abbreviations: CEU = Congenital ectropion uvea, CIES = Congenital Iris Ectropion Syndrome, ASD = Anterior segment dysgenesis syndrome, BCVA = Best-corrected visual acuity, IOP = Intraocular pressure

Optic Nerve Sheath Calcification

A 72-year-old man underwent computed tomography to evaluate upper airway congestion. Imaging revealed the incidental findings of calcification in each optic nerve sheath (Fig A, B) and the falx cerebri (Fig B, arrow). The patient had undergone a parathyroidectomy 5 years prior for primary hyperparathyroidism. His serum calcium levels subsequently remained within normal limits. Examination showed no evidence of optic nerve dysfunction, although there was asymmetric optic disc cupping (Fig C, D). Optic nerve sheath calcification may lead to optic atrophy with reduced visual acuity or remain asymptomatic, as in our patient (Magnified version of Fig A-D is available online at www.aaojournal.org).

Graph deep network for optic disc and optic cup segmentation for glaucoma disease using retinal imaging.

The fundus imaging method of eye screening detects eye diseases by segmenting the optic disc (OD) and optic cup (OC). OD and OC are still challenging to segment accurately. This work proposes three-layer graph-based deep architecture with an enhanced fusion method for OD and OC segmentation. CNN encoder-decoder architecture, extended graph network, and approximation via fusion-based rule are explored for connecting local and global information. A graph-based model is developed for combining local and overall knowledge. By extending feature masking, regularization of repetitive features with fusion for combining channels has been done. The performance of the proposed network is evaluated through the analysis of different metric parameters such as dice similarity coefficient (DSC), intersection of union (IOU), accuracy, specificity, sensitivity. Experimental verification of this methodology has been done using the four benchmarks publicly available datasets DRISHTI-GS, RIM-ONE for OD, and OC segmentation. In addition, DRIONS-DB and HRF fundus imaging datasets were analyzed for optimizing the model's performance based on OD segmentation. DSC metric of methodology achieved 0.97 and 0.96 for DRISHTI-GS and RIM-ONE, respectively. Similarly, IOU measures for DRISHTI-GS and RIM-ONE datasets were 0.96 and 0.93, respectively, for OD measurement. For OC segmentation, DSC and IOU were measured as 0.93 and 0.90 respectively for DRISHTI-GS and 0.83 and 0.82 for RIM-ONE data. The proposed technique improved value of metrics with most of the existing methods in terms of DSC and IOU of the results metric of the experiments for OD and OC segmentation.

Joint optic disk and cup segmentation for glaucoma screening using a region-based deep learning network.

To develop and validate an end-to-end region-based deep convolutional neural network (R-DCNN) to jointly segment the optic disc (OD) and optic cup (OC) in retinal fundus images for precise cup-to-disc ratio (CDR) measurement and glaucoma screening. In total, 2440 retinal fundus images were retrospectively obtained from 2033 participants. An R-DCNN was presented for joint OD and OC segmentation, where the OD and OC segmentation problems were formulated into object detection problems. We compared R-DCNN's segmentation performance on our in-house dataset with that of four ophthalmologists while performing quantitative, qualitative and generalization analyses on the publicly available both DRISHIT-GS and RIM-ONE v3 datasets. The Dice similarity coefficient (DC), Jaccard coefficient (JC), overlapping error (E), sensitivity (SE), specificity (SP) and area under the curve (AUC) were measured. On our in-house dataset, the proposed model achieved a 98.51% DC and a 97.07% JC for OD segmentation, and a 97.63% DC and a 95.39% JC for OC segmentation, achieving a performance level comparable to that of the ophthalmologists. On the DRISHTI-GS dataset, our approach achieved 97.23% and 94.17% results in DC and JC results for OD segmentation, respectively, while it achieved a 94.56% DC and an 89.92% JC for OC segmentation. Additionally, on the RIM-ONE v3 dataset, our model generated DC and JC values of 96.89% and 91.32% on the OD segmentation task, respectively, whereas the DC and JC values acquired for OC segmentation were 88.94% and 78.21%, respectively. The proposed approach achieved very encouraging performance on the OD and OC segmentation tasks, as well as in glaucoma screening. It has the potential to serve as a useful tool for computer-assisted glaucoma screening.

Accidentally diagnosed ocular abnormalities identified during close screening for retinopathy of prematurity.

To report the variety of ocular findings which have been identified serendipitously during the screening for retinopathy of prematurity (ROP) in a tertiary referral center during seven-year period. The charts of 1568 preterm infants who screened for ROP were reviewed retrospectively. Any ocular lesion except for ROP were noted. All infants had undergone routine ocular examination of the external eye, pupillary light reflex, anterior and posterior segment. Wide-angle digital retinal image acquisition system for any vitreoretinal pathology requiring a close follow-up had been utilized. Abnormal ocular findings other than ROP were diagnosed in 296 infants (19.2%). Tunica vasculosa lentis was the most common finding (25%) followed by vitreous or retinal hemorrhages (17.2%) and retinal white lesions (16.6%). Retina was the most frequently involved anatomic site. Other frequent ocular findings included optic disc cupping, congenital cataract, optic nerve hypoplasia, choroidal nevus, persistent fetal vasculature, lid hemangioma, and tilted disc. However, life-threatening pathologies such as lipemia retinalis and even retinoblastoma were also diagnosed. A duly ophthalmologic examination is mandatory in premature infants for ROP screening. During such examinations, ophthalmologists must be aware of coexisting ocular findings; which could be sight-threatening or even life-threatening.

Graph-Based Region and Boundary Aggregation for Biomedical Image Segmentation.

Segmentation is a fundamental task in biomedical image analysis. Unlike the existing region-based dense pixel classification methods or boundary-based polygon regression methods, we build a novel graph neural network (GNN) based deep learning framework with multiple graph reasoning modules to explicitly leverage both region and boundary features in an end-to-end manner. The mechanism extracts discriminative region and boundary features, referred to as initialized region and boundary node embeddings, using a proposed Attention Enhancement Module (AEM). The weighted links between cross-domain nodes (region and boundary feature domains) in each graph are defined in a data-dependent way, which retains both global and local cross-node relationships. The iterative message aggregation and node update mechanism can enhance the interaction between each graph reasoning module's global semantic information and local spatial characteristics. Our model, in particular, is capable of concurrently addressing region and boundary feature reasoning and aggregation at several different feature levels due to the proposed multi-level feature node embeddings in different parallel graph reasoning modules. Experiments on two types of challenging datasets demonstrate that our method outperforms state-of-the-art approaches for segmentation of polyps in colonoscopy images and of the optic disc and optic cup in colour fundus images. The trained models will be made available at: https://github.com/smallmax00/Graph_Region_Boudnary.

Machine Learning and Deep Learning Techniques for Optic Disc and Cup Segmentation - A Review.

BackgroundGlobally, glaucoma is the second leading cause of blindness. Detecting glaucoma in the early stages is essential to avoid disease complications, which lead to blindness. Thus, computer-aided diagnosis systems are powerful tools to overcome the shortage of glaucoma screening programs.MethodsA systematic search of public databases, including PubMed, Google Scholar, and other sources, was performed to identify relevant studies to overview the publicly available fundus image datasets used to train, validate, and test machine learning and deep learning methods. Additionally, existing machine learning and deep learning methods for optic cup and disc segmentation were surveyed and critically reviewed.ResultsEight fundus images datasets were publicly available with 15,445 images labeled with glaucoma or non-glaucoma, and manually annotated optic disc and cup boundaries were found. Five metrics were identified for evaluating the developed models. Finally, three main deep learning architectural designs were commonly used for optic disc and optic cup segmentation.ConclusionWe provided future research directions to formulate robust optic cup and disc segmentation systems. Deep learning can be utilized in clinical settings for this task. However, many challenges need to be addressed before using this strategy in clinical trials. Finally, two deep learning architectural designs have been widely adopted, such as U-net and its variants.

An Efficient Hierarchical Optic Disc and Cup Segmentation Network Combined with Multi-task Learning and Adversarial Learning.

Automatic and accurate segmentation of optic disc (OD) and optic cup (OC) in fundus images is a fundamental task in computer-aided ocular pathologies diagnosis. The complex structures, such as blood vessels and macular region, and the existence of lesions in fundus images bring great challenges to the segmentation task. Recently, the convolutional neural network-based methods have exhibited its potential in fundus image analysis. In this paper, we propose a cascaded two-stage network architecture for robust and accurate OD and OC segmentation in fundus images. In the first stage, the U-Net like framework with an improved attention mechanism and focal loss is proposed to detect accurate and reliable OD location from the full-scale resolution fundus images. Based on the outputs of the first stage, a refined segmentation network in the second stage that integrates multi-task framework and adversarial learning is further designed for OD and OC segmentation separately. The multi-task framework is conducted to predict the OD and OC masks by simultaneously estimating contours and distance maps as auxiliary tasks, which can guarantee the smoothness and shape of object in segmentation predictions. The adversarial learning technique is introduced to encourage the segmentation network to produce an output that is consistent with the true labels in space and shape distribution. We evaluate the performance of our method using two public retinal fundus image datasets (RIM-ONE-r3 and REFUGE). Extensive ablation studies and comparison experiments with existing methods demonstrate that our approach can produce competitive performance compared with state-of-the-art methods.

Dry Eyes and Glaucoma: A Report of Sixty Eyes

Introduction: Glaucoma is an optic neuropathy defined by optic disc cupping, characteristic visual field defects, and often an increase in intraocular pressure. It remains a leading cause of irreversible blindness, with an estimated 60.5 million people affected worldwide. It is a chronic disease that frequently requires long-term treatment with topical ocular hypotensive eyedrops. The use of antiglaucoma medications has been associated with dry eye. Another reason for the correlation between glaucoma and dry eyes, is that prevalence for both increases with age. Patients and methods: Sixty eyes of 30 patients with medically treated primary open-angle glaucoma (POAG) were included in this study. Dry eye syndrome was searched with Schirmer test (less than 10 mm in 5 minutes) and tear break-up time (BUT: less than 10 seconds). Results: Forty-eight eyes medically treated primary open-angle glaucoma (80%) present dry eye syndrome, this is consistent with previously published literature. All the POAG patients with dry eyes, were medically treated with preserved glaucoma eye drop. They were compliant to their treatment, thinking their vision is blurry because their glaucoma is getting worse and they’re dealing with what they think is irreversible. After solving the dry eye syndrome, we effect a control of the Visual Field test, we can say that dry eye’s treatment can significantly improve reliability and visual field index of glaucoma patients undergoing visual field examination. Conclusion: this study shows that dry eye syndrome associated with glaucoma might lead to ocular discomfort and low-quality retinal images which can decreased compliance of treatment, and deteriorate the Visual field test results. Protecting the integrity of the ocular surface while treating ocular conditions like glaucoma is necessary.

Segmentation of the optic disc and optic cup using a machine learning-based biregional contour evolution model for the cup-to-disc ratio

Segmentation of the optic disc and optic cup using a machine learning-based biregional contour evolution model for the cup-to-disc ratio

An Efficient Deep Learning Approach to Automatic Glaucoma Detection Using Optic Disc and Optic Cup Localization.

Glaucoma is an eye disease initiated due to excessive intraocular pressure inside it and caused complete sightlessness at its progressed stage. Whereas timely glaucoma screening-based treatment can save the patient from complete vision loss. Accurate screening procedures are dependent on the availability of human experts who performs the manual analysis of retinal samples to identify the glaucomatous-affected regions. However, due to complex glaucoma screening procedures and shortage of human resources, we often face delays which can increase the vision loss ratio around the globe. To cope with the challenges of manual systems, there is an urgent demand for designing an effective automated framework that can accurately identify the Optic Disc (OD) and Optic Cup (OC) lesions at the earliest stage. Efficient and effective identification and classification of glaucomatous regions is a complicated job due to the wide variations in the mass, shade, orientation, and shapes of lesions. Furthermore, the extensive similarity between the lesion and eye color further complicates the classification process. To overcome the aforementioned challenges, we have presented a Deep Learning (DL)-based approach namely EfficientDet-D0 with EfficientNet-B0 as the backbone. The presented framework comprises three steps for glaucoma localization and classification. Initially, the deep features from the suspected samples are computed with the EfficientNet-B0 feature extractor. Then, the Bi-directional Feature Pyramid Network (BiFPN) module of EfficientDet-D0 takes the computed features from the EfficientNet-B0 and performs the top-down and bottom-up keypoints fusion several times. In the last step, the resultant localized area containing glaucoma lesion with associated class is predicted. We have confirmed the robustness of our work by evaluating it on a challenging dataset namely an online retinal fundus image database for glaucoma analysis (ORIGA). Furthermore, we have performed cross-dataset validation on the High-Resolution Fundus (HRF), and Retinal Image database for Optic Nerve Evaluation (RIM ONE DL) datasets to show the generalization ability of our work. Both the numeric and visual evaluations confirm that EfficientDet-D0 outperforms the newest frameworks and is more proficient in glaucoma classification.

The review of optic disc and optic cup segmentation applications in computer-aided glaucoma diagnosis

The review of optic disc and optic cup segmentation applications in computer-aided glaucoma diagnosis

Optic Disc and Optic Cup Segmentation on Retinal Image Based on Multimap Localization and U-Net Convolutional Neural Network

Optic Disc and Optic Cup Segmentation on Retinal Image Based on Multimap Localization and U-Net Convolutional Neural Network

Hydrostatic Pressure Gradients and a New Membrane-glymphatic Theory of Primary Glaucoma

The paper describes a novel theory of primary glaucoma etiopathogenesis based on a new understanding of the glymphatic system of the eye. Glymphatic flow routes of the eye include the vitreous body, uveal tract, paravascular spaces, and the retina. There are different pressure gradients of the intraocular fluid flow within the divided central channel of the vitreous body. In normal conditions above the foveola there is highly positive pressure gradient whereas above the optic disc it is 3 times lower. The retinal pigment epithelium is responsible for water and metabolic exchanges between choroid and the vitreous, therefore may contribute to intra-ocular fluid accumulation. Impairment of the outer blood-ocular and/or blood-aqueous barriers is the key pathogenic element of glaucoma including secondary forms due to uveitis. A disbalance of the pressure gradients above the macula and above the optic disc leads to fluid accumulation above the latter. An abundant intra-retinal fluid flow through the glymphatic pathways in the outer and inner plexiform layers, formed by Muller cell processes, contributes to accumulation of the interstitial fluid before the lamina cribrosa. Excessive interstitial fluid from the retina passes not only through the retinal pigment epithelium to choroid but also through paravascular spaces to the optic disc cup and discharges partly back to the vitreous body through the prevascular vitreous fissures. There is an interstitial fluid flow existing through paravascular spaces, the plexiform layers of retina and along the axons of ganglion cells to the optic nerve and its sheath where excessive fluid is absorbed into the subarachnoid space. The RPE plays a major role in glaucoma etiopahogenesis, probably acting as a regulator of fluid transport in the eye.

Optic disc cupping after circumpapillary Pd-103 slotted plaque radiation therapy

ObjectifExaminer les effets vasculaires et morphologiques sur le disque optique après une radiothérapie par plaque fendue dans le traitement d'un mélanome choroïdien touchant le disque optique. NatureÉtude transversale rétrospective. ParticipantsTrente-neuf patients atteints d'un mélanome choroïdien touchant le nerf optique. MéthodesChaque mélanome a été traité au moyen d'une plaque radioactive fendue (palladium-103) incluant et/ou entourant la gaine du nerf optique entre 2005 et 2019. Les images du nerf optique avant et après la radiothérapie ont permis de documenter et d’évaluer les changements au chapitre de la pâleur du nerf optique et du rapport cupule/disque (C/D). On a mesuré le rapport C/D à la tomographie par cohérence optique (OCT) de même que la pression intraoculaire (PIO) avant le traitement et pendant le suivi, et 22 patients ont également fait l'objet d'une angiographie-tomographie par cohérence optique (OCT-A) dont la qualité des images permettait l’évaluation de la densité et de la longueur des vaisseaux sanguins. On a noté une corrélation entre les différences quant aux mesures du rapport C/D et les modifications de la densité et de la longueur des vaisseaux sanguins visibles à l'OCT-A. RésultatsSuivant une radiothérapie par plaque fendue, on n'a pas observé de hausse significative de la PIO ni d'augmentation de la pâleur du nerf optique. Les images OCT et les photographies couleur ont mis au jour une augmentation significative (p < 0,001 pour les 2 techniques) du rapport C/D à la fin du suivi, comparativement à ce qui avait été mesuré avant le traitement. On notait une corrélation significative entre la hausse du rapport C/D selon l'OCT et les variations de la longueur des vaisseaux sanguins mesurées à l'OCT-A (p = 0,027). De même, on a établi une corrélation significative entre une augmentation du rapport C/D dans les photographies du fond d’œil et les variations de la densité (p = 0,043) et de la longueur (p = 0,019) des vaisseaux sanguins mesurées à l'OCT-A. ConclusionLes photographies du fond d’œil et les images OCT ont mis en lumière une augmentation du phénomène d'excavation du disque optique au décours d'une radiothérapie par plaque fendue. L'excavation était associée à une occlusion et à une atténuation vasculaires progressives synchrones de la zone péripapillaire, visibles à l'OCT-A. On en conclut que l'ischémie péripapillaire et papillaire due à la radiothérapie par plaque fendue s'accompagne d'une hausse du rapport C/D et d'une aggravation de l'excavation du disque optique.

S-CUDA: Self-cleansing unsupervised domain adaptation for medical image segmentation.

Medical image segmentation tasks hitherto have achieved excellent progresses with large-scale datasets, which empowers us to train potent deep convolutional neural networks (DCNNs). However, labeling such large-scale datasets is laborious and error-prone, which leads the noisy (or incorrect) labels to be an ubiquitous problem in the real-world scenarios. In addition, data collected from different sites usually exhibit significant data distribution shift (or domain shift). As a result, noisy label and domain shift become two common problems in medical imaging application scenarios, especially in medical image segmentation, which degrade the performance of deep learning models significantly. In this paper, we identify a novel problem hidden in medical image segmentation, which is unsupervised domain adaptation on noisy labeled data, and propose a novel algorithm named "Self-Cleansing Unsupervised Domain Adaptation" (S-CDUA) to address such issue. S-CUDA sets up a realistic scenario to solve the above problems simultaneously where training data (i.e., source domain) not only shows domain shift w.r.t. unsupervised test data (i.e., target domain) but also contains noisy labels. The key idea of S-CUDA is to learn noise-excluding and domain invariant knowledge from noisy supervised data, which will be applied on the highly corrupted data for label cleansing and further data-recycling, as well as on the test data with domain shift for supervised propagation. To this end, we propose a novel framework leveraging noisy-label learning and domain adaptation techniques to cleanse the noisy labels and learn from trustable clean samples, thus enabling robust adaptation and prediction on the target domain. Specifically, we train two peer adversarial networks to identify high-confidence clean data and exchange them in companions to eliminate the error accumulation problem and narrow the domain gap simultaneously. In the meantime, the high-confidence noisy data are detected and cleansed in order to reuse the contaminated training data. Therefore, our proposed method can not only cleanse the noisy labels in the training set but also take full advantage of the existing noisy data to update the parameters of the network. For evaluation, we conduct experiments on two popular datasets (REFUGE and Drishti-GS) for optic disc (OD) and optic cup (OC) segmentation, and on another public multi-vendor dataset for spinal cord gray matter (SCGM) segmentation. Experimental results show that our proposed method can cleanse noisy labels efficiently and obtain a model with better generalization performance at the same time, which outperforms previous state-of-the-art methods by large margin. Our code can be found at https://github.com/zzdxjtu/S-cuda.