Automated techniques for blood vessels segmentation through fundus retinal images: A review

High-precision retinal blood vessel segmentation based on a multi-stage and dual-channel deep learning network

Objective.The high-precision segmentation of retinal vessels in fundus images is important for the early diagnosis of ophthalmic diseases. However, the extraction for microvessels is challenging due to their characteristics of low contrast and high structural complexity. Although some works have been developed to improve the segmentation ability in thin vessels, they have only been successful in recognizing small vessels with relatively high contrast.Approach.Therefore, we develop a deep learning (DL) framework with a multi-stage and dual-channel network model (MSDC_NET) to further improve the thin-vessel segmentation with low contrast. Specifically, an adaptive image enhancement strategy combining multiple preprocessing and the DL method is firstly proposed to elevate the contrast of thin vessels; then, a two-channel model with multi-scale perception is developed to implement whole- and thin-vessel segmentation; and finally, a series of post-processing operations are designed to extract more small vessels in the predicted maps from thin-vessel channels.Main results.Experiments on DRIVE, STARE and CHASE_DB1 demonstrate the superiorities of the proposed MSDC_NET in extracting more thin vessels in fundus images, and quantitative evaluations on several parameters based on the advanced ground truth further verify the advantages of our proposed DL model. Compared with the previous multi-branch method, the specificity and F1score are improved by about 2.18%, 0.68%, 1.73% and 2.91%, 0.24%, 8.38% on the three datasets, respectively.Significance.This work may provide richer information to ophthalmologists for the diagnosis and treatment of vascular-related ophthalmic diseases.

A Global Review of Publicly Available Datasets Containing Fundus Images: Characteristics, Barriers to Access, Usability, and Generalizability

This article provides a comprehensive and up-to-date overview of the repositories that contain color fundus images. We analyzed them regarding availability and legality, presented the datasets' characteristics, and identified labeled and unlabeled image sets. This study aimed to complete all publicly available color fundus image datasets to create a central catalog of available color fundus image datasets.

A Deep-Learning-Based Collaborative Edge-Cloud Telemedicine System for Retinopathy of Prematurity

Retinopathy of prematurity is an ophthalmic disease with a very high blindness rate. With its increasing incidence year by year, its timely diagnosis and treatment are of great significance. Due to the lack of timely and effective fundus screening for premature infants in remote areas, leading to an aggravation of the disease and even blindness, in this paper, a deep learning-based collaborative edge-cloud telemedicine system is proposed to mitigate this issue. In the proposed system, deep learning algorithms are mainly used for classification of processed images. Our algorithm is based on ResNet101 and uses undersampling and resampling to improve the data imbalance problem in the field of medical image processing. Artificial intelligence algorithms are combined with a collaborative edge-cloud architecture to implement a comprehensive telemedicine system to realize timely screening and diagnosis of retinopathy of prematurity in remote areas with shortages or a complete lack of expert medical staff. Finally, the algorithm is successfully embedded in a mobile terminal device and deployed through the support of a core hospital of Guangdong Province. The results show that we achieved 75% ACC and 60% AUC. This research is of great significance for the development of telemedicine systems and aims to mitigate the lack of medical resources and their uneven distribution in rural areas.

Prevalence of Monosodium Urate (MSU) Deposits in Cadavers Detected by Dual-Energy Computed Tomography (DECT)

Background: Dual-energy computed tomography (DECT) allows direct visualization of monosodium urate (MSU) deposits in joints and soft tissues. Purpose: To describe the distribution of MSU deposits in cadavers using DECT in the head, body trunk, and feet. Materials and Methods: A total of 49 cadavers (41 embalmed and 8 fresh cadavers; 20 male, 29 female; mean age, 79.5 years; SD ± 11.3; range 52–95) of unknown clinical history underwent DECT to assess MSU deposits in the head, body trunk, and feet. Lens, thoracic aorta, and foot tendon dissections of fresh cadavers were used to verify MSU deposits by polarizing light microscopy. Results: 33/41 embalmed cadavers (80.5%) showed MSU deposits within the thoracic aorta. 11/41 cadavers (26.8%) showed MSU deposits within the metatarsophalangeal (MTP) joints and 46.3% of cadavers demonstrated MSU deposits within foot tendons, larger than and equal to 5 mm. No MSU deposits were detected in the cranium/intracerebral vessels, or the coronary arteries. Microscopy used as a gold standard could verify the presence of MSU deposits within the lens, thoracic aorta, or foot tendons in eight fresh cadavers. Conclusions: Microscopy confirmed the presence of MSU deposits in fresh cadavers within the lens, thoracic aorta, and foot tendons, whereas no MSU deposits could be detected in cranium/intracerebral vessels or coronary arteries. DECT may offer great potential as a screening tool to detect MSU deposits and measure the total uric acid burden in the body. The clinical impact of this cadaver study in terms of assessment of MSU burden should be further proven.

DBFU-Net: Double branch fusion U-Net with hard example weighting train strategy to segment retinal vessel

BACKGROUND

Many fundus imaging modalities measure ocular changes. Automatic retinal vessel segmentation (RVS) is a significant fundus image-based method for the diagnosis of ophthalmologic diseases. However, precise vessel segmentation is a challenging task when detecting micro-changes in fundus images, e.g., tiny vessels, vessel edges, vessel lesions and optic disc edges.

METHODS

In this paper, we will introduce a novel double branch fusion U-Net model that allows one of the branches to be trained by a weighting scheme that emphasizes harder examples to improve the overall segmentation performance. A new mask, we call a hard example mask, is needed for those examples that include a weighting strategy that is different from other methods. The method we propose extracts the hard example mask by morphology, meaning that the hard example mask does not need any rough segmentation model. To alleviate overfitting, we propose a random channel attention mechanism that is better than the drop-out method or the L2-regularization method in RVS.

RESULTS

We have verified the proposed approach on the DRIVE, STARE and CHASE datasets to quantify the performance metrics. Compared to other existing approaches, using those dataset platforms, the proposed approach has competitive performance metrics. (DRIVE: F1-Score = 0.8289, G-Mean = 0.8995, AUC = 0.9811; STARE: F1-Score = 0.8501, G-Mean = 0.9198, AUC = 0.9892; CHASE: F1-Score = 0.8375, G-Mean = 0.9138, AUC = 0.9879).

DISCUSSION

The segmentation results showed that DBFU-Net with RCA achieves competitive performance in three RVS datasets. Additionally, the proposed morphological-based extraction method for hard examples can reduce the computational cost. Finally, the random channel attention mechanism proposed in this paper has proven to be more effective than other regularization methods in the RVS task.

Detection of microscopic glaucoma through fundus images using deep transfer learning approach

Glaucoma disease in humans can lead to blindness if it progresses to the point where it affects the oculus' optic nerve head. It is not easily detected since there are no symptoms, but it can be detected using tonometry, ophthalmoscopy, and perimeter. However, advances in artificial intelligence approaches have permitted machine learning techniques to diagnose at an early stage. Numerous methods have been proposed using Machine Learning to diagnose glaucoma with different data sets and techniques but these are complex methods. Although, medical imaging instruments are used as glaucoma screening methods, fundus imaging specifically is the most used screening technique for glaucoma detection. This study presents a novel DenseNet and DarkNet combination to classify normal and glaucoma affected fundus image. These frameworks have been trained and tested on three data sets of high-resolution fundus (HRF), RIM 1, and ACRIMA. A total of 658 images have been used for healthy eyes and 612 images for glaucoma-affected eyes classification. It has also been observed that the fusion of DenseNet and DarkNet outperforms the two CNN networks and achieved 99.7% accuracy, 98.9% sensitivity, 100% specificity for the HRF database. In contrast, for the RIM1 database, 89.3% accuracy, 93.3% sensitivity, 88.46% specificity has been attained. Moreover, for the ACRIMA database, 99% accuracy, 100% sensitivity, 99% specificity has been achieved. Therefore, the proposed method is robust and efficient with less computational time and complexity compared to the literature available.

Macular Edema and Visual Acuity Observation after Cataract Surgery in Patients with Diabetic Retinopathy

Objective. The purpose was to explore the effect of cataract surgery on postoperative macular edema and visual acuity in patients with diabetic retinopathy. Methods. 88 patients with diabetic retinopathy treated in our hospital (December 2019–December 2020) were chosen as research subjects and divided into experimental group of 44 patients (52 eyes) and control group of 44 patients (54 eyes) according to the odd and even admission numbers. The control group received laser photocoagulation treatment, while the experimental group underwent cataract surgery. The central macular thickness (CMT) and visual acuity of the two groups after treatment were detected to evaluate the therapeutic effect of different treatment methods on diabetic retinopathy. Results. No obvious differences in sex ratio, average age, average course of disease, average weight, average BMI, average glycosylated hemoglobin, and residence were found between the two groups $\left(P>0.05\right)$ . The total clinical effective rate in the experimental group was obviously higher compared with the control group $\left(P<0.05\right)$ . The CMT at T1, T2, and T3 in the experimental group was obviously lower compared with the control group $\left(P<0.05\right)$ . The BCVA in the experimental group at 1 month and 3 months after treatment was obviously higher compared with the control group $\left(P<0.05\right)$ . The VEGF levels of both groups after treatment were obviously lower $\left(P<0.001\right)$ , and the VEGF level in the experimental group after treatment was obviously lower compared with the control group $\left(P<0.001\right)$ . The total incidence of complications in the experimental group was obviously lower compared with the control group $\left(P<0.05\right)$ . Conclusion. Cataract surgery is a reliable method to improve visual acuity and reduce serum inflammatory indicators in patients with diabetic retinopathy, with better clinical effect than laser photocoagulation, which is recommended for the treatment of diabetic retinopathy.

Microscopic retinal blood vessels detection and segmentation using support vector machine and K-nearest neighbors

The retina is the deepest layer of texture covering the rear of the eye, recorded by fundus images. Vessel detection and segmentation are useful in disease diagnosis. The retina's blood vessels could help diagnose maladies such as glaucoma, diabetic retinopathy, and blood pressure. A mix of supervised and unsupervised strategies exists for the detection and segmentation of blood vessels images. The tree structure of retinal blood vessels, their random area, and different thickness have caused vessel detection difficulties at machine learning calculations. Since the green band of retinal images conveys more information about the vessels, they are utilized for microscopic vessels detection. The current research proposes an administered calculation for segmentation of retinal vessels, where two upgrading stages depending on filtering and comparative histogram were applied after pre-processing and image quality improvement. At that point, statistical features of vessel tracking, maximum curvature and curvelet coefficient are extracted for each pixel. The extracted features are classified by support vector machine and the k-nearest neighbors. The morphological operators then enhance the classified image at the final stage to segment with higher accuracy. The dice coefficient is utilized for the evaluation of the proposed method. The proposed approach is concluded to be better than different strategies with a normal of 92%.

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.

Construction and application of color fundus image segmentation algorithm based on Multi-Scale local combined global enhancement

Objective

Aiming at the problem of low accuracy in extracting small blood vessels from existing retinal blood vessel images, a retinal blood vessel segmentation method based on a combination of a multi-scale linear detector and local and global enhancement is proposed.

Methods

The multi-scale line detector is studied, and it is divided into two parts: small scale and large scale. The small scale is used to detect the locally enhanced image and the large scale is used to detect the globally enhanced image. Fusion the response functions at different scales to get the final retinal vascular structure.

Results

Experiments on two databases STARE and DRIVE, show that the average vascular accuracy rates obtained by the algorithm reach 96.62% and 96.45%, and the average true positive rates reach 75.52% and 83.07%, respectively.

Conclusion

The segmentation accuracy is high, and better blood vessel segmentation results can be obtained.

Automatic detection of papilledema through fundus retinal images using deep learning

Papilledema is a syndrome of the retina in which retinal optic nerve is inflated by elevation of intracranial pressure. The papilledema abnormalities such as retinal nerve fiber layer (RNFL) opacification may lead to blindness. These abnormalities could be seen through capturing of retinal images by means of fundus camera. This paper presents a deep learning-based automated system that detects and grades the papilledema through U-Net and Dense-Net architectures. The proposed approach has two main stages. First, optic disc and its surrounding area in fundus retinal image are localized and cropped for input to Dense-Net which classifies the optic disc as papilledema or normal. Second, consists of preprocessing of Dense-Net classified papilledema fundus image by Gabor filter. The preprocessed papilledema image is input to U-Net to achieve the segmented vascular network from which the vessel discontinuity index (VDI) and vessel discontinuity index to disc proximity (VDIP) are calculated for grading of papilledema. The VDI and VDIP are standard parameter to check the severity and grading of papilledema. The proposed system is evaluated on 60 papilledema and 40 normal fundus images taken from STARE dataset. The experimental results for classification of papilledema through Dense-Net are much better in terms of sensitivity 98.63%, specificity 97.83%, and accuracy 99.17%. Similarly, the grading results for mild and severe papilledema classification through U-Net are also much better in terms of sensitivity 99.82%, specificity 98.65%, and accuracy 99.89%. The deep learning-based automated detection and grading of papilledema for clinical purposes is first effort in state of art.

Impact of illumination spectrum and eye pigmentation on image quality from a fundus camera using transscleral illumination

SIGNIFICANCE

The use of the transscleral illumination approach has the potential to simplify the optical design of fundus cameras. In particular, this approach could allow the use of smaller and cheaper cameras that are easier to use by non-specialists, thereby facilitating a wider spread of eye disease screening programs.

AIM

Our aim was to investigate the suitability of transscleral illumination in a fundus camera system. In particular, we explored the impact of the illumination spectrum and the eye pigmentation on the quality of the image. These factors have never been systematically investigated before in the literature on transscleral illumination.

APPROACH

A fundus camera was constructed using transscleral illumination. We studied the influence of eye pigmentation and choice of illumination spectra on the image quality for a group of 10 individuals with varied skin pigmentation, ranging from pale white (North-European) to darkest brown (African). The influence of the light source spectrum on the image quality was assessed using wavelength filters.

RESULTS

There was a difference of a factor of 100 in the signal level of retinal images between individuals with low and high skin pigmentation. The image contrast was highest using illumination wavelengths of 500 to 600 nm. The illumination level can be adjusted to obtain high-quality images for highly pigmented eyes while keeping the system eye-safe.

CONCLUSIONS

We have demonstrated that a fundus camera with transscleral illumination can provide high-quality images. However, the variations observed in scleral and retinal pigmentation in a practical setting require a system that must be able to adapt illumination and/or exposure to the individual patient.

Identification of cardiovascular high risk groups from dynamic retinal vessel signals using untargeted machine learning

AIMS

Dynamic retinal vessel analysis (DVA) provides a noninvasive way to asses microvascular function in patients and potentially to improve predictions of individual cardiovascular (CV) risk. The aim of our study was to use untargeted machine learning on DVA in order to improve CV mortality prediction and to identify corresponding response alterations.

METHODS AND RESULTS

We adopted a workflow consisting of noise reduction and extraction of independent components within DVA signals. Predictor performance was assessed in survival random forest models. Applying our technique to the prediction of all-cause mortality in a cohort of 214 hemodialysis patients resulted in the selection of a component which was highly correlated to maximal venous dilation following flicker stimulation (vMax), a previously identified predictor, confirming the validity of our approach. When fitting for CV mortality as the outcome of interest, a combination of three components derived from the arterial signal resulted in a marked improvement in predictive performance. Clustering analysis suggested that these independent components identified groups of patients with substantially higher CV mortality.

CONCLUSIONS

Our results provide a machine learning workflow to improve the predictive performance of DVA and identify groups of hemodialysis patients at high risk of CV mortality. Our approach may also prove to be promising for DVA signal analysis in other CV disease states.

TRANSLATIONAL PERSPECTIVE

DVA is a noninvasive technique which can help to assess microvascular dysfunction, one of the driving factors of CV disease. This study demonstrates a machine learning method which improves DVA interpretation for the estimation of CV risk in hemodialysis patients. Similar techniques can help doctors to identify high risk patients for timely therapeutic interventions and to monitor the effects of these interventions.

Retinal vessel changes in cerebrovascular disease

PURPOSE OF REVIEW

The retina is growingly recognized as a window into cerebrovascular and systemic vascular conditions. The utility of noninvasive retinal vessel biomarkers in cerebrovascular risk assessment has expanded due to advances in retinal imaging techniques and machine learning-based digital analysis. The purpose of this review is to underscore the latest evidence linking retinal vascular abnormalities with stroke and vascular-related cognitive disorders; to highlight modern developments in retinal vascular imaging modalities and software-based vasculopathy quantification.

RECENT FINDINGS

Longitudinal studies undertaken for extended periods indicate that retinal vascular changes can predict cerebrovascular disorders (CVD). Cerebrovascular ties to dementia provoked recent explorations of retinal vessel imaging tools for conceivable early cognitive decline detection. Innovative biomedical engineering technologies and advanced dynamic and functional retinal vascular imaging methods have recently been added to the armamentarium, allowing an unbiased and comprehensive analysis of the retinal vasculature. Improved artificial intelligence-based deep learning algorithms have boosted the application of retinal imaging as a clinical and research tool to screen, risk stratify, and monitor with precision CVD and vascular cognitive impairment.

SUMMARY

Mounting evidence supports the use of quantitative retinal vessel analysis in predicting CVD, from clinical stroke to neuroimaging markers of stroke and neurodegeneration.

Retinal findings in patients with COVID-19: Results from the SERPICO-19 study

BACKGROUND

Coronavirus disease 2019 (COVID-19) has been associated to microvascular alterations. We screened the fundus of patients with COVID-19 to detect alterations of the retina and its vasculature and to assess possible correlations with clinical parameters.

METHODS

Cross-sectional study. The presence of retinal alterations in patients with COVID-19 and subjects unexposed to the virus was assessed using fundus photographs and their prevalence was compared. Mean arteries diameter (MAD) and mean veins diameter (MVD) were compared between patients and unexposed subjects with multiple linear regression including age, sex, ethnicity, body mass index, smoking/alcohol consumption, hypertension, hyperlipidaemia, diabetes as covariates. The influence of clinical/lab parameters on retinal findings was tested in COVID-19 patients.

FINDINGS

54 patients and 133 unexposed subjects were enrolled. Retinal findings in COVID-19 included: haemorrhages (9·25%), cotton wools spots (7·4%), dilated veins (27·7%), tortuous vessels (12·9%). Both MAD and MVD were higher in COVID-19 patients compared to unexposed subjects (98·3 ± 15·3 µm vs 91·9 ± 11·7 µm, p = 0.006 and 138·5 ± 21·5 µm vs 123·2 ± 13·0 µm, p<0.0001, respectively). In multiple regression accounting for covariates MVD was positively associated with COVID-19 both in severe (coefficient 30·3, CI95% 18·1-42·4) and non-severe (coefficient 10·3, CI95% 1·6-19·0) cases compared to unexposed subjects. In COVID-19 patients MVD was negatively correlated with the time from symptoms onset (coefficient -1·0, CI 95% -1·89 to -0·20) and positively correlated with disease severity (coefficient 22·0, CI 95% 5·2-38·9).

INTERPRETATION

COVID-19 can affect the retina. Retinal veins diameter seems directly correlated with the disease severity. Its assessment could have possible applications in the management of COVID-19.

FUNDING

None.

Diabetic retinopathy detection through novel tetragonal local octa patterns and extreme learning machines

Diabetic retinopathy (DR) is an eye disease that victimize the people suffering from diabetes from many years. The severe form of DR results in form of the blindness that can initially be controlled by the DR-screening oriented treatment. The effective screening programs require the trained human resource that manually grade the fundus images to understand the severity of the disease. But due to the complexity of this process, and the insufficient number of the trained workers, the precise manual grading is an expensive process. The CAD-based solutions try to address these limitations but most of the existing DR detection systems are as evaluated over small sets and become ineffective when applied in real scenarios. Therefore, in this paper we proposed a novel technique to precisely detect the various stages of the DR by extending the research of the content-based image retrieval domain. To achieve the human-level performance over the large-scale DR-datasets (i.e. Kaggle-DR), the fundus images are represented by the novel tetragonal local octa pattern (T-LOP) features, that are then classified through the extreme learning machine (ELM). To justify the significance of the method, the proposed scheme is compared against several state-of-the-art methods including the deep learning-based methods over four DR-datasets of variational lengths (i.e. Kaggle-DR, DRIVE, Review-DB, STARE). The experimental results confirm the significance of the DR-detection scheme to serve as a stand-alone solution for providing the precise information of the severity of the DR in an efficient manner.