Enhancing the structure-function relationship in glaucoma using anatomical compensation of retinal nerve fibre layer.

BACKGROUND/AIMS: To investigate whether compensating retinal nerve fibre layer (RNFL) thickness measurements for demographic and anatomical ocular factors can strengthen the structure-function relationship in patients with glaucoma. METHODS: 600 eyes from 412 patients with glaucoma (mean deviation of the visual field (MD VF) -6.53±5.55 dB) were included in this cross-sectional study. Participants underwent standard automated perimetry and spectral-domain optical coherence tomography imaging (Cirrus; Carl Zeiss Meditec). Compensated RNFL thickness was computed considering age, refractive error, optic disc parameters and retinal vessel density. The relationship between MD VF and RNFL thickness measurements, with or without demographic and anatomical compensation, was evaluated sectorally and focally. RESULTS: The superior arcuate sector exhibited the highest correlation between measured RNFL and MD VF, with a correlation of 0.49 (95% CI 0.37 to 0.59). Applying the compensated RNFL data increased the correlation substantially to 0.62 (95% CI 0.52 to 0.70; p<0.001). Only 61% of the VF locations showed a significant relationship (Spearman's correlation of at least 0.30) between structural and functional aspects using measured RNFL data, and this increased to 78% with compensated RNFL measurements. In the 10°-20° VF region, the slope below the breakpoint for compensated RNFL thickness demonstrated a more robust correlation (slope=1.66±0.18 µm/dB; p<0.001) than measured RNFL (slope=0.27±0.67 µm/dB; p=0.688). CONCLUSION: Compensated RNFL data improve the correlation between RNFL measurements and VF parameters. This indicates that creating structure-to-function maps that consider anatomical variances may aid in identifying localised structural and functional loss in glaucoma.

Choriocapillaris microvasculature dysfunction in systemic hypertension.

We examined the choriocapillaris microvasculature using a non-invasive swept-source optical coherence tomography angiography (SS-OCTA) in 41 healthy controls and 71 hypertensive patients and determined possible correlations with BP and renal parameters. BP levels, serum creatinine and urine microalbumin/creatinine ratio (MCR) specimens were collected. The estimated glomerular filtration rate (eGFR) was calculated based on CKD-EPI Creatinine Equation. The main outcome was choriocapillaris flow deficits (CFD) metrics (density, size and numbers). The CFD occupied a larger area and were fewer in number in the hypertensive patients with poor BP control (407 ± 10 µm2; 3260 ± 61) compared to the hypertensives with good BP control (369 ± 5 µm2; 3551 ± 41) and healthy controls (365 ± 11 µm2; 3581 ± 84). Higher systolic BP (ß = 9.90, 95% CI, 2.86-16.93), lower eGFR (ß = - 0.85; 95% CI, - 1.58 to - 0.13) and higher urine MCR (ß = 1.53, 95% CI, 0.32-2.78) were associated with larger areas of CFD. Similar significant associations with systolic BP, eGFR and urine MCR were found with number of CFD. These findings highlight the potential role of choriocapillaris imaging using SS-OCTA as an indicator of systemic microvascular abnormalities secondary to hypertensive disease.

Association Between Structure-function Characteristics and Visual Field Outcomes in Glaucoma Subjects With Intraocular Pressure Reduction After Trabeculectomy.

PRECIS: Improvements in post-trabeculectomy visual field (VF) outcomes were found to be significantly associated with preoperative nerve fiber layer thickness parameters extracted from the sectorized structure-function relationship, baseline VF, and severity of glaucoma. OBJECTIVE: To determine whether the preoperative structure-function relationship helps to predict visual outcomes at 1-year post-trabeculectomy. PATIENTS AND METHODS: In total, 91 eyes from 87 participants who successfully underwent trabeculectomy were included in our study. All eyes received optical coherence tomography imaging and VF assessment using 30-2 standard automated perimetry preoperatively at baseline and postoperatively 1 year after trabeculectomy. The linear mixed-model analysis was used to assess the association of structure and function at baseline, and multivariate analysis to investigate factors associated with postoperative VF outcomes. RESULTS: Results from multivariate and univariate analysis for VF 1 year after trabeculectomy showed that a positive preoperative retinal nerve fiber layer thickness deviation from the structure-function model was found to be significantly associated with improved postoperative VF outcomes [ß=0.06 dB/µm; 95% confidence interval (CI), 0.03-0.09]. Other significant factors included baseline VF MD (ß=-0.18; 95% CI, -0.23 to -0.13) and the presence of severe glaucoma (ß=-1.69; 95% CI, -2.80 to -0.57). Intraocular pressure was positively associated with improved VF outcomes only in univariate analysis (ß=0.06; 95% CI, 0.01-0.11). CONCLUSIONS AND RELEVANCE: Characteristics derived from the baseline structure-function relationship were found to be strongly associated with postoperative VF outcomes. These findings suggest that the structure-function relationship could potentially have a role in predicting VF progression after trabeculectomy.

Angle-Closure Detection in Anterior Segment OCT Based on Multilevel Deep Network.

Irreversible visual impairment is often caused by primary angle-closure glaucoma, which could be detected via anterior segment optical coherence tomography (AS-OCT). In this paper, an automated system based on deep learning is presented for angle-closure detection in AS-OCT images. Our system learns a discriminative representation from training data that captures subtle visual cues not modeled by handcrafted features. A multilevel deep network is proposed to formulate this learning, which utilizes three particular AS-OCT regions based on clinical priors: 1) the global anterior segment structure; 2) local iris region; and 3) anterior chamber angle (ACA) patch. In our method, a sliding window-based detector is designed to localize the ACA region, which addresses ACA detection as a regression task. Then, three parallel subnetworks are applied to extract AS-OCT representations for the global image and at clinically relevant local regions. Finally, the extracted deep features of these subnetworks are concatenated into one fully connected layer to predict the angle-closure detection result. In the experiments, our system is shown to surpass previous detection methods and other deep learning systems on two clinical AS-OCT datasets.

A Deep Learning System for Automated Angle-Closure Detection in Anterior Segment Optical Coherence Tomography Images.

PURPOSE: Anterior segment optical coherence tomography (AS-OCT) provides an objective imaging modality for visually identifying anterior segment structures. An automated detection system could assist ophthalmologists in interpreting AS-OCT images for the presence of angle closure. DESIGN: Development of an artificial intelligence automated detection system for the presence of angle closure. METHODS: A deep learning system for automated angle-closure detection in AS-OCT images was developed, and this was compared with another automated angle-closure detection system based on quantitative features. A total of 4135 Visante AS-OCT images from 2113 subjects (8270 anterior chamber angle images with 7375 open-angle and 895 angle-closure) were examined. The deep learning angle-closure detection system for a 2-class classification problem was tested by 5-fold cross-validation. The deep learning system and the automated angle-closure detection system based on quantitative features were evaluated against clinicians' grading of AS-OCT images as the reference standard. RESULTS: The area under the receiver operating characteristic curve of the system using quantitative features was 0.90 (95% confidence interval [CI] 0.891-0.914) with a sensitivity of 0.79 ± 0.037 and a specificity of 0.87 ± 0.009, while the area under the receiver operating characteristic curve of the deep learning system was 0.96 (95% CI 0.953-0.968) with a sensitivity of 0.90 ± 0.02 and a specificity of 0.92 ± 0.008, against clinicians' grading of AS-OCT images as the reference standard. CONCLUSIONS: The results demonstrate the potential of the deep learning system for angle-closure detection in AS-OCT images.

Structure-Preserving Guided Retinal Image Filtering and Its Application for Optic Disk Analysis.

Retinal fundus photographs have been used in the diagnosis of many ocular diseases such as glaucoma, pathological myopia, age-related macular degeneration, and diabetic retinopathy. With the development of computer science, computer aided diagnosis has been developed to process and analyze the retinal images automatically. One of the challenges in the analysis is that the quality of the retinal image is often degraded. For example, a cataract in human lens will attenuate the retinal image, just as a cloudy camera lens which reduces the quality of a photograph. It often obscures the details in the retinal images and posts challenges in retinal image processing and analyzing tasks. In this paper, we approximate the degradation of the retinal images as a combination of human-lens attenuation and scattering. A novel structure-preserving guided retinal image filtering (SGRIF) is then proposed to restore images based on the attenuation and scattering model. The proposed SGRIF consists of a step of global structure transferring and a step of global edge-preserving smoothing. Our results show that the proposed SGRIF method is able to improve the contrast of retinal images, measured by histogram flatness measure, histogram spread, and variability of local luminosity. In addition, we further explored the benefits of SGRIF for subsequent retinal image processing and analyzing tasks. In the two applications of deep learning-based optic cup segmentation and sparse learning-based cup-to-disk ratio (CDR) computation, our results show that we are able to achieve more accurate optic cup segmentation and CDR measurements from images processed by SGRIF.

Joint Optic Disc and Cup Segmentation Based on Multi-Label Deep Network and Polar Transformation.

Glaucoma is a chronic eye disease that leads to irreversible vision loss. The cup to disc ratio (CDR) plays an important role in the screening and diagnosis of glaucoma. Thus, the accurate and automatic segmentation of optic disc (OD) and optic cup (OC) from fundus images is a fundamental task. Most existing methods segment them separately, and rely on hand-crafted visual feature from fundus images. In this paper, we propose a deep learning architecture, named M-Net, which solves the OD and OC segmentation jointly in a one-stage multi-label system. The proposed M-Net mainly consists of multi-scale input layer, U-shape convolutional network, side-output layer, and multi-label loss function. The multi-scale input layer constructs an image pyramid to achieve multiple level receptive field sizes. The U-shape convolutional network is employed as the main body network structure to learn the rich hierarchical representation, while the side-output layer acts as an early classifier that produces a companion local prediction map for different scale layers. Finally, a multi-label loss function is proposed to generate the final segmentation map. For improving the segmentation performance further, we also introduce the polar transformation, which provides the representation of the original image in the polar coordinate system. The experiments show that our M-Net system achieves state-of-the-art OD and OC segmentation result on ORIGA data set. Simultaneously, the proposed method also obtains the satisfactory glaucoma screening performances with calculated CDR value on both ORIGA and SCES datasets.

Disc-Aware Ensemble Network for Glaucoma Screening From Fundus Image.

Glaucoma is a chronic eye disease that leads to irreversible vision loss. Most of the existing automatic screening methods first segment the main structure and subsequently calculate the clinical measurement for the detection and screening of glaucoma. However, these measurement-based methods rely heavily on the segmentation accuracy and ignore various visual features. In this paper, we introduce a deep learning technique to gain additional image-relevant information and screen glaucoma from the fundus image directly. Specifically, a novel disc-aware ensemble network for automatic glaucoma screening is proposed, which integrates the deep hierarchical context of the global fundus image and the local optic disc region. Four deep streams on different levels and modules are, respectively, considered as global image stream, segmentation-guided network, local disc region stream, and disc polar transformation stream. Finally, the output probabilities of different streams are fused as the final screening result. The experiments on two glaucoma data sets (SCES and new SINDI data sets) show that our method outperforms other state-of-the-art algorithms.

Extract-and-match geometric corner and step pattern approach for registration of fluoroscopic X-ray sequences.

This paper presents a method to extract-and-match robust corner features based on connecting edges from the edge maps, mainly formed by coronary vascular junctions in fluoroscopic x-ray sequence images. Such images are challenging due to the aperture problem. To overcome this, existing approaches attempt to extract vessels for registration. However, they are ineffective in poor quality images. Our approach describes the extracted robust corner features in a rotation invariant manner using step patterns, followed by matching them effectively. Experimental results show that our approach performs very well (above 80%) in a dataset of poor quality fluoroscopic x-ray image sequences without extensive processing such as segmentation or learning.

Fibroatheroma identification in Intravascular Optical Coherence Tomography images using deep features.

Identifying vulnerable plaque is important in coronary heart disease diagnosis. Recent emerged imaging modality, Intravascular Optical Coherence Tomography (IVOCT), has been proved to be able to characterize the appearance of vulnerable plaques. Comparing with the manual method, automated fibroatheroma identification would be more efficient and objective. Deep convolutional neural networks have been adopted in many medical image analysis tasks. In this paper, we introduce deep features to resolve fibroatheroma identification problem. Deep features which extracted using four deep convolutional neural networks, AlexNet, GoogLeNet, VGG-16 and VGG-19, are studied. And a dataset of 360 IVOCT images from 18 pullbacks are constructed to evaluate these features. Within these 360 images, 180 images are normal IVOCT images and the rest 180 images are IVOCT images with fibroatheroma. Here, one pullback belongs to one patient; leave-one-patient-out cross-validation is employed for evaluation. Data augmentation is applied on training set for each classification scheme. Linear support vector machine is conducted to classify the normal IVOCT image and IVOCT image with fibroatheroma. The experimental results show that deep features could achieve relatively high accuracy in fibroatheroma identification.

Similarity regularized sparse group lasso for cup to disc ratio computation.

Automatic cup to disc ratio (CDR) computation from color fundus images has shown to be promising for glaucoma detection. Over the past decade, many algorithms have been proposed. In this paper, we first review the recent work in the area and then present a novel similarity-regularized sparse group lasso method for automated CDR estimation. The proposed method reconstructs the testing disc image based on a set of reference disc images by integrating the similarity between testing and the reference disc images with the sparse group lasso constraints. The reconstruction coefficients are then used to estimate the CDR of the testing image. The proposed method has been validated using 650 images with manually annotated CDRs. Experimental results show an average CDR error of 0.0616 and a correlation coefficient of 0.7, outperforming other methods. The areas under curve in the diagnostic test reach 0.843 and 0.837 when manual and automatically segmented discs are used respectively, better than other methods as well.

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.

Segmentation and Quantification for Angle-Closure Glaucoma Assessment in Anterior Segment OCT.

Angle-closure glaucoma is a major cause of irreversible visual impairment and can be identified by measuring the anterior chamber angle (ACA) of the eye. The ACA can be viewed clearly through anterior segment optical coherence tomography (AS-OCT), but the imaging characteristics and the shapes and locations of major ocular structures can vary significantly among different AS-OCT modalities, thus complicating image analysis. To address this problem, we propose a data-driven approach for automatic AS-OCT structure segmentation, measurement, and screening. Our technique first estimates initial markers in the eye through label transfer from a hand-labeled exemplar data set, whose images are collected over different patients and AS-OCT modalities. These initial markers are then refined by using a graph-based smoothing method that is guided by AS-OCT structural information. These markers facilitate segmentation of major clinical structures, which are used to recover standard clinical parameters. These parameters can be used not only to support clinicians in making anatomical assessments, but also to serve as features for detecting anterior angle closure in automatic glaucoma screening algorithms. Experiments on Visante AS-OCT and Cirrus high-definition-OCT data sets demonstrate the effectiveness of our approach.

Speckle Reduction in 3D Optical Coherence Tomography of Retina by A-Scan Reconstruction.

Optical coherence tomography (OCT) is a micrometer-scale, cross-sectional imaging modality for biological tissue. It has been widely used for retinal imaging in ophthalmology. Speckle noise is problematic in OCT. A raw OCT image/volume usually has very poor image quality due to speckle noise, which often obscures the retinal structures. Overlapping scan is often used for speckle reduction in a 2D line-scan. However, it leads to an increase of the data acquisition time. Therefore, it is unpractical in 3D scan as it requires a much longer data acquisition time. In this paper, we propose a new method for speckle reduction in 3D OCT. The proposed method models each A -scan as the sum of underlying clean A -scan and noise. Based on the assumption that neighboring A -scans are highly similar in the retina, the method reconstructs each A -scan from its neighboring scans. In the method, the neighboring A -scans are aligned/registered to the A -scan to be reconstructed and form a matrix together. Then low rank matrix completion using bilateral random projection is utilized to iteratively estimate the noise and recover the underlying clean A -scan. The proposed method is evaluated through the mean square error, peak signal to noise ratio and the mean structure similarity index using high quality line-scan images as reference. Experimental results show that the proposed method performs better than other methods. In addition, the subsequent retinal layer segmentation also shows that the proposed method makes the automatic retinal layer segmentation more accurate. The technology can be embedded into current OCT machines to enhance the image quality for visualization and subsequent analysis such as retinal layer segmentation.

Automatic anterior chamber angle structure segmentation in AS-OCT image based on label transfer.

The anterior chamber angle (ACA) plays an important role for diagnosis and treatment of angle-closure glaucoma. Anterior Segment Optical Coherence Tomography (AS-OCT) imaging is qualitative and quantitative assessment for the ACA structure. In this paper, we propose a novel fully automatic segmentation method for anterior chamber angle structure in AS-OCT. In our method, the initial labels are obtained by using label transfer from the AS-OCT reference dataset. Then, these labels are refined and utilized as the landmarks to support the structure segmentation. Finally, the major clinical structures: corneal boundary, iris region, and trabecular-iris contact, are extracted as the segmentation result. Experiments show that our proposed method achieve the satisfactory segmentation performance on the clinical AS-OCT dataset. Our proposed method has potential in the applications of clinical ACA parameter measurement and automatic glaucoma classification.

Automated basal cell carcinoma detection in high-definition optical coherence tomography.

Basal cell carcinoma (BCC) is the most common non-melanoma skin cancer. Conventional diagnosis of BCC requires invasive biopsies. Recently, a high-definition optical coherence tomography (HD-OCT) technique has been developed, which provides a non-invasive in vivo imaging method of skin. Good agreements of BCC features between HD-OCT images and histopathological architecture have been found. Therefore it is possible to automatically detect BCC using HD-OCT. This paper presents a novel BCC detection method that consists of four steps: graph based skin surface segmentation, surface flattening, deep feature extraction and the BCC classification. The effectiveness of the proposed method is well demonstrated on a dataset of 5,040 images. It can therefore serve as an automatic tool for screening BCC.

Automated in vivo 3D high-definition optical coherence tomography skin analysis system.

The in vivo assessment and visualization of skin structures can be performed through the use of high resolution optical coherence tomography imaging, also known as HD-OCT. However, the manual assessment of such images can be exhaustive and time consuming. In this paper, we present an analysis system to automatically identify and quantify the skin characteristics such as the topography of the surface of the skin and thickness of the epidermis in HD-OCT images. Comparison of this system with manual clinical measurements demonstrated its potential for automatic objective skin analysis and diseases diagnosis. To our knowledge, this is the first report of an automated system to process and analyse HD-OCT skin images.

Sparse dissimilarity-constrained coding for glaucoma screening.

OBJECTIVE: Glaucoma is an irreversible chronic eye disease that leads to vision loss. As it can be slowed down through treatment, detecting the disease in time is important. However, many patients are unaware of the disease because it progresses slowly without easily noticeable symptoms. Currently, there is no effective method for low-cost population-based glaucoma detection or screening. Recent studies have shown that automated optic nerve head assessment from 2-D retinal fundus images is promising for low-cost glaucoma screening. In this paper, we propose a method for cup to disc ratio (CDR) assessment using 2-D retinal fundus images. METHODS: In the proposed method, the optic disc is first segmented and reconstructed using a novel sparse dissimilarity-constrained coding (SDC) approach which considers both the dissimilarity constraint and the sparsity constraint from a set of reference discs with known CDRs. Subsequently, the reconstruction coefficients from the SDC are used to compute the CDR for the testing disc. RESULTS: The proposed method has been tested for CDR assessment in a database of 650 images with CDRs manually measured by trained professionals previously. Experimental results show an average CDR error of 0.064 and correlation coefficient of 0.67 compared with the manual CDRs, better than the state-of-the-art methods. Our proposed method has also been tested for glaucoma screening. The method achieves areas under curve of 0.83 and 0.88 on datasets of 650 and 1676 images, respectively, outperforming other methods. CONCLUSION: The proposed method achieves good accuracy for glaucoma detection. SIGNIFICANCE: The method has a great potential to be used for large-scale population-based glaucoma screening.

Epidermal segmentation in high-definition optical coherence tomography.

Epidermis segmentation is a crucial step in many dermatological applications. Recently, high-definition optical coherence tomography (HD-OCT) has been developed and applied to imaging subsurface skin tissues. In this paper, a novel epidermis segmentation method using HD-OCT is proposed in which the epidermis is segmented by 3 steps: the weighted least square-based pre-processing, the graph-based skin surface detection and the local integral projection-based dermal-epidermal junction detection respectively. Using a dataset of five 3D volumes, we found that this method correlates well with the conventional method of manually marking out the epidermis. This method can therefore serve to effectively and rapidly delineate the epidermis for study and clinical management of skin diseases.