Color Fundus Images Research Articles

Multiple Ocular Disease Diagnosis Using Fundus Images Based on Multi-Label Deep Learning Classification

Designing computer-aided diagnosis (CAD) systems that can automatically detect ocular diseases (ODs) has become an active research field in the health domain. Although the human eye might have more than one OD simultaneously, most existing systems are designed to detect specific eye diseases. Therefore, it is crucial to develop new CAD systems that can detect multiple ODs simultaneously. This paper presents a novel multi-label convolutional neural network (ML-CNN) system based on ML classification (MLC) to diagnose various ODs from color fundus images. The proposed ML-CNN-based system consists of three main phases: the preprocessing phase, which includes normalization and augmentation using several transformation processes, the modeling phase, and the prediction phase. The proposed ML-CNN consists of three convolution (CONV) layers and one max pooling (MP) layer. Then, two CONV layers are performed, followed by one MP and dropout (DO). After that, one flatten layer is performed, followed by one fully connected (FC) layer. We added another DO once again, and finally, one FC layer with 45 nodes is performed. The system outputs the probabilities of all 45 diseases in each image. We validated the model by using cross-validation (CV) and measured the performance by five different metrics: accuracy (ACC), recall, precision, Dice similarity coefficient (DSC), and area under the curve (AUC). The results are 94.3%, 80%, 91.5%, 99%, and 96.7%, respectively. The comparisons with the existing built-in models, such as MobileNetV2, DenseNet201, SeResNext50, InceptionV3, and InceptionresNetv2, demonstrate the superiority of the proposed ML-CNN model.

Assessment of image quality on color fundus retinal images using the automatic retinal image analysis

Image quality assessment is essential for retinopathy detection on color fundus retinal image. However, most studies focused on the classification of good and poor quality without considering the different types of poor quality. This study developed an automatic retinal image analysis (ARIA) method, incorporating transfer net ResNet50 deep network with the automatic features generation approach to automatically assess image quality, and distinguish eye-abnormality-associated-poor-quality from artefact-associated-poor-quality on color fundus retinal images. A total of 2434 retinal images, including 1439 good quality and 995 poor quality (483 eye-abnormality-associated-poor-quality and 512 artefact-associated-poor-quality), were used for training, testing, and 10-ford cross-validation. We also analyzed the external validation with the clinical diagnosis of eye abnormality as the reference standard to evaluate the performance of the method. The sensitivity, specificity, and accuracy for testing good quality against poor quality were 98.0%, 99.1%, and 98.6%, and for differentiating between eye-abnormality-associated-poor-quality and artefact-associated-poor-quality were 92.2%, 93.8%, and 93.0%, respectively. In external validation, our method achieved an area under the ROC curve of 0.997 for the overall quality classification and 0.915 for the classification of two types of poor quality. The proposed approach, ARIA, showed good performance in testing, 10-fold cross validation and external validation. This study provides a novel angle for image quality screening based on the different poor quality types and corresponding dealing methods. It suggested that the ARIA can be used as a screening tool in the preliminary stage of retinopathy grading by telemedicine or artificial intelligence analysis.

Automated glaucoma assessment from color fundus images using structural and texture features

Glaucoma is an eye disease that damages optic nerves in eye leading to loss of vision. It is recognized to be the one of the major cause of blindness across the world. The detection of glaucoma in its early stage followed by treatment is the only way forward because damage done by the disease is irreversible. Hence there is a requirement of a large scale glaucoma screening. Manual screening of glaucoma at large scale is quite a challenging task due to lack of skilled manpower in ophthalmology. To address the above issue, many works have been proposed towards automating glaucoma detection suitable for large scale screening. On similar line, we in this proposed work present a novel method for automated glaucoma assessment from color fundus images (CFI) using structural features and texture features. Structural features such as cup to disk ratio (CDR), rim to disc ratio (RDR) and texture features extracted through the Grey-Level Co occurrence Matrix (GLCM) and entropy of images, histogram of gradient (HOG) features are used for glaucoma assessment. The classifiers used are Linear Discriminant Analysis (LDA), Random forest, Support Vector Machine (SVM) and Naïve Bayesian. The results obtained by the proposed work yield an overall efficiency of over 89%.

Early Glaucoma Detection by Using Style Transfer to Predict Retinal Nerve Fiber Layer Thickness Distribution on the Fundus Photograph.

ObjectiveWe aimed to develop a deep learning (DL)–based algorithm for early glaucoma detection based on color fundus photographs that provides information on defects in the retinal nerve fiber layer (RNFL) and its thickness from the mapping and translating relations of spectral domain OCT (SD-OCT) thickness maps.DesignDeveloping and evaluating an artificial intelligence detection tool.SubjectsPretraining paired data of color fundus photographs and SD-OCT images from 189 healthy participants and 371 patients with early glaucoma were used.MethodsThe variational autoencoder (VAE) network training architecture was used for training, and the correlation between the fundus photographs and RNFL thickness distribution was determined through the deep neural network. The reference standard was defined as a vertical cup-to-disc ratio of ≥0.7, other typical changes in glaucomatous optic neuropathy, and RNFL defects. Convergence indicates that the VAE has learned a distribution that would enable us to produce corresponding synthetic OCT scans.Main Outcome MeasuresSimilarly to wide-field OCT scanning, the proposed model can extract the results of RNFL thickness analysis. The structural similarity index measure (SSIM) and peak signal-to-noise ratio (PSNR) were used to assess signal strength and the similarity in the structure of the color fundus images converted to an RNFL thickness distribution model. The differences between the model-generated images and original images were quantified.ResultsWe developed and validated a novel DL-based algorithm to extract thickness information from the color space of fundus images similarly to that from OCT images and to use this information to regenerate RNFL thickness distribution images. The generated thickness map was sufficient for clinical glaucoma detection, and the generated images were similar to ground truth (PSNR: 19.31 decibels; SSIM: 0.44). The inference results were similar to the OCT-generated original images in terms of the ability to predict RNFL thickness distribution.ConclusionsThe proposed technique may aid clinicians in early glaucoma detection, especially when only color fundus photographs are available.

Optic disc localization in fundus images through accumulated directional and radial blur analysis

Optic disc localization, a key preprocessing step in the analysis of color fundus images for diagnoses of eye diseases and the localization of various anatomical structures, is particularly challenging when input retina images contain abnormalities. In such cases, the disc can be confused with other anatomical structures such as fovea, exudates, vessel tree extraction, and retinopathy-related lesions. Herein, we present a method for effective optic disc detection and localization based on color and blur analysis. In this method, the input color fundus image is converted to CIE L*a*b* color space to enhance optic disc appearance and contrast, and the accumulated directional blur and extended-maxima transform are then applied to precisely extract optic disc candidates. Subsequently, radial blur is applied to each candidate to obtain better profiles and thus distinguish the optic disc from other candidates. Finally, the full width at 80% maximum (FW80M) metric is used to select the optic disc. The performance of the proposed method is evaluated using well-studied data sets, and comparison of the obtained results with those of state-of-the-art techniques reveals the effectiveness of our method and shows that it can precisely locate the disc position not only in normal cases but also in the presence of exudates and abnormalities.

En‐ConvNet: A novel approach for glaucoma detection from color fundus images using ensemble of deep convolutional neural networks

AbstractGlaucomatous optic neuropathy is the preeminent cause of incurable vision impairment and blindness across the world. Manual interpretation of the pathological structures in fundus images is time‐consuming and requires the expertise of a competent specialist. With the development of deep learning approaches, automated glaucoma diagnosis is easy and effective for larger screening. Convolutional neural networks, in particular, have emerged as a promising choice for glaucoma detection from fundus images due to their remarkable success in image classification. Transferring the optimized weights from a pre‐trained model expedites and simplifies the training process of deep neural network. In this paper, a deep ensemble model using the stacking ensemble learning technique is developed to attain the optimum performance for the classification of glaucomatous and normal images. Thirteen pre‐trained models such as Alexnet, Googlenet, VGG‐16, VGG‐19, Squeezenet, Resnet‐18, Resnet‐50, Resnet‐101, Efficientnet‐b0, Mobilenet‐v2, Densenet‐201, Inception‐v3, and Xception are implemented. Their performance is compared in 65 different configurations, comprising 13 CNN architectures and five various classification approaches. A two‐stage ensemble selection technique is proposed to select the optimal configurations. Selected configurations are pooled using a probability averaging technique. The final classification is performed using an SVM classifier. In this work, publicly available databases are modified (such as: DRISHTI‐GS1‐R, ORIGA‐R, RIM‐ONE2‐R, LAG‐R, and ACRIMA‐R) based on oversampling data‐level technique for validating the performance of deep ensemble model. Ensembling the best configurations reports an overall classification accuracy of 93.4%, 79.6%, 91.3%, 99.5%, and 99.6% in DRISHTI‐GS1‐R, ORIGA‐R, RIM‐ONE2‐R, ACRIMA‐R, and LAG‐R databases, respectively.

Discriminative convolution neural network architecture for diagnosis of diabetic retinopathy through classification and progression prediction of lesions grading in color fundus images of retina

Diabetic Retinopathy (DR) is a common complication of diabetes mellitus, which causes lesions on the retina that effect vision. Many researchers have supported the ophthalmologists to diagnosis and classify the stage of the disease in the retinal fundus images using machine learning model. Machine learning models are less impressive for several staging diseases due to clinical grading. In order to develop a system capable of classifying the lesion grading images on disease pathology, a unique deep learning architecture named as discriminative Convolution Neural Network has been employed towards diagnosis of diabetic retinopathy through classification and progression prediction of lesion grading in fundus image of Retina. Initially Image Pre-processing has been carried out using wiener filter to remove the noise and Contrast limited adaptive histogram equalization for image enhancement. Pre-processed image has been processed using Oriented Fast and Rotated Brief for feature descriptors. It contains like Optic distance, Fovea, blood Vessel, Blot haemorrhages, Exudate number, exudates area, Macular Edema, Bifurcation, Shannon entrophy, Kapur Entropy and Renyis Entropy, LBP entropy, LBP energy and Microaneurysm. On these obtained features, feature reduction has to be carried out using principle component analysis to eliminate the irrelevant features before the onset of the process.

Ar-HGSO: Autoregressive-Henry Gas Sailfish Optimization enabled deep learning model for diabetic retinopathy detection and severity level classification

Diabetic Retinopathy (DR) is one the most important problems of diabetics and it directs to the main cause of blindness. When proper treatment is afforded for DR patients, almost 90% of patients are protected from visual damage. DR does not produce any symptoms at the initial phase of the disease, thus various physical assessments, namely pupil dilation, visual acuity test, and so on are needed for DR disease detection. It is more complex to detect the DR during manual testing, because of the variations and complications of DR. The early detection and appropriate treatment assist to prevent vision loss for DR patients. Thus, it is very indispensable to categorize the levels and severity of DR for recommendation of essential treatment. In this paper, Autoregressive-Henry Gas Sailfish Optimization (Ar-HGSO)-based deep learning technique is proposed for DR detection and severity level classification of DR and Macular Edema (ME) based on color fundus images. The segmentation process is more essential for proper detection and classification process, which segments the image into various subgroups. The Deep Learning approach is utilized for effective identification of DR and severity classification of DR and ME. Moreover, the deep learning technique is trained by the designed Ar-HGSO scheme for obtaining better performance. The performance of the devised technique is evaluated using the IDRID dataset and DDR dataset. The introduced Ar-HGSO-based deep learning approach obtained better performance than other existing DR detection and classification techniques with regards to testing accuracy, sensitivity, and specificity of 0.9142, 0.9254, and 0.9142 using the IDRID dataset.

Reflectance evaluation of eye fundus structures with a visible and near-infrared multispectral camera.

We examined the spectral reflectance of fundus structures in the visible and near-infrared (400-1300 nm) range for contributing to the medical diagnosis of fundus diseases. Spectral images of healthy eye fundus and other ocular diseases were acquired using a novel multispectral fundus camera. Reflectance metrics were computed based on contrast to analyze the spectral features. Significant differences were observed among the structures in healthy and diseased eye fundus. Specifically, near-infrared analysis allows imaging of deeper layers, such as the choroid, which, to date, has not been retrieved using traditional color fundus cameras. Pathological structures, which were hardly observable in color fundus images owing to metamerism, were also revealed by the developed multispectral fundus camera.

A computer-aided diagnosis system for detecting various diabetic retinopathy grades based on a hybrid deep learning technique

Diabetic retinopathy (DR) is a serious disease that may cause vision loss unawares without any alarm. Therefore, it is essential to scan and audit the DR progress continuously. In this respect, deep learning techniques achieved great success in medical image analysis. Deep convolution neural network (CNN) architectures are widely used in multi-label (ML) classification. It helps in diagnosing normal and various DR grades: mild, moderate, and severe non-proliferative DR (NPDR) and proliferative DR (PDR). DR grades are formulated by appearing multiple DR lesions simultaneously on the color retinal fundus images. Many lesion types have various features that are difficult to segment and distinguished by utilizing conventional and hand-crafted methods. Therefore, the practical solution is to utilize an effective CNN model. In this paper, we present a novel hybrid, deep learning technique, which is called E-DenseNet. We integrated EyeNet and DenseNet models based on transfer learning. We customized the traditional EyeNet by inserting the dense blocks and optimized the resulting hybrid E-DensNet model’s hyperparameters. The proposed system based on the E-DenseNet model can accurately diagnose healthy and different DR grades from various small and large ML color fundus images. We trained and tested our model on four different datasets that were published from 2006 to 2019. The proposed system achieved an average accuracy (ACC), sensitivity (SEN), specificity (SPE), Dice similarity coefficient (DSC), the quadratic Kappa score (QKS), and the calculation time (T) in minutes (m) equal 91.2%, 96%, 69%, 92.45%, 0.883, and 3.5m respectively. The experiments show promising results as compared with other systems.Graphical abstract

MULTIMODAL ASSESSMENTS OF DRUSENOID PIGMENT EPITHELIAL DETACHMENTS IN THE AGE-RELATED EYE DISEASE STUDY 2 ANCILLARY SPECTRAL-DOMAIN OPTICAL COHERENCE TOMOGRAPHY STUDY COHORT.

To identify features correlating with drusenoid pigment epithelial detachment (DPED) progression in the Age-Related Eye Disease Study 2 Ancillary spectral-domain optical coherence tomography study cohort. In this retrospective analysis of a prospective longitudinal study, eyes with intermediate age-related macular degeneration and DPEDs were followed longitudinally with annual multimodal imaging. Thirty-one eyes of 25 participants (mean age 72.6 years) in the Age-Related Eye Disease Study 2 Ancillary spectral-domain OCT substudy (A2A study) had DPED identified in color fundus images. Spectral-domain optical coherence tomography inspection confirmed a subretinal pigment epithelium drusenoid elevation of ≥433 µm diameter in 25 eyes (80.6%). Twenty-four of these eyes were followed longitudinally (median 4.0 years), during which 7 eyes (29.2%) underwent DPED collapse (with 3/7 further progressing to geographic atrophy), 6 (25.0%) developing neovascular age-related macular degeneration, and 11 (45.8%) maintaining DPED persistence without late age-related macular degeneration. On Kaplan-Meier analysis, mean time to DPED collapse was 3.9 years. Both DPED collapse and progression to neovascular age-related macular degeneration were preceded by the presence of hyperreflective foci over the DPED. The natural history of DPED comprises collapse (sometimes followed by the development of atrophy), vascularization followed by exudation, or DPED persistence. Spectral-domain optical coherence tomography can confirm retinal pigment epithelial elevation caused by drusenoid accumulation and facilitate the identification of high-risk features that correlate with progression.

LightEyes: A Lightweight Fundus Segmentation Network for Mobile Edge Computing.

Fundus is the only structure that can be observed without trauma to the human body. By analyzing color fundus images, the diagnosis basis for various diseases can be obtained. Recently, fundus image segmentation has witnessed vast progress with the development of deep learning. However, the improvement of segmentation accuracy comes with the complexity of deep models. As a result, these models show low inference speeds and high memory usages when deploying to mobile edges. To promote the deployment of deep fundus segmentation models to mobile devices, we aim to design a lightweight fundus segmentation network. Our observation comes from the fact that high-resolution representations could boost the segmentation of tiny fundus structures, and the classification of small fundus structures depends more on local features. To this end, we propose a lightweight segmentation model called LightEyes. We first design a high-resolution backbone network to learn high-resolution representations, so that the spatial relationship between feature maps can be always retained. Meanwhile, considering high-resolution features means high memory usage; for each layer, we use at most 16 convolutional filters to reduce memory usage and decrease training difficulty. LightEyes has been verified on three kinds of fundus segmentation tasks, including the hard exudate, the microaneurysm, and the vessel, on five publicly available datasets. Experimental results show that LightEyes achieves highly competitive segmentation accuracy and segmentation speed compared with state-of-the-art fundus segmentation models, while running at 1.6 images/s Cambricon-1A speed and 51.3 images/s GPU speed with only 36k parameters.

Using artificial intelligence reading label system in diabetic retinopathy grading training of junior ophthalmology residents and medical students

PurposeEvaluate the efficiency of using an artificial intelligence reading label system in the diabetic retinopathy grading training of junior ophthalmology resident doctors and medical students.MethodsLoading 520 diabetic retinopathy patients’ colour fundus images into the artificial intelligence reading label system. Thirteen participants, including six junior ophthalmology residents and seven medical students, read the images randomly for eight rounds. They evaluated the grading of images and labeled the typical lesions. The sensitivity, specificity, and kappa scores were determined by comparison with the participants’ results and diagnosis gold standards.ResultsThrough eight rounds of reading, the average kappa score was elevated from 0.67 to 0.81. The average kappa score for rounds 1 to 4 was 0.77, and the average kappa score for rounds 5 to 8 was 0.81. The participants were divided into two groups. The participants in Group 1 were junior ophthalmology resident doctors, and the participants in Group 2 were medical students. The average kappa score of Group 1 was elevated from 0.71 to 0.76. The average kappa score of Group 2 was elevated from 0.63 to 0.84.ConclusionThe artificial intelligence reading label system is a valuable tool for training resident doctors and medical students in performing diabetic retinopathy grading.

Automated lesion segmentation and quantification for prediction of paradoxical worsening in patients with tubercular serpiginous-like choroiditis

To develop and evaluate a fully automated pipeline that analyzes color fundus images in patients with tubercular serpiginous-like choroiditis (TB SLC) for prediction of paradoxical worsening (PW). In this retrospective study, patients with TB SLC with a follow-up of 9 months after initiation of anti-tubercular therapy were included. A fully automated custom-designed pipeline was developed which was initially tested using 12 baseline color fundus photographs for assessment of repeatability. After confirming reliability using Bland–Altman plots and intraclass correlation coefficient (ICC), the pipeline was deployed for all patients. The images were preprocessed to exclude the optic nerve from the fundus photo using a single-shot trainable WEKA segmentation algorithm. Two automatic thresholding algorithms were applied, and quantitative metrics were generated. These metrics were compared between PW + and PW- groups using non-parametric tests. A logistic regression model was used to predict probability of PW for assessing binary classification performance and receiver operator curves were generated to choose a sensitivity-optimized threshold. The study included 139 patients (139 eyes; 92 males and 47 females; mean age: 44.8 ± 11.3 years) with TB SLC. Pilot analysis of 12 images showed an excellent ICC for measuring the mean area, intensity, and integrated pixel intensity (all ICC > 0.89). The PW + group had significantly higher mean lesion area (p = 0.0152), mean pixel intensity (p = 0.0181), and integrated pixel intensity (p < 0.0001) compared to the PW- group. Using a sensitivity optimized threshold cut-off for mean pixel intensity, an area under the curve of 0.87 was achieved (sensitivity: 96.80% and specificity: 72.09%). Automated calculation of lesion metrics such as mean pixel intensity and segmented area in TB SLC is a novel approach with good repeatability in predicting PW during the follow-up.

Detection of microaneurysms in color fundus images based on local Fourier transform

Retinal microaneurysm (MA) is one of the early clinical signs of diabetic retinopathy (DR), which is essential for the early diagnosis of DR. Its shape is often shown as a red dot, which makes detection of MA challenging because of its tiny size. There are many effective MA automatic detection methods having been proposed, but these methods are hard to describe the overall shape of Mas. This paper proposes an automatic MA detection method based on local Fourier transform which splits the original image into small independent images and calculates their local Fourier transforms in isolation. After that, the object is not the whole image but a series of independent small images. We also extract a statistical feature to check for normality and design a new feature to represent the difference among every direction. After that, a group of candidate points are extracted from the images. Finally, some features are extracted from the candidate points, and the confidence scores of the candidate points are obtained by using random forest algorithm for classification. The proposed method obtained a FROC curve score of 0.847 on the E-Ophtha MA database, which is better than most existing methods. In the ROC database, although the FROC curve score is only 0.283, the AUC can reach 0.961, which is better than most methods. It can be seen that the proposed method can achieve a better result with low noise and large sample size.

Hybrid multi-kernel SVM algorithm for detection of microaneurysm in color fundus images.

Diabetic retinopathy (DR) is a chronic disease that may cause vision loss in diabetic patients. Microaneurysms which are characterized by small red spots on the retina due to fluid or blood leakage from the weak capillary wall often occur during the early stage of DR, making screening at this stage is essential. In this paper, an automatic screening system for early detection of DR in retinal images is developed using a combined shape and texture features. Due to minimum number of hand-crafted features, the computational burden is much reduced. The proposed hybrid multi-kernel support vector machine classifier is constructed by learning a kernel model formed as a combination of the base kernels. This approach outperforms the recent deep learning techniques in terms of the evaluation metrics. The efficiency of the proposed scheme is experimentally validated on three public datasets - Retinopathy Online Challenge, DIARETdB1, MESSIDOR, and AGAR300 (developed for this study). Studies reveal that the proposed model produced the best results of 0.503 in ROC dataset, 0.481 in DIARETdB1, and 0.464 in the MESSIDOR dataset in terms of FROC score. The AGAR300 database outperforms the existing MA detection algorithm in terms of FROC, AUC, F1 score, precision, sensitivity, and specificity which guarantees the robustness of this system.

A Novel Un-Supervised GAN for Fundus Image Enhancement with Classification Prior Loss

Fundus images captured for clinical diagnosis usually suffer from degradation factors due to variation in equipment, operators, or environment. These degraded fundus images need to be enhanced to achieve better diagnosis and improve the results of downstream tasks. As there is no paired low- and high-quality fundus image, existing methods mainly focus on supervised or semi-supervised learning methods for color fundus image enhancement (CFIE) tasks by utilizing synthetic image pairs. Consequently, domain gaps between real images and synthetic images arise. With respect to existing unsupervised methods, the most important low scale pathological features and structural information in degraded fundus images are prone to be erased after enhancement. To solve these problems, an unsupervised GAN is proposed for CFIE tasks utilizing adversarial training to enhance low quality fundus images. Synthetic image pairs are no longer required during the training. A specially designed U-Net with skip connection in our enhancement network can effectively remove degradation factors while preserving pathological features and structural information. Global and local discriminators adopted in the GAN lead to better illumination uniformity in the enhanced fundus image. To better improve the visual quality of enhanced fundus images, a novel non-reference loss function based on a pretrained fundus image quality classification network was designed to guide the enhancement network to produce high quality images. Experiments demonstrated that our method could effectively remove degradation factors in low-quality fundus images and produce a competitive result compared with previous methods in both quantitative and qualitative metrics.

Long-term Retinal Morphology and Functional Associations in Treated Neovascular Age-Related Macular Degeneration: Findings from the Inhibition of VEGF in Age-Related Choroidal Neovascularisation Trial

To describe the frequency of long-term morphologic features and their relationships with visual function in participants who exited the Inhibition of VEGF in Age-Related Choroidal Neovascularisation (IVAN; ISRCTN92166560) trial. Multicenter cohort study up to 7 years after enrollment. Patients enrolled in the IVAN trial, excluding participants who died or withdrew during the trial. Multimodal fundus images, best-corrected visual acuity (BCVA), and low-luminance visual acuity (LLVA) were obtained for a subset of 199 participants who attended a research visit. Clinical sites (n = 20) also provided all visual acuity and clinical information from usual care records for 532 participants and submitted the most recent color, OCT, and other fundus images for 468 participants to a reading center. Assessed the following from the most recent images: intralesional macular atrophy (ILMA) within the footprint of the neovascular lesion; hyperreflective material (HRM); intraretinal fluid (IRF); subretinal fluid (SRF); pigment epithelial detachment (PED); and disorganized retinal outer layers (DROLs). Cross-sectional relationships between morphologic features and BCVA/LLVA were estimated. Intralesional macular atrophy was present in 31.8% of the study eyes at IVAN exit (mean follow-up, 1.96 years) and 89.5% at the most recent imaging visit (mean follow-up, 6.18 years). Hyperreflective material, IRF, SRF, PED, and DROLs were present in 78.8%, 47.7%, 7.6%, 94.5%, and 55% of the study eyes, respectively. In the subset with complete imaging data, in eyes without DROL, the BCVA was worst in the thinnest outer fovea tertile (thinnest minus middle and thickest tertiles,-19.7 and-19.5 letters, respectively), whereas in eyes with DROL, the BCVA was worst in the thickest (thinnest and middle tertiles minus thickest, 12.5 and 12.2, respectively). Regression models showed that the presence of ILMA and HRM was independently associated with BCVA (22 letters worse [95% confidence interval {CI}, -11.2 to -32.8; P < 0.001] and 9.8 letters worse [95% CI, -0.1 to -19.4; P = 0.047], respectively). Subretinal fluid and foveal PED were associated with better BCVA (5.9 letters [95% CI,-7.9 to 19.7; P = 0.399] and 6.4 letters [95% CI,-1.1 to 14.0; P = 0.094], respectively). The model with LLVA was similar. A sensitivity analysis involving the entire eligible cohort yielded similar estimates. Macular atrophy and HRM were common after 7 years of follow-up and strongly associated with visual outcomes.

DCNN-based prediction model for detection of age-related macular degeneration from color fundus images

DCNN-based prediction model for detection of age-related macular degeneration from color fundus images

Cross-Camera External Validation for Artificial Intelligence Software in Diagnosis of Diabetic Retinopathy.

Aims To investigate the applicability of deep learning image assessment software VeriSee DR to different color fundus cameras for the screening of diabetic retinopathy (DR). Methods Color fundus images of diabetes patients taken with three different nonmydriatic fundus cameras, including 477 Topcon TRC-NW400, 459 Topcon TRC-NW8 series, and 471 Kowa nonmyd 8 series that were judged as “gradable” by one ophthalmologist were enrolled for validation. VeriSee DR was then used for the diagnosis of referable DR according to the International Clinical Diabetic Retinopathy Disease Severity Scale. Gradability, sensitivity, and specificity were calculated for each camera model. Results All images (100%) from the three camera models were gradable for VeriSee DR. The sensitivity for diagnosing referable DR in the TRC-NW400, TRC-NW8, and non-myd 8 series was 89.3%, 94.6%, and 95.7%, respectively, while the specificity was 94.2%, 90.4%, and 89.3%, respectively. Neither the sensitivity nor the specificity differed significantly between these camera models and the original camera model used for VeriSee DR development (p = 0.40, p = 0.065, respectively). Conclusions VeriSee DR was applicable to a variety of color fundus cameras with 100% agreement with ophthalmologists in terms of gradability and good sensitivity and specificity for the diagnosis of referable DR.