Automated segmentation of the optic disc from stereo color photographs using physiologically plausible features

Quadratic divergence regularized SVM for optic disc segmentation

Machine learning has been used in many retinal image processing applications such as optic disc segmentation. It assumes that the training and testing data sets have the same feature distribution. However, retinal images are often collected under different conditions and may have different feature distributions. Therefore, the models trained from one data set may not work well for another data set. However, it is often too expensive and time consuming to label the needed training data and rebuild the models for all different data sets. In this paper, we propose a novel quadratic divergence regularized support vector machine (QDSVM) to transfer the knowledge from domains with sufficient training data to domains with limited or even no training data. The proposed method simultaneously minimizes the distribution difference between the source domain and target domain while training the classifier. Experimental results show that the proposed transfer learning based method reduces the classification error in superpixel level from 14.2% without transfer learning to 2.4% with transfer learning. The proposed method is effective to transfer the label knowledge from source to target domain, which enables it to be used for optic disc segmentation in data sets with different feature distributions.

Automated Axon Counting in Rodent Optic Nerve Sections with AxonJ

We have developed a publicly available tool, AxonJ, which quantifies the axons in optic nerve sections of rodents stained with paraphenylenediamine (PPD). In this study, we compare AxonJ's performance to human experts on 100x and 40x images of optic nerve sections obtained from multiple strains of mice, including mice with defects relevant to glaucoma. AxonJ produced reliable axon counts with high sensitivity of 0.959 and high precision of 0.907, high repeatability of 0.95 when compared to a gold-standard of manual assessments and high correlation of 0.882 to the glaucoma damage staging of a previously published dataset. AxonJ allows analyses that are quantitative, consistent, fully-automated, parameter-free, and rapid on whole optic nerve sections at 40x. As a freely available ImageJ plugin that requires no highly specialized equipment to utilize, AxonJ represents a powerful new community resource augmenting studies of the optic nerve using mice.

Software-Assisted Depth Analysis of Optic Nerve Stereoscopic Images in Telemedicine

Background. Software guided optic nerve assessment can assist in process automation and reduce interobserver disagreement. We tested depth analysis software (DAS) in assessing optic nerve cup-to-disc ratio (VCD) from stereoscopic optic nerve images (SONI) of normal eyes. Methods. In a prospective study, simultaneous SONI from normal subjects were collected during telemedicine screenings using a Kowa 3Wx nonmydriatic simultaneous stereoscopic retinal camera (Tokyo, Japan). VCD was determined from SONI pairs and proprietary pixel DAS (Kowa Inc., Tokyo, Japan) after disc and cup contour line placement. A nonstereoscopic VCD was determined using the right channel of a stereo pair. Mean, standard deviation, t-test, and the intraclass correlation coefficient (ICCC) were calculated. Results. 32 patients had mean age of 40 ± 14 years. Mean VCD on SONI was 0.36 ± 0.09, with DAS 0.38 ± 0.08, and with nonstereoscopic 0.29 ± 0.12. The difference between stereoscopic and DAS assisted was not significant (p = 0.45). ICCC showed agreement between stereoscopic and software VCD assessment. Mean VCD difference was significant between nonstereoscopic and stereoscopic (p < 0.05) and nonstereoscopic and DAS (p < 0.005) recordings. Conclusions. DAS successfully assessed SONI and showed a high degree of correlation to physician-determined stereoscopic VCD.

Glaucoma classification using Regional Wavelet Features of the ONH and its surroundings

Glaucoma is one of the leading cause of blindness but the detection at its earliest stage and subsequent treatment can aid patients to preserve blindness. The existing work has been focusing on global features such as texture, grayscale and wavelet energy of the Optic Nerve Head (ONH) and its surrounding to differentiate between normal and glaucoma images. In contrast to previous approaches which focus on global information only, this work proposes a new approach to automatically classify between the normal and glaucoma images based on Regional Wavelet Features of the ONH and different regions around it. These regions are usually used for diagnosis of glaucoma by clinicians through visual observation only. Our method automatically determines different clinically observed regions around the ONH and performs classification on the basis of wavelet energy at different frequency subbands. We have conducted experiments based on different global features and regional features respectively and applied it to RIMONE (An Open Retinal Image Database for Optic Nerve Evaluation) database with 158 images. The experimental evaluation demonstrated that the classification accuracy of normal and glaucoma images using Regional Wavelet Features of different regions with 93% outperforms all other feature sets.

An adaptive threshold based image processing technique for improved glaucoma detection and classification

Glaucoma is an optic neuropathy which is one of the main causes of permanent blindness worldwide. This paper presents an automatic image processing based method for detection of glaucoma from the digital fundus images. In this proposed work, the discriminatory parameters of glaucoma infection, such as cup to disc ratio (CDR), neuro retinal rim (NRR) area and blood vessels in different regions of the optic disc has been used as features and fed as inputs to learning algorithms for glaucoma diagnosis. These features which have discriminatory changes with the occurrence of glaucoma are strategically used for training the classifiers to improve the accuracy of identification. The segmentation of optic disc and cup based on adaptive threshold of the pixel intensities lying in the optic nerve head region. Unlike existing methods the proposed algorithm is based on an adaptive threshold that uses local features from the fundus image for segmentation of optic cup and optic disc making it invariant to the quality of the image and noise content which may find wider acceptability. The experimental results indicate that such features are more significant in comparison to the statistical or textural features as considered in existing works. The proposed work achieves an accuracy of 94.11% with a sensitivity of 100%. A comparison of the proposed work with the existing methods indicates that the proposed approach has improved accuracy of classification glaucoma from a digital fundus which may be considered clinically significant.

Local patch reconstruction framework for optic cup localization in glaucoma detection

Optic cup localization/segmentation has attracted much attention from medical imaging researchers, since it is the primary image component clinically used for identifying glaucoma, which is a leading cause of blindness. In this work, we present an optic cup localization framework based on local patch reconstruction, motivated by the great success achieved by reconstruction approaches in many computer vision applications recently. Two types of local patches, i.e. grids and superpixels are used to show the variety, generalization ability and robustness of the proposed framework. Tested on the ORIGA clinical dataset, which comprises of 325 fundus images from a population-based study, both implementations under the proposed frameworks achieved higher accuracy than the state-of-the-art techniques.

Multimodal Segmentation of Optic Disc and Cup From SD-OCT and Color Fundus Photographs Using a Machine-Learning Graph-Based Approach

In this work, a multimodal approach is proposed to use the complementary information from fundus photographs and spectral domain optical coherence tomography (SD-OCT) volumes in order to segment the optic disc and cup boundaries. The problem is formulated as an optimization problem where the optimal solution is obtained using a machine-learning theoretical graph-based method. In particular, first the fundus photograph is registered to the 2D projection of the SD-OCT volume. Three in-region cost functions are designed using a random forest classifier corresponding to three regions of cup, rim, and background. Next, the volumes are resampled to create radial scans in which the Bruch's Membrane Opening (BMO) endpoints are easier to detect. Similar to in-region cost function design, the disc-boundary cost function is designed using a random forest classifier for which the features are created by applying the Haar Stationary Wavelet Transform (SWT) to the radial projection image. A multisurface graph-based approach utilizes the in-region and disc-boundary cost images to segment the boundaries of optic disc and cup under feasibility constraints. The approach is evaluated on 25 multimodal image pairs from 25 subjects in a leave-one-out fashion (by subject). The performances of the graph-theoretic approach using three sets of cost functions are compared: 1) using unimodal (OCT only) in-region costs, 2) using multimodal in-region costs, and 3) using multimodal in-region and disc-boundary costs. Results show that the multimodal approaches outperform the unimodal approach in segmenting the optic disc and cup.

Self-assessment for optic disc segmentation

Optic disc segmentation from retinal fundus image is a fundamental but important step in many applications such as automated glaucoma diagnosis. Very often, one method might work well on many images but fail on some other images and it is difficult to have a single method or model to cover all scenarios. Therefore, it is important to combine results from several methods to minimize the risk of failure. For this purpose, this paper computes confidence scores for three methods and combine their results for an optimal one. The experimental results show that the combined result from three methods is better than the results by any individual method. It reduces the mean overlapping error by 7.4% relatively compared with best individual method. Simultaneously, the number of failed cases with large overlapping errors is also greatly reduced. This is important to enhance the clinical deployment of the automated disc segmentation.

Comparison of Retinal Thickness Measurements between the Topcon Algorithm and a Graph-Based Algorithm in Normal and Glaucoma Eyes

PURPOSE

To assess the correlation and agreement between the Topcon built-in algorithm and our graph-based algorithm in measuring the total and regional macular thickness for normal and glaucoma subjects.

METHODS

A total of 228 normal eyes and 93 glaucomatous eyes were enrolled in our study. All patients underwent comprehensive ophthalmic examination and Topcon 3D-OCT 2000 scan. One eye was randomly selected for each subject. The thickness of each layer and the total and regional macular thickness on an Early Treatment of Diabetic Retinopathy Study (ETDRS) chart were measured using the Topcon algorithm and our three-dimensional graph-based algorithm. Correlation and agreement analyses between these two algorithms were performed.

RESULTS

Our graph search algorithm exhibited a strong correlation with Topcon algorithm. The macular GCC thickness values for normal and glaucoma subjects ranged from 0.86 to 0.91 and from 0.78 to 0.90, and the regional macular thickness values ranged from 0.79 to 0.96 and 0.70 to 0.95, respectively. Small differences were observed between the Topcon algorithm and our graph-based algorithm. The span of 95% limits of agreement of macular GCC thickness was less than 28 μm in both normal and glaucoma subjects, respectively. These limits of total and regional macular thickness were 15.5 μm and 23.1 μm for normal subjects and 29.1 μm and 46.4 μm for glaucoma subjects, respectively.

CONCLUSION

Our graph-based algorithm exhibited a high degree of agreement with the Topcon algorithm with respect to thickness measurements in normal and glaucoma subjects. Moreover, our graph-based algorithm can segment the retina into more layers than the Topcon algorithm does.

Optic disc segmentation by balloon snake with texture from color fundus image

A well-established method for diagnosis of glaucoma is the examination of the optic nerve head based on fundus image as glaucomatous patients tend to have larger cup-to-disc ratios. The difficulty of optic segmentation is due to the fuzzy boundaries and peripapillary atrophy (PPA). In this paper a novel method for optic nerve head segmentation is proposed. It uses template matching to find the region of interest (ROI). The method of vessel erasing in the ROI is based on PDE inpainting which will make the boundary smoother. A novel optic disc segmentation approach using image texture is explored in this paper. A cluster method based on image texture is employed before the optic disc segmentation step to remove the edge noise such as cup boundary and vessels. We replace image force in the snake with image texture and the initial contour of the balloon snake is inside the optic disc to avoid the PPA. The experimental results show the superior performance of the proposed method when compared to some traditional segmentation approaches. An average segmentation dice coefficient of 94% has been obtained.

Obtaining optic disc center and pixel region by automatic thresholding methods on morphologically processed fundus images

Development of automatic retinal disease diagnosis systems based on retinal image computer analysis can provide remarkably quicker screening programs for early detection. Such systems are mainly focused on the detection of the earliest ophthalmic signs of illness and require previous identification of fundal landmark features such as optic disc (OD), fovea or blood vessels. A methodology for accurate center-position location and OD retinal region segmentation on digital fundus images is presented in this paper. The methodology performs a set of iterative opening-closing morphological operations on the original retinography intensity channel to produce a bright region-enhanced image. Taking blood vessel confluence at the OD into account, a 2-step automatic thresholding procedure is then applied to obtain a reduced region of interest, where the center and the OD pixel region are finally obtained by performing the circular Hough transform on a set of OD boundary candidates generated through the application of the Prewitt edge detector. The methodology was evaluated on 1200 and 1748 fundus images from the publicly available MESSIDOR and MESSIDOR-2 databases, acquired from diabetic patients and thus being clinical cases of interest within the framework of automated diagnosis of retinal diseases associated to diabetes mellitus. This methodology proved highly accurate in OD-center location: average Euclidean distance between the methodology-provided and actual OD-center position was 6.08, 9.22 and 9.72 pixels for retinas of 910, 1380 and 1455 pixels in size, respectively. On the other hand, OD segmentation evaluation was performed in terms of Jaccard and Dice coefficients, as well as the mean average distance between estimated and actual OD boundaries. Comparison with the results reported by other reviewed OD segmentation methodologies shows our proposal renders better overall performance. Its effectiveness and robustness make this proposed automated OD location and segmentation method a suitable tool to be integrated into a complete prescreening system for early diagnosis of retinal diseases.

Optic disc and macula detection in fundus images by means of template matching

Various methods for detecting optic disc and macula in fundus images have been developed. Our aim is to propose a fairly easy method for detecting both features jointly. This is achieved by first correcting in homogenous luminosity using a polynomial approximation of the background of the images. Secondly, the use of the cross-correlation in the frequency domain between the images and a steerable template which contains both structures. The 38 photographs used in this work belong to a local database of patients suffering diabetic retinopathy along its four severity stages. Our results showed 100% optic disc centers located within the OD area and 90% macula centers located within the MC area.

A computational modeling for the detection of diabetic retinopathy severity

Prolonged diabetes ultimately leads to Diabetic Retinopathy (DR) which is one of the leading causes of preventable blindness in the world. Through advanced image analysis techniques are used for abnormalities detection in retina that define and correlate the severity of DR. A thorough study is done in this area in recent past years and on the basis of these studies we have developed a computer based prediction model that is used to determine the severity of DR. To identify severity DR, we have analyzed the human eye image. We have extracted some important features from human eye image i.e. Blood Artery, Optical disc, Exudates. Based on these image and data we have designed an automated system for the determination of DR severity. This automated DR severity assessment methods can be used to predict the clinical case and conditions when young clinicians would agree or disagree with their more experienced fellow members. The algorithms described in this study may be used in clinical practice to validate or invalidate the diagnoses. Algorithms or method developed here may also be used for pooling diagnostic knowledge for serving mankind. Here we have described a computational based low cost retinal diagnostic approach which can aid an ophthalmologist to quickly diagnose the various stages of DR. This system can accept retinal images and can successfully detect any pathological condition associated with DR.

A survey on computer aided diagnosis for ocular diseases

Background

Computer Aided Diagnosis (CAD), which can automate the detection process for ocular diseases, has attracted extensive attention from clinicians and researchers alike. It not only alleviates the burden on the clinicians by providing objective opinion with valuable insights, but also offers early detection and easy access for patients.

Method

We review ocular CAD methodologies for various data types. For each data type, we investigate the databases and the algorithms to detect different ocular diseases. Their advantages and shortcomings are analyzed and discussed.

Result

We have studied three types of data (i.e., clinical, genetic and imaging) that have been commonly used in existing methods for CAD. The recent developments in methods used in CAD of ocular diseases (such as Diabetic Retinopathy, Glaucoma, Age-related Macular Degeneration and Pathological Myopia) are investigated and summarized comprehensively.

Conclusion

While CAD for ocular diseases has shown considerable progress over the past years, the clinical importance of fully automatic CAD systems which are able to embed clinical knowledge and integrate heterogeneous data sources still show great potential for future breakthrough.

Retinal imaging as a source of biomarkers for diagnosis, characterization and prognosis of chronic illness or long-term conditions

The black void behind the pupil was optically impenetrable before the invention of the ophthalmoscope by von Helmholtz over 150 years ago. Advances in retinal imaging and image processing, especially over the past decade, have opened a route to another unexplored landscape, the retinal neurovascular architecture and the retinal ganglion pathways linking to the central nervous system beyond. Exploiting these research opportunities requires multidisciplinary teams to explore the interface sitting at the border between ophthalmology, neurology and computing science. It is from the detail and depth of retinal phenotyping that novel metrics and candidate biomarkers are likely to emerge. Confirmation that in vivo retinal neurovascular measures are predictive of microvascular change in the brain and other organs is likely to be a major area of research activity over the next decade. Unlocking this hidden potential within the retina requires integration of structural and functional data sets, that is, multimodal mapping and longitudinal studies spanning the natural history of the disease process. And with further advances in imaging, it is likely that this area of retinal research will remain active and clinically relevant for many years to come. Accordingly, this review looks at state-of-the-art retinal imaging and its application to diagnosis, characterization and prognosis of chronic illness or long-term conditions.

Optic disc and cup segmentation from color fundus photograph using graph cut with priors

For automatic segmentation of optic disc and cup from color fundus photograph, we describe a fairly general energy function that can naturally fit into a global optimization framework with graph cut. Distinguished from most previous work, our energy function includes priors on the shape & location of disc & cup, the rim thickness and the geometric interaction of "disc contains cup". These priors together with the effective optimization of graph cut enable our algorithm to generate reliable and robust solutions. Our approach is able to outperform several state-of-the-art segmentation methods, as shown by a set of experimental comparisons with manual delineations and a series of results of correlations with the assessments of a merchant-provided software from Optical Coherence Tomography (OCT) regarding several cup and disc parameters.

Incorporating privileged genetic information for fundus image based glaucoma detection

Visual features extracted from retinal fundus images have been increasingly used for glaucoma detection, as those images are generally easy to acquire. In recent years, genetic researchers have found that some single nucleic polymorphisms (SNPs) play important roles in the manifestation of glaucoma and also show superiority over fundus images for glaucoma detection. In this work, we propose to use the SNPs to form the so-called privileged information and deal with a practical problem where both fundus images and privileged genetic information exist for the training subjects, while the test objects only have fundus images. To solve this problem, we present an effective approach based on the learning using privileged information (LUPI) paradigm to train a predictive model for the image visual features. Extensive experiments demonstrate the usefulness of our approach in incorporating genetic information for fundus image based glaucoma detection.

Improved automated optic cup segmentation based on detection of blood vessel bends in retinal fundus images

Glaucoma is a leading cause of permanent blindness. Retinal imaging is useful for early detection of glaucoma. In order to evaluate the presence of glaucoma, ophthalmologists may determine the cup and disc areas and diagnose glaucoma using a vertical optic cup-to-disc (C/D) ratio and a rim-to-disc (R/D) ratio. Previously we proposed a method to determine cup edge by analyzing a vertical profile of pixel values, but this method provided a cup edge smaller than that of an ophthalmologist. This paper describes an improved method using the locations of the blood vessel bends. The blood vessels were detected by a concentration feature determined from the density gradient. The blood vessel bends were detected by tracking the blood vessels from the disc edge to the primary cup edge, which was determined by our previous method. Lastly, the vertical C/D ratio and the R/D ratio were calculated. Using forty-four images, including 32 glaucoma images, the AUCs of both the vertical C/D ratio and R/D ratio by this proposed method were 0.966 and 0.936, respectively.

An ensembling approach for optic cup detection based on spatial heuristic analysis in retinal fundus images

Optic cup detection remains a challenging task in retinal image analysis, and is of particular importance for glaucoma evaluation, where disease severity is often assessed by the size of the optic cup. In this paper, we propose spatial heuristic ensembling (SHE), an approach which aims to fuse the advantages of each method based on the specific performance in each defined sector. In this way, we generate an ensembled optic cup which is obtained from the optimal combination of the component methods. We conduct experiments on the ORIGA data set of 650 retinal images and show that the ensemble approach performs better than the individual segmentations, reducing the relative overlap error, and CDR errors by as much as 0.04 CDR units. The results are promising for the continued development of such an approach for improving optic cup segmentation.

Superpixel classification for initialization in model based optic disc segmentation

Optic disc segmentation in retinal fundus image is important in ocular image analysis and computer aided diagnosis. Because of the presence of peripapillary atrophy which affects the deformation, it is important to have a good initialization in deformable model based optic disc segmentation. In this paper, a superpixel classification based method is proposed for the initialization. It uses histogram of superpixels from the contrast enhanced image as features. In the training, bootstrapping is adopted to handle the unbalanced cluster issue due to the presence of peripapillary atrophy. A self-assessment reliability score is computed to evaluate the quality of the initialization and the segmentation. The proposed method has been tested in a database of 650 images with optic disc boundaries marked by trained professionals manually. The experimental results show an mean overlapping error of 10.0% and standard deviation of 7.5% in the best scenario. The results also show an increase in overlapping error as the reliability score reduces, which justifies the effectiveness of the self-assessment. The method can be used for optic disc boundary initialization and segmentation in computer aided diagnosis system and the self-assessment can be used as an indicator of cases with large errors and thus enhance the usage of the automatic segmentation.

Superpixel classification based optic disc and optic cup segmentation for glaucoma screening

Glaucoma is a chronic eye disease that leads to vision loss. As it cannot be cured, detecting the disease in time is important. Current tests using intraocular pressure (IOP) are not sensitive enough for population based glaucoma screening. Optic nerve head assessment in retinal fundus images is both more promising and superior. This paper proposes optic disc and optic cup segmentation using superpixel classification for glaucoma screening. In optic disc segmentation, histograms, and center surround statistics are used to classify each superpixel as disc or non-disc. A self-assessment reliability score is computed to evaluate the quality of the automated optic disc segmentation. For optic cup segmentation, in addition to the histograms and center surround statistics, the location information is also included into the feature space to boost the performance. The proposed segmentation methods have been evaluated in a database of 650 images with optic disc and optic cup boundaries manually marked by trained professionals. Experimental results show an average overlapping error of 9.5% and 24.1% in optic disc and optic cup segmentation, respectively. The results also show an increase in overlapping error as the reliability score is reduced, which justifies the effectiveness of the self-assessment. The segmented optic disc and optic cup are then used to compute the cup to disc ratio for glaucoma screening. Our proposed method achieves areas under curve of 0.800 and 0.822 in two data sets, which is higher than other methods. The methods can be used for segmentation and glaucoma screening. The self-assessment will be used as an indicator of cases with large errors and enhance the clinical deployment of the automatic segmentation and screening.

Validating retinal fundus image analysis algorithms: issues and a proposal

This paper concerns the validation of automatic retinal image analysis (ARIA) algorithms. For reasons of space and consistency, we concentrate on the validation of algorithms processing color fundus camera images, currently the largest section of the ARIA literature. We sketch the context (imaging instruments and target tasks) of ARIA validation, summarizing the main image analysis and validation techniques. We then present a list of recommendations focusing on the creation of large repositories of test data created by international consortia, easily accessible via moderated Web sites, including multicenter annotations by multiple experts, specific to clinical tasks, and capable of running submitted software automatically on the data stored, with clear and widely agreed-on performance criteria, to provide a fair comparison.

Comparison of disc analysis algorithms provided by cirrus OCT and stereo optic-disc photography in normal and open angle glaucoma patients

PURPOSE

To compare optic nerve head (ONH) parameters obtained by semi-automated disc analysis of stereo optic-disc photography (ODP) with those by spectral domain optical coherence tomography (SD-OCT) regarding their reproducibility, agreement and relationship with the retinal nerve fiber layer (RNFL) thickness.

PATIENTS AND METHODS

In this cross-sectional retrospective study, 91 eyes (50 normal and 41 with open angle glaucoma) from 47 patients were examined. ONH parameters were obtained by ODP and SD-OCT. Agreement and reproducibility were assessed by intra-class correlation coefficients (ICCs). Inter-device agreement and precision were also evaluated. The structural factors related to the ONH measurements by both instruments and the disparity between the measurements were evaluated by the generalized estimating equation model.

RESULTS

ODP showed good intra-reader and inter-reader reproducibility, comparable to that of SD-OCT (ICC: >0.970). Disc area (DA) (ICC: 0.950) showed better inter-device agreement than rim area (RA) (ICC: 0.859). RNFL thickness was correlated with RA, but not with DA, for both devices. SD-OCT RA showed significant correlation with RNFL thickness in both the normal and glaucoma groups, whereas ODP RA correlated with RNFL only in the glaucoma group. The measurement disparity between the devices was influenced by the DA in both groups. The inter-device difference in RA was not correlated with RA, but instead with the RNFL thickness in the glaucoma group.

CONCLUSION

ONH parameters obtained by semi-automated algorithms of ODP showed excellent reproducibility and good agreement with those measured by SD-OCT. SD-OCT RA was a better indicator of RNFL thickness, compared with ODP RA for both normal patients and glaucoma patients; however, the inter-device disparity of the RA was less in glaucoma patients who had a thinner RNFL.

Efficient reconstruction-based optic cup localization for glaucoma screening

We present a reconstruction-based learning technique to localize the optic cup in fundus images for glaucoma screening. In contrast to previous approaches which rely on low-level visual cues, our method instead considers the input image as a whole and infers its optic cup parameters from a codebook of manually labeled reference images based on their similarity to the input and their contribution towards reconstructing the input image. We show that this approach can be formulated as a closed-form solution without any search, which leads to highly efficient and 100% repeatable computation. Our tests on the ORIGA and SCES datasets show that the performance of this method compares favorably to those of previous techniques while operating at faster speeds. This suggests much promise for this approach to be used in practice for screening.

A combined machine-learning and graph-based framework for the segmentation of retinal surfaces in SD-OCT volumes

Optical coherence tomography is routinely used clinically for the detection and management of ocular diseases as well as in research where the studies may involve animals. This routine use requires that the developed automated segmentation methods not only be accurate and reliable, but also be adaptable to meet new requirements. We have previously proposed the use of a graph-theoretic approach for the automated 3-D segmentation of multiple retinal surfaces in volumetric human SD-OCT scans. The method ensures the global optimality of the set of surfaces with respect to a cost function. Cost functions have thus far been typically designed by hand by domain experts. This difficult and time-consuming task significantly impacts the adaptability of these methods to new models. Here, we describe a framework for the automated machine-learning based design of the cost function utilized by this graph-theoretic method. The impact of the learned components on the final segmentation accuracy are statistically assessed in order to tailor the method to specific applications. This adaptability is demonstrated by utilizing the method to segment seven, ten and five retinal surfaces from SD-OCT scans obtained from humans, mice and canines, respectively. The overall unsigned border position errors observed when using the recommended configuration of the graph-theoretic method was 6.45 ± 1.87 μm, 3.35 ± 0.62 μm and 9.75 ± 3.18 μm for the human, mouse and canine set of images, respectively.

Validation of tablet-based evaluation of color fundus images

PURPOSE

To compare diabetic retinopathy (DR) referral recommendations made by viewing fundus images using a tablet computer with those made using a standard desktop display.

METHODS

A tablet computer (iPad) and a desktop computer with a high-definition color display were compared. For each platform, 2 retinal specialists independently rated 1,200 color fundus images from patients at risk for DR using an annotation program Truthseeker. The specialists determined whether each image had referable DR and also how urgently each patient should be referred for medical examination. Graders viewed and rated the randomly presented images independently and were masked to their ratings on the alternative platform. Tablet-based and desktop display-based referral ratings were compared using cross-platform intraobserver kappa as the primary outcome measure. Additionally, interobserver kappa, sensitivity, specificity, and area under the receiver operating characteristic were determined.

RESULTS

A high level of cross-platform intraobserver agreement was found for the DR referral ratings between the platforms (κ = 0.778) and for the 2 graders (κ = 0.812). Interobserver agreement was similar for the 2 platforms (κ = 0.544 and κ = 0.625 for tablet and desktop, respectively). The tablet-based ratings achieved a sensitivity of 0.848, a specificity of 0.987, and an area under the receiver operating characteristic of 0.950 compared with desktop display-based ratings.

CONCLUSION

In this pilot study, tablet-based rating of color fundus images for subjects at risk for DR was consistent with desktop display-based rating. These results indicate that tablet computers can be reliably used for clinical evaluation of fundus images for DR.

Variability between experts in defining the edge and area of the optic nerve head

OBJECTIVE

Estimation of the error rate in the subjective determination of the optic nerve head edge and area.

METHOD

1) 169 images of optic nerve disc were evaluated by five experts for the defining of the edges in 8 positions (every 45°). 2) The estimated areas of 26 cases were compared with the measurements of the Cirrus Optical Coherence Tomography (OCT-Cirrus).

RESULTS

1) The mean variation of the estimated radius was ±5.2%, with no significant differences between sectors. Specific differences were found between the 5 experts (P <.001), each one compared with the others. 2) The disc area measured by the OCT-Cirros was 1.78 mm² (SD =0.27). The results corresponding to the experts who detected smaller areas were better correlated to the area detected by the OCT-Cirrus (r=0.77-0.88) than the results corresponding to larger areas (r =0.61-0.69) (P <.05 in extreme cases).

CONCLUSIONS

There are specific patterns in each expert for defining the disc edges and involve 20% variation in the estimation of the optic nerve area. The experts who detected smaller areas have a higher agreement with the objective method used. A web tool is proposed for self-assessment and training in this task.

Automated detection of optic disk in retinal fundus images using intuitionistic fuzzy histon segmentation

The human eye is one of the most sophisticated organs, with perfectly interrelated retina, pupil, iris cornea, lens, and optic nerve. Automatic retinal image analysis is emerging as an important screening tool for early detection of eye diseases. Uncontrolled diabetic retinopathy (DR) and glaucoma may lead to blindness. The identification of retinal anatomical regions is a prerequisite for the computer-aided diagnosis of several retinal diseases. The manual examination of optic disk (OD) is a standard procedure used for detecting different stages of DR and glaucoma. In this article, a novel automated, reliable, and efficient OD localization and segmentation method using digital fundus images is proposed. General-purpose edge detection algorithms often fail to segment the OD due to fuzzy boundaries, inconsistent image contrast, or missing edge features. This article proposes a novel and probably the first method using the Attanassov intuitionistic fuzzy histon (A-IFSH)–based segmentation to detect OD in retinal fundus images. OD pixel intensity and column-wise neighborhood operation are employed to locate and isolate the OD. The method has been evaluated on 100 images comprising 30 normal, 39 glaucomatous, and 31 DR images. Our proposed method has yielded precision of 0.93, recall of 0.91, F-score of 0.92, and mean segmentation accuracy of 93.4%. We have also compared the performance of our proposed method with the Otsu and gradient vector flow (GVF) snake methods. Overall, our result shows the superiority of proposed fuzzy segmentation technique over other two segmentation methods.

Simple methods for segmentation and measurement of diabetic retinopathy lesions in retinal fundus images

Diabetic retinopathy (DR) is one of the most important complications of diabetes mellitus, which causes serious damages in the retina, consequently visual loss and sometimes blindness if necessary medical treatment is not applied on time. One of the difficulties in this illness is that the patient with diabetes mellitus requires a continuous screening for early detection. So far, numerous methods have been proposed by researchers to automate the detection process of DR in retinal fundus images. In this paper, we developed an alternative simple approach to detect DR. This method was built on the inverse segmentation method, which we suggested before to detect Age Related Macular Degeneration (ARMDs). Background image approach along with inverse segmentation is employed to measure and follow up the degenerations in retinal fundus images. Direct segmentation techniques generate unsatisfactory results in some cases. This is because of the fact that the texture of unhealthy areas such as DR is not homogenous. The inverse method is proposed to exploit the homogeneity of healthy areas rather than dealing with varying structure of unhealthy areas for segmenting bright lesions (hard exudates and cotton wool spots). On the other hand, the background image, dividing the retinal image into high and low intensity areas, is exploited in segmentation of hard exudates and cotton wool spots, and microaneurysms (MAs) and hemorrhages (HEMs), separately. Therefore, a complete segmentation system is developed for segmenting DR, including hard exudates, cotton wool spots, MAs, and HEMs. This application is able to measure total changes across the whole retinal image. Hence, retinal images that belong to the same patients are examined in order to monitor the trend of the illness. To make a comparison with other methods, a Naïve Bayes method is applied for segmentation of DR. The performance of the system, tested on different data sets including various qualities of retinal fundus images, is over 95% in detection of the optic disc (OD), and 90% in segmentation of the DR.

Three-dimensional segmentation of fluid-associated abnormalities in retinal OCT: probability constrained graph-search-graph-cut

An automated method is reported for segmenting 3-D fluid-associated abnormalities in the retina, so-called symptomatic exudate-associated derangements (SEAD), from 3-D OCT retinal images of subjects suffering from exudative age-related macular degeneration. In the first stage of a two-stage approach, retinal layers are segmented, candidate SEAD regions identified, and the retinal OCT image is flattened using a candidate-SEAD aware approach. In the second stage, a probability constrained combined graph search-graph cut method refines the candidate SEADs by integrating the candidate volumes into the graph cut cost function as probability constraints. The proposed method was evaluated on 15 spectral domain OCT images from 15 subjects undergoing intravitreal anti-VEGF injection treatment. Leave-one-out evaluation resulted in a true positive volume fraction (TPVF), false positive volume fraction (FPVF) and relative volume difference ratio (RVDR) of 86.5%, 1.7%, and 12.8%, respectively. The new graph cut-graph search method significantly outperformed both the traditional graph cut and traditional graph search approaches (p < 0.01, p < 0.04) and has the potential to improve clinical management of patients with choroidal neovascularization due to exudative age-related macular degeneration.

Quantitative evaluation of papilledema from stereoscopic color fundus photographs

PURPOSE

To derive a computerized measurement of optic disc volume from digital stereoscopic fundus photographs for the purpose of diagnosing and managing papilledema.

METHODS

Twenty-nine pairs of stereoscopic fundus photographs and optic nerve head (ONH) centered spectral domain optical coherence tomography (SD-OCT) scans were obtained at the same visit in 15 patients with papilledema. Some patients were imaged at multiple visits in order to assess their changes. Three-dimensional shape of the ONH was estimated from stereo fundus photographs using an automated multi-scale stereo correspondence algorithm. We assessed the correlation of the stereo volume measurements with the SD-OCT volume measurements quantitatively, in terms of volume of retinal surface elevation above a reference plane and also to expert grading of papilledema from digital fundus photographs using the Frisén grading scale.

RESULTS

The volumetric measurements of retinal surface elevation estimated from stereo fundus photographs and OCT scans were positively correlated (correlation coefficient r(2) = 0.60; P < 0.001) and were positively correlated with Frisén grade (Spearman correlation coefficient r = 0.59; P < 0.001).

CONCLUSIONS

Retinal surface elevation among papilledema patients obtained from stereo fundus photographs compares favorably with that from OCT scans and with expert grading of papilledema severity. Stereoscopic color imaging of the ONH combined with a method of automated shape reconstruction is a low-cost alternative to SD-OCT scans that has potential for a more cost-effective diagnosis and management of papilledema in a telemedical setting. An automated three-dimensional image analysis method was validated that quantifies the retinal surface topography with an imaging modality that has lacked prior objective assessment.

Model-based optic nerve head segmentation on retinal fundus images

The optic nerve head (optic disc) plays an important role in the diagnosis of retinal diseases. Automatic localization and segmentation of the optic disc is critical towards a good computer-aided diagnosis (CAD) system. In this paper, we propose a method that combines edge detection, the Circular Hough Transform and a statistical deformable model to detect the optic disc from retinal fundus images. The algorithm was evaluated against a data set of 325 digital color fundus images, which includes both normal images and images with various pathologies. The result shows that the average error in area overlap is 11.3% and the average absolute area error is 10.8%, which outperforms existing methods. The result indicates a high correlation with ground truth segmentation and thus demonstrates a good potential for this system to be integrated with other retinal CAD systems.

Automatic measurement of cup to disc ratio based on line profile analysis in retinal images

Retinal image examination is useful for early detection of glaucoma, which is a leading cause of permanent blindness. In order to evaluate the presence of glaucoma, ophthalmologists may determine the cup and disc areas and diagnose glaucoma using a vertical cup-to-disc ratio. However, determination of the cup area based on computation algorithm is very difficult, thus we propose a method to measure the cup-to-disc ratio using a vertical profile on the optic disc. The edge of optic disc was then detected by use of a Canny edge detection filter. The profile was then obtained around the center of the optic disc. Subsequently, the edges of the cup area were determined by classification of the profiles based on zero-crossing method. Lastly, the vertical cup-to-disc ratio was calculated. Using forty five images, including twenty three glaucoma images, the AUC of 0.947 was achieved with this method.

Automatic glaucoma diagnosis from fundus image

Glaucoma is currently diagnosed by glaucoma specialists using specialized imaging devices like HRT and OCT. Fundus imaging is a modality widely used in primary healthcare. An automatic glaucoma diagnosis system based on fundus image can be deployed to primary healthcare clinics and has potential for early disease diagnosis. A mass glaucoma screening program can also be facilitated using such a system. We present an automatic fundus image based cup-to-disc ratio measurement system; and demonstrate its potential for automatic objective glaucoma diagnosis and screening. It provides strong support to use fundus image as the modality for automatic glaucoma diagnosis.

Depth discontinuity-based cup segmentation from multiview color retinal images

Accurate segmentation of the cup region from retinal images is needed to derive relevant measurements for glaucoma assessment. A novel, depth discontinuity (in the retinal surface)-based approach to estimate the cup boundary is proposed in this paper. The proposed approach shifts focus from the cup region used by existing approaches to cup boundary. The given set of images, acquired sequentially, are related via a relative motion model and the depth discontinuity at the cup boundary is determined from cues such as motion boundary and partial occlusion. The information encoded by these cues is used to approximate the cup boundary with a set of best-fitting circles. The final boundary is found by considering points on these circles at different sectors using a confidence measure. Four different kinds of data sets ranging from synthetic to real image pairs, covering different multiview scenarios, have been used to evaluate the proposed method. The proposed method was found to yield an error reduction of 16% for cup-to-disk vertical diameter ratio (CDR) and 13% for cup-to-disk area ratio (CAR) estimation, over an existing monocular image-based cup segmentation method. The error reduction increased to 33% in CDR and 18% in CAR with the addition of a third view (image) which indicates the potential of the proposed approach.

Efficient optic cup localization using regional propagation based on retinal structure priors

We present a regional propagation approach based on retinal structure priors to localize the optic cup in 2D fundus images, which is the primary image component clinically used for identifying glaucoma. This method provides three major contributions. First, it proposes processing of the fundus images at the superpixel level, which leads to more descriptive and effective features than those employed by pixel based techniques, without additional computational cost. Second, the proposed approach does not need manually labeled training samples, but uses the structural priors on relative cup and disc positions. Third, a refinement scheme that utilizes local context information is adopted to further improve the accuracy. Tested on the ORIGA-light clinical dataset, which comprises of 325 images from a population-based study, the proposed method achieves a 34.9% non-overlap ratio with manually-labeled ground-truth and a 0.104 absolute cup-to-disc ratio (CDR) error. This level of accuracy is much higher than the state-of-the-art pixel based techniques, with a comparable or even less computational cost.

Automated discovery and quantification of image-based complex phenotypes: a twin study of drusen phenotypes in age-related macular degeneration

PURPOSE

Determining the relationships between phenotype and genotype of many disorders can improve clinical diagnoses, identify disease mechanisms, and enhance therapy. Most genetic disorders result from interaction of many genes that obscure the discovery of such relationships. The hypothesis for this study was that image analysis has the potential to enable formalized discovery of new visible phenotypes. It was tested in twins affected with age-related macular degeneration (AMD).

METHODS

Fundus images from 43 monozygotic (MZ) and 32 dizygotic (DZ) twin pairs with AMD were examined. First, soft and hard drusen were segmented. Then newly defined phenotypes were identified by using drusen distribution statistics that significantly separate MZ from DZ twins. The ACE model was used to identify the contributions of additive genetic (A), common environmental (C), and nonshared environmental (E) effects on drusen distribution phenotypes.

RESULTS

Four drusen distribution characteristics significantly separated MZ from DZ twin pairs. One encoded the quantity, and the remaining three encoded the spatial distribution of drusen, achieving a zygosity prediction accuracy of 76%, 74%, 68%, and 68%. Three of the four phenotypes had a 55% to 77% genetic effect in an AE model, and the fourth phenotype showed a nonshared environmental effect (E model).

CONCLUSIONS

Computational discovery of genetically determined features can reveal quantifiable AMD phenotypes that are genetically determined without explicitly linking them to specific genes. In addition, it can identify phenotypes that appear to result predominantly from environmental exposure. The approach is rapid and unbiased, suitable for large datasets, and can be used to reveal unknown phenotype-genotype relationships.

Sliding window and regression based cup detection in digital fundus images for glaucoma diagnosis

We propose a machine learning framework based on sliding windows for glaucoma diagnosis. In digital fundus photographs, our method automatically localizes the optic cup, which is the primary structural image cue for clinically identifying glaucoma. This localization uses a bundle of sliding windows of different sizes to obtain cup candidates in each disc image, then extracts from each sliding window a new histogram based feature that is learned using a group sparsity constraint. An epsilon-SVR (support vector regression) model based on non-linear radial basis function (RBF) kernels is used to rank each candidate, and final decisions are made with a non-maximal suppression (NMS) method. Tested on the large ORIGA(-light) clinical dataset, the proposed method achieves a 73.2% overlap ratio with manually-labeled ground-truth and a 0.091 absolute cup-to-disc ratio (CDR) error, a simple yet widely used diagnostic measure. The high accuracy of this framework on images from low-cost and widespread digital fundus cameras indicates much promise for developing practical automated/assisted glaucoma diagnosis systems.

Robust multiscale stereo matching from fundus images with radiometric differences

A robust multiscale stereo matching algorithm is proposed to find reliable correspondences between low contrast and weakly textured retinal image pairs with radiometric differences. Existing algorithms designed to deal with piecewise planar surfaces with distinct features and Lambertian reflectance do not apply in applications such as 3D reconstruction of medical images including stereo retinal images. In this paper, robust pixel feature vectors are formulated to extract discriminative features in the presence of noise in scale space, through which the response of low-frequency mechanisms alter and interact with the response of high-frequency mechanisms. The deep structures of the scene are represented with the evolution of disparity estimates in scale space, which distributes the matching ambiguity along the scale dimension to obtain globally coherent reconstructions. The performance is verified both qualitatively by face validity and quantitatively on our collection of stereo fundus image sets with ground truth, which have been made publicly available as an extension of standard test images for performance evaluation.

Automated determination of cup-to-disc ratio for classification of glaucomatous and normal eyes on stereo retinal fundus images

Early diagnosis of glaucoma, which is the second leading cause of blindness in the world, can halt or slow the progression of the disease. We propose an automated method for analyzing the optic disc and measuring the cup-to-disc ratio (CDR) on stereo retinal fundus images to improve ophthalmologists' diagnostic efficiency and potentially reduce the variation on the CDR measurement. The method was developed using 80 retinal fundus image pairs, including 25 glaucomatous, and 55 nonglaucomatous eyes, obtained at our institution. A disc region was segmented using the active contour method with the brightness and edge information. The segmentation of a cup region was performed using a depth map of the optic disc, which was reconstructed on the basis of the stereo disparity. The CDRs were measured and compared with those determined using the manual segmentation results by an expert ophthalmologist. The method was applied to a new database which consisted of 98 stereo image pairs including 60 and 30 pairs with and without signs of glaucoma, respectively. Using the CDRs, an area under the receiver operating characteristic curve of 0.90 was obtained for classification of the glaucomatous and nonglaucomatous eyes. The result indicates potential usefulness of the automated determination of CDRs for the diagnosis of glaucoma.