Search

Early detection of Diabetic Retinopathy (DR) and its grading has been a growing demand among researchers in this community. Computer-aided diagnostic (CAD) systems have the potential to enhance the sensitivity and effectiveness of early diagnoses, benefiting ophthalmic specialists by offering additional insights for more efficient treatment options. The proposed study addresses the challenges of improved detection of mild stage and the limited number of samples with fewer parameters. Fundus images are initially pre-processed for this task using resizing, augmentation and oversampling. Oversampling is employed to guarantee the balanced inclusion of images from every grade category throughout the training stage. The proposed approach utilizes a Convolutional Neural Network (CNN) to extract texture and vessel features separately from the fundus images. This methodology exploited Local Binary Pattern (LBP) for improved texture features before applying CNN. Similarly, we utilized Contrast Limited Adaptive Histogram Equalization (CLAHE) to enhance the blood vessels of the fundus images, enabling the extraction of relevant features using CNN. The extracted features are combined and classified using fully connected layers. The proposed approach is validated using standard datasets such as IDRiD, APTOS, DDR, and EyePACS with limited samples. The experimental results demonstrate that the proposed model in this research outperforms state-of-the-art models across all standard performance metrics, with classification accuracies of 92.46%, 98.08%, 95.66% and 88.84%. • Tackles early-stage DR detection challenges; implemented with limited sample sizes. • By using Local Binary Pattern for texture enhancement, Contrast Limited Adaptive Histogram Equalization for better vessel visibility, and fully connected layers for classification. • The proposed model achieves accuracies of 92.46%, 98.08%, 95.66%, and 88.84% on IDRiD, APTOS, DDR, and EyePACS datasets. [ABSTRACT FROM AUTHOR]

Optic nerve head (ONH) detection has been a crucial area of study in ophthalmology for years. However, the significant discrepancy between fundus image datasets, each generated using a single type of fundus camera, poses challenges to the generalizability of ONH detection approaches developed based on semantic segmentation networks. Despite the numerous recent advancements in general-purpose semantic segmentation methods using convolutional neural networks (CNNs) and Transformers, there is currently a lack of benchmarks for these state-of-the-art (SoTA) networks specifically trained for ONH detection. Therefore, in this article, we make contributions from three key aspects: network design, the publication of a dataset, and the establishment of a comprehensive benchmark. Our newly developed ONH detection network, referred to as ODFormer, is based upon the Swin Transformer architecture and incorporates two novel components: a multi-scale context aggregator and a lightweight bidirectional feature recalibrator. Our published large-scale dataset, known as TongjiU-DROD, provides multi-resolution fundus images for each participant, captured using two distinct types of cameras. Our established benchmark involves three datasets: DRIONS-DB, DRISHTI-GS1, and TongjiU-DROD, created by researchers from different countries and containing fundus images captured from participants of diverse races and ages. Extensive experimental results demonstrate that our proposed ODFormer outperforms other state-of-the-art (SoTA) networks in terms of performance and generalizability. Our dataset and source code are publicly available at https://mias.group/ODFormer. • A novel network, referred to as ODFormer, for fundus image segmentation. • A dataset for ONH detection, captured using two different cameras. • A benchmark of state-of-the-art semantic segmentation CNNs and Transformers. [ABSTRACT FROM AUTHOR]

Unsupervised anomaly detection methods aim to reduce the cost of manually annotating abnormal medical image datasets. However, since they are not trained on massive abnormal images, their discriminative capability may be low. In this paper, we present AugPaste, a novel one-shot anomaly detection framework for detecting diabetic retinopathy (DR) from fundus images. AugPaste utilizes true anomalies from a single annotated DR sample to synthesize a large amount of artificial DR fundus images. The framework begins with constructing a DR lesion bank through augmentation of randomly selected DR lesion patches. Synthesized DR samples are then generated by pasting lesion patches selected from the lesion bank into normal images using various prior knowledge-guided strategies. We finally train a classification network on the synthetic abnormal images along with true normal images for anomaly detection. Our tests on four public fundus image datasets show that AugPaste outperforms leading unsupervised and few-shot methods and rivals fully-supervised methods. The source code is available at https://github.com/Aidanvk/AugPaste. • AugPaste: a novel OSAD framework for DR detection in fundus images. • First to synthesize DR fundus images from real lesion data for anomaly detection. • AugPaste evaluated across four datasets, showing superior SOTA performance. [ABSTRACT FROM AUTHOR]

• Proposition of an accurate and fast method for glaucoma screening based on the vessel displacement. • Achievement of higher performance detection with smartphone captured fundus images. • Implementation of the method into a smartphone to provide a mobile aided screening system for glaucoma. The glaucoma is an eye disease that causes blindness when it progresses in an advanced stage. Early glaucoma diagnosis is essential to prevent the vision loss. However, early detection is not covered due to the lack of ophthalmologists and the limited accessibility to retinal image capture devices. In this paper, we present an automated method for glaucoma screening dedicated for Smartphone Captured Fundus Images (SCFIs). The implementation of the method into a smartphone associated to an optical lens for retina capturing leads to a mobile aided screening system for glaucoma. The challenge consists in insuring higher performance detection despite the moderate quality of SCFIs, with a reduced execution time to be adequate for the clinical use. The main idea consists in deducing glaucoma based on the vessel displacement inside the Optic Disk (OD), where the vessel tree remains sufficiently modeled on SCFIs. Within this objective, our major contribution consists in proposing: (1) a robust processing for locating vessel centroids in order to adequately model the vessel distribution, and (2) a feature vector that relevantly reflect two main glaucoma biomarkers in terms of vessel displacement. Furthermore, all processing steps are carefully chosen based on lower complexity, to be suitable for fast clinical screening. A first evaluation of our method is performed using the two public DRISHTI-DB and DRIONS-DB databases, where 99% and 95% accuracy, 96.77% and 97,5% specificity and 100% and 95% sensitivity are respectively achieved. Thereafter, the method is evaluated using two fundus image databases respectively captured through a smartphone and retinograph for the same persons. We achieve 100% accuracy using both databases which assesses the robustness of our method. In addition, the detection is performed on 0.027 and 0.029 second when executed respectively on the Samsung-M51 on the Samsung-A70 smartphone devices. Our proposed smartphone app provides a cost-effective and widely accessible mobile platform for early screening of glaucoma in remote clinics or areas with limited access to fundus cameras and ophthalmologists. [ABSTRACT FROM AUTHOR]

Automated segmentation of diabetic retinopathy (DR) lesions in fundus images is significant in computer-aided diagnosis. While numerous studies have tackled lesion segmentation, most have relied on traditional backbone networks for direct multi-scale learning, primarily focusing on the effective integration of multi-scale features. However, few studies have delved into efficient multi-scale learning specifically tailored for DR lesion segmentation. In previous research, parallel convolutional operations with varying kernel sizes or residual connections have been commonly employed for multi-scale learning. Nevertheless, the spatial resolution of features remains unchanged throughout the whole process. This limitation hinders multi-resolution interaction within these blocks, preventing them from fully embedding multi-resolution information. To this end, we present a Multi-step Scale Embedding Block (MSEB), comprising two branches that operate on different spatial resolutions to enhance multi-resolution interaction. Furthermore, to facilitate improved multi-scale learning, we have constructed a VGG-like encoder network based on MSEB. This encoder is complemented by a lightweight decoder designed to seamlessly integrate multi-scale information and restore lesion details. By combining the strengths of both the encoder and decoder networks, we have developed SEBNet for DR multi-lesion segmentation. Extensive experiments are conducted on the IDRiD, DDR, and FGADR datasets, and experimental results demonstrate that SEBNet achieves competitive performance compared to state-of-the-art methods. The source code will be available at https://github.com/guomugong/SEBNet. • We introduce a multi-step scale embedding block for multi-scale feature learning. • We present SEBNet for DR multi-lesion segmentation. • SEBNet achieves SOTA performance on three benchmark datasets. [ABSTRACT FROM AUTHOR]

• Deep GD model offers 99.35% accuracy for glaucoma classification and segmentation. • Uses EfficientNetB4 for classification with Aquila optimization for parameter tuning. • Outperforms AG-CNN, C-RGBGs-IF, U-Net, and cGAN by up to 21.5% in accuracy. • Segments glaucoma into OP, AC, NT, Secondary, and Pigmentary categories with V-Net. Glaucoma is a neuropathic ocular degenerative disease that decreases peripheral vision and, if left untreated, eventually leads to blindness. In this research, a novel Deep GD model has been proposed for early diagnosis of glaucoma from retinal fundus images. The input images are fed into the preprocessing such as enhancement, normalization and augmentation by denoising the image. The preprocessed images are fed into Cosine Cyclic Learning Rate Schedule for Snapshot Ensemble Convolutional Neural Network (CCRS Snap ECNN) for extracting the features. The efficientNetB4 (ENB4) network is utilized to classify the normal and glaucoma. Aquila optimization (AQO) algorithm is used for normalizing the parameters in the Deep GD to improve efficiency with low level of network complexity. Finally, the classified glaucoma images are fed to the segmentation process using V-Net. The segmentation results are open-angle (OP), angle-closure (AC), normal-tension (NT), secondary(S) and pigmentary(P). The classification accuracy of Glaucoma disease is 99.35% on average when it comes to classification and segmentation. The overall accuracy of Deep GD is improved over attention-based convolutional neural network for glaucoma detection (AG-CNN), concatenated RGBGs image features (C-RGBGs-IF), U-Net and generative adversarial network to the conditional configuration (cGAN) by 3.17%, 14.49%, 2.05% and 21.5% respectively. These results showed how accurate the planned Deep GD classification system is for Glaucoma disease. [ABSTRACT FROM AUTHOR]

Accurate segmentation of the fundus optic disc and optic cup is essential for the early detection of glaucoma, a leading cause of blindness. Despite the advancements in deep learning-based segmentation models, the reliance on large amounts of annotated data and the weak generalization ability remain a challenge. The Segment Anything Model (SAM) has demonstrated robust zero-shot generalization in various natural image segmentation tasks. However, its dependency on manual interactive prompts and the domain gap between natural and medical images are obstacles to its application in medical image segmentation. This study introduces OSAM-Fundus, a novel training-free framework for optic disc and optic cup segmentation using one reference image and its annotation. First, We propose an automatic prompt generation method that incorporates medical prior knowledge and hard-negative sampling into patch-level bidirectional matching, enabling cross-image and cross-domain prompt generation without manual clinician intervention. Then, to overcome SAM's inherent challenge in segmenting fuzzy edges, we employ a feature re-matching strategy to refine the false-positive regions potentially produced by SAM, ensuring consistent semantic object outputs in the target image. Evaluation on four public datasets, both in-domain and out-of-domain, our method achieves state-of-the-art performance with average dice scores of 80.43%, which is 7.28% higher than that of SAM using prompts from ground truth. This demonstrates that OSAM-Fundus is a highly efficient and robust segmentation method for fundus images. [Display omitted] • A training-free framework for optic cup and disc segmentation with one shot. • Automatically generates prompts for SAM without training or human interaction. • Medical prior knowledge aids feature matching across images and domains. • Feature rematching is proposed to correct SAM's false-positive prediction. [ABSTRACT FROM AUTHOR]

Many early ophthalmology-related diseases can be detected through fundus images. The cup-to-disc ratio (CDR) is an important criterion for the early screening and diagnosis of glaucoma. However, accurately calculating it is a challenging task primarily due to blurred boundaries and the physiological structure of the fundus. To alleviate these issues, this paper designs a location-constrained joint optic disc (OD) and optic cup (OC) segmentation method using the multiplex aggregation network (LC-MANet). This method incorporates independent segmentation and joint segmentation within a coarse-to-fine framework. In the independent segmentation stage, this paper utilizes the initial OD and OC as prior location auto-context information to constrain the joint segmentation of OD and OC. The proposed multiplex aggregation network combines features from the corresponding layer and across layers to capture global and local information. This paper employs a multi-channel fusion strategy in image preprocessing to suppress vascular interference in OD independent segmentation. To emphasize the superiority of our method, it is important to highlight the significant number of test images used in the evaluation. Our experiments are conducted on three public datasets, namely RIM-ONE (60 images), Drishti-GS (51 images), and REFUGE (400 images). Experimental results demonstrate that our method outperforms most others in terms of Dice Coefficient and absolute error. The accuracy of glaucoma screening achieved by our proposed method for the Drishti-GS, REFUGE, and RIM-ONE-r3 datasets is 0.9216, 0.9600, and 0.9000, respectively. The proposed framework successfully designs a multiplex aggregation network for a coarse-to-fine joint segmentation model and achieves state-of-the-art performance in OD and OC segmentation. Furthermore, it can be applied to other medical image segmentation tasks. [ABSTRACT FROM AUTHOR]

• Introduce regularizer based on the number of isolated objects for small vessel segmentation. • The number of isolated objects estimate from the topology of segmentation result by using Euler characteristics. • Euler characteristic-based regularizers proposed to pay attention to the small and disconnected vessel regions. Segmentation of retinal blood vessels is important for the analysis of diabetic retinopathy (DR). Existing methods do not prioritize the small and disconnected vessels for DR. With the aim of paying attention to the small and disconnected vessel regions, this study introduced Euler characteristics (EC) from topology to calculate the number of isolated objects on segmented vessel regions, which is the key contribution of this study. In addition, we utilized the number of isolated objects in a U-Net-like deep convolutional neural network (CNN) architecture as a regularizer to train the network for improving the connectivity between the pixels of the vessel regions. The proposed network performance of the regularizer based on EC in reconstructing vessel regions is compared over the network without our regularizer. Furthermore, the capacity of the proposed regularizer approach in enhancing the smoothness and pixel connectivity of the vessels is compared with graph-based smoothing (GS) and combined GS with isolated objects (GISO) regularizers for delineating blood vessel regions. The proposed approach achieved the area under the curve value of 0.982, which is much higher than the state-of-the-arts, and thus it is suggested that the proposed system could support accuracy and reliability in decision-making for DR detection. [ABSTRACT FROM AUTHOR]

• Gamma Correction of Illumination Component (GCIC) – A dedicated devignetting scheme for fundus images. • Has good capability to suppress unevenness of background illumination. • Generates output images which are natural in appearance. • Does not produce any colour artefacts. • Computationally fast. • Does not intolerably boost or scale down the background illumination. Fundus photography is an imaging modality exclusively used in ophthalmology for visualizing structures like macula, retina, and optic disc. The fundus camera has only one illumination source, which is situated at its center. Hence, structures away from the center will appear darker than naturally they are. This adverse effect caused by the uneven illumination is called as 'vignetting'. An algorithm termed as Gamma Correction of Illumination Component (GCIC) for devignetting fundus images is proposed in this paper. Inspired by the Retinex theory, the illumination component is computed with the help of a Maximum a Posteriori (MAP) estimator. The estimated illumination component after normalization is subjected to the Gamma correction to suppress its unevenness. GCIC exhibited comparatively low values of Average Gradient of the Illumination Component (AGIC), Lightness Order Error (LOE), and computational time. The proposed method gave a comparatively better performance in terms of the performance metrics, namely contrast-to-noise-ratio (CNR), peak-signal-to-noise-ratio (PSNR), structure similarity index (SSIM), and entropy. With respect to the cumulative performance, GCIC has been observed to be better than other devignetting algorithms in the literature, like Illumination Equalization model, Homomorphic Filtering, Adaptive Gamma Correction (AGC), Modified Sigmoid Transform (MST), Imran Qureshi et al. (2019), Zheng et al., Variation-based Fusion (VF) and Zhou's et al. GCIC corrects the uneven background illumination without scaling or boosting it intolerably. It produces output images, which are natural in appearance, free from color artefacts, and maintaining the sharpness of the fundus features. It is computationally fast as well. [ABSTRACT FROM AUTHOR]

This paper proposes a rectified complex spherical fuzzy set (rCTSFS) model that enables the phase term of a complex number to function truthfully to its inherent meaning of representing directions, phases, or color hues. In addition, this paper proposes the score and accuracy functions for rectified complex spherical fuzzy numbers (rCTSFn), which allows the three types of membership degrees of an rCTSFn to fulfill human judgment/intuition. The proposed rCTSFS model proves to be a productive extension of the complex spherical fuzzy set (CSFS), complex fuzzy set (CFS), and spherical fuzzy set (SFS) models. On the other hand, Dombi t -norms prove more flexible and comprehensive than some of the other families of triangular norms, such as the algebraic t -norms and the Einstein t -norms, due to the presence of a parameter γ. The parameter γ determines the amount of aggressiveness at estimating the maximum and the minimum of a population based on a sample obtained. Therefore, this paper proposes two Dombi-Choquet integral operators, namely, the rectified complex t -spherical fuzzy arithmetic Dombi-Choquet integral (r C T S F A γ , w λ ) operator and the rectified complex t -spherical fuzzy geometric Dombi-Choquet integral (r C T S F G γ , w λ ) operator. A multi-criteria decision making (abbr. MCDM) algorithm utilizing the two Dombi-Choquet integral operators is innovated. The proposed Dombi-Choquet MCDM algorithm for the rCTSFS model is applied to an MCDM problem related to diabetic retinopathy detection on five real-life fundus images of different severity levels taken from the Messidor2 dataset. Our newly proposed algorithm proves to be the only algorithm that yields the correct diagnostic results that match the hard truth. On the other hand, none of the 50 algorithms observed among recent works in literature can produce the correct diagnostic results, even after lending the fuzzification procedure innovated in this work to them. [ABSTRACT FROM AUTHOR]

Glaucoma is a progressive eye condition that can lead to permanent vision loss. Therefore, on-time detection of glaucoma is critical for making an effective treatment plan. In recent years, enormous attempts have been made to develop automated glaucoma classification systems using CNNs through images. In contrast, limited methods have been proposed for diagnosing different glaucoma stages. It is mainly owing to the lack of large publicly available labeled datasets. Also, fundus images exhibit a high inter-stage resemblance, redundant features and minute size variations of lesions, making the conventional CNNs difficult to classify multiple stages of glaucoma accurately. To address these challenges, this paper proposes a novel adapter and enhanced self-attention based CNN framework named AES-Net for effective classification of glaucoma stages. In particular, we propose a spatial adapter module on top of the backbone network for learning better feature representations and an enhanced self-attention module (ESAM) to capture global feature correlations among the relevant channels and spatial positions. The ESAM assists in capturing stage-specific and detailed-lesion features from the fundus images. Extensive experiments on two multi-stage glaucoma datasets indicate that our AES-Net surpasses CNN-based existing approaches. The Grad-CAM ++ visualization maps further confirm the effectiveness of our AES-Net. • A spatial adapter and enhanced self-attention based CNN, AES-Net, is proposed. • An enhanced self-attention module is proposed to learn class-specific features. • A comparative analysis with state-of-the-art attention mechanisms is performed. • Results on two datasets indicate the effectiveness of AES-Net over current methods. • The visualization results demonstrate the explainability and interpretability of the model. [ABSTRACT FROM AUTHOR]

Early cataract screening is essential to help patients prevent irreversible vision loss. Existing data-driven methods can automatically extract deep fundus features for automatic cataract classification. However, due to the lack of medical domain knowledge, these deep features may contain large pathological irrelevant information, thus reducing the classification accuracy of data-driven methods. To fully take the advantages of domain knowledge to enhance the discriminative representation and avoid the irrelevant impact, we comprehensively investigate how to incorporate it into data-driven models. Considering the vessel information and multi-granularity representation are two key points of domain knowledge for cataract diagnosis, we propose a vessel attention guided multi-granularity network (VAM-Net), which contains three main components. Specifically, the vessel-level segmentation subnet is firstly designed by leveraging U-shape segmentation network. The refined vessel information obtained in this step can be regarded as visual attention to enable deep network extract discriminative features. With the guidance of vessel attention, the global-local classification subnet is further proposed to eliminate the interference of irrelevant information and enhance the multi-granularity feature learning (including global structural features and local subtle features). Finally, the multiple levels predictions are integrated into the final cataract diagnosis decision via constructing multi-granularity ensemble network. In this manner, both domain knowledge and deep networks are organically combined for pursuing better diagnosis performance. By setting extensive comparative and ablation experiments, we evaluate the effectiveness of the proposed VAM-Net on the real-world cataract dataset (92.78% detection accuracy and 87.68% grading accuracy). • A attention guided multi-granularity network is proposed for cataract diagnosis. • We aim to incorporate medical domain information into data-driven method. • Extensive experiments are conducted to prove the effectiveness of our method. [ABSTRACT FROM AUTHOR]

Since the labels for medical images are challenging to collect in real scenarios, especially for rare diseases, fully supervised methods cannot achieve robust performance for clinical anomaly detection. Recent research tried to tackle this problem by training the anomaly detection framework using only normal data. Reconstruction-based methods, e.g., auto-encoder, achieved impressive performances in the anomaly detection task. However, most existing methods adopted the straightforward backbone architecture (i.e., encoder-and-decoder) for image reconstruction. The design of a skip connection, which can directly transfer information between the encoder and decoder, is rarely used. Since the existing U-Net has demonstrated the effectiveness of skip connections for image reconstruction tasks, in this paper, we first use the dynamic gating strategy to achieve the usage of skip connections in existing reconstruction-based anomaly detection methods and then propose a novel gating highway connection module to adaptively integrate skip connections into the framework and boost its anomaly detection performance, namely GatingAno. Furthermore, we formulate an auxiliary task, namely histograms of oriented gradients (HOG) prediction, to encourage the framework to exploit contextual information from fundus images in a self-driven manner, which increases the robustness of feature representation extracted from the healthy samples. Last but not least, to improve the model generalization for anomalous data, we introduce an adversarial strategy for the training of our multi-task framework. Experimental results on the publicly available datasets, i.e., IDRiD and ADAM, validate the superiority of our method for detecting abnormalities in retinal fundus images. The source code is available at https://github.com/WentianZhang-ML/GatingAno. • A gating highway connection module is proposed, which can be applied in a reconstruction-based anomaly detection method to exploit and constrain the intrinsic features during training for better anomaly detection performance. • We integrate histograms of oriented gradients (HOG) prediction as an auxiliary task in the reconstruction network to improve its sensitivity to abnormal images. • Adversarial learning is combined in our multi-task encoder–decoder network in the training phase to improve generalization ability for anomaly detection. • We evaluate our model on two publicly available datasets. The experimental results on image-level and pixel-level tasks prove the effectiveness of gating highway connections, HOG prediction, and adversarial learning for anomaly detection. [ABSTRACT FROM AUTHOR]

Glaucoma is one of the leading cause of blindness for over 60 million people around the world. Since a cure for glaucoma doesn't yet exist, early screening and diagnosis become critical for the prevention of the disease. Optic disc and optic cup evaluation are one of the preeminent steps for glaucoma diagnosis. A novel approach is developed in this paper for the identification of glaucoma using a segmentation based approach on the optic disc and optic cup. The Dhristi dataset was used to help improve performance on a small dataset. A custom UNET++ model is built for the segmentation task by tuning the hyper-parameters in addition to a custom loss function. The developed loss function helps tackle the class imbalance occurring due to small size of the optic nerve head. The proposed approach achieves 96% accuracy in classifying glaucomatous and non-glaucomatous images based on clinical feature identification. The improvised model is able to achieve state-of-art results for Intersection over Union (IOU) scores, 0.9477 for optic disc and 0.9321 for optic cup, along with providing an enhancement in reducing the training time. The model was tested on publicly available datasets RIM-ONE, DRIONS-DB and ORIGA and is able to achieve an accuracy of 91%, 92% and 90% respectively. The developed approach is validated by training it over RIM-ONE dataset independently, without changing any model parameters. The model provides significant improvement in segmentation of the optic disc and optic cup along with improvement in training time. [ABSTRACT FROM AUTHOR]

Segmentation of lesions from fundus images is an essential prerequisite for accurate severity assessment of diabetic retinopathy. Due to variation in morphologies, number and size of lesions, the manual grading process becomes extremely challenging and time-consuming. This necessitates the need of an automatic segmentation system that can precisely define the region of interest boundaries and assist ophthalmologists in speedy diagnosis along with diabetic retinopathy severity grading. The paper presents a modified U-Net architecture based on residual network and employs periodic shuffling with sub-pixel convolution initialized to convolution nearest neighbour resize. The proposed architecture has been trained and validated for microaneurysm and hard exudate segmentation on two publicly available datasets namely IDRiD and e-ophtha. For IDRiD dataset, the network obtains 99.88% accuracy, 99.85% sensitivity, 99.95% specificity and dice score of 0.9998 for both microaneurysm and exudate segmentation. Further, when trained on e-ophtha and validated on IDRiD dataset, the network shows 99.98% accuracy, 99.88% sensitivity, 99.89% specificity and dice score of 0.9998 for microaneurysm segmentation. For exudates segmentation, the model obtained 99.98% accuracy, 99.88% sensitivity, 99.89% specificity and dice score of 0.9999, when trained on e-ophtha and validated on IDRiD dataset. In comparison to existing literature, the proposed model provides state-of-the-art results for retinal lesion segmentation. [ABSTRACT FROM AUTHOR]

Diabetic retinopathy is one of the leading reasons that causes blindness. And the segmentation of hard exudates in color fundus images is crucial for early diagnosis of diabetic retinopathy, which is a difficult task due to its uncertainty in size, shape and contrast. Class-balanced cross entropy (CBCE) loss is the most popular objective function for image segmentation task to solve the class-unbalance problem. However, we show that background pixels tend to be misclassified to hard exudates in CBCE since the loss for a misclassified background pixels is much smaller than that for a misclassified hard exudate pixel, which is called loss-unbalance problem here. A top-k loss is proposed in this paper, which considers the cases of both class-unbalance and loss-unbalance by focusing more over the hard-to-classify pixels. Moreover, a fast version of the top-k loss, named bin loss, is implemented for efficiency, which reduces the time complexity from O (n log n) of top-k loss to O (n), where n is the number of background pixels. We evaluated the proposed bin loss over two public datasets for hard exudates segmentation task, including e-ophtha EX and IDRiD. Furthermore, three popular models for image segmentation, HED, DeepLab v2, and FCRN, were used to evaluate the versatility of bin loss. Extensive experiments show that each model with the proposed bin loss performs better than that with CBCE loss, which demonstrates bin loss is versatile so that it can be applied to different models for performance improvement. Specially, for DeepLab over e-ophtha EX, the F-score increases 5.2 percentage points, and the area under the SE-PPV curve (AUC) increases 10.6 percentage points. Moreover, the AUC increases more than 4 percentage points over IDRiD dataset for both DeepLab and FCRN. The source code of bin loss is available at: https://github.com/guomugong/bin_loss. [ABSTRACT FROM AUTHOR]

Color fundus image analysis for detecting the retinal abnormalities requires an improved visualization of image attributes with sufficient luminosity, contrast and accurate edge details. A hybrid technique based on singular value equalization using shearlet transform and adaptive gamma correction, followed by contrast limited adaptive histogram equalization (CLAHE) is proposed for the enhancement of luminosity and contrast in color fundus images. The low frequency components of the original and adaptive gamma transformed value channel in HSV color space obtained by applying shearlet transform are considered for singular value equalization. The high frequency components of the unchanged value channel, denoised using soft thresholding are applied while performing inverse shearlet transform. Luminosity component in L * a * b * colorspace is considered for performing CLAHE on the singular value equalized image. Subjective analysis is done based on visualization of the image attributes and the objective analysis is carried out based on the parameters such as Peak signal to noise ratio, entropy, feature similarity index, edge-based contrast measure, quality index and noise suppression measure. The simulation results evince superior noise performance, sufficient luminosity adjustment and improved contrast along with excellent edge detail preservation when compared with the existing state-of-the-art techniques. [ABSTRACT FROM AUTHOR]

• Utilized D2_UNet with dense blocks for enhanced vessel segmentation results. • SE-ResCA-GTNet for retinopathy classification, featuring a transformer model and GO algorithm for enhanced performance. • Proposed model excels in segmentation & classification, enabling precise DR detection and diagnosis. An essential complication of the diabetes disease is termed as Diabetic Retinopathy (DR), and it leads visual impairment in long-term vision. In general, DR is a kind of eye disease occurred by diabetes mellitus and also it is a general cause of blindness around the world. In this scenario, early detection of DR is more necessary to prevent people from being afflicted by blindness. Classification is difficult since there are so many different fundus imaging options, especially in Proliferative DR, which includes bleeding and the development of new arteries in the retina. The publicly accessible fundus image datasets from Kaggle (EyePACS) are used in this study together with real-time datasets for training and testing a transformer network on the datasets. The proposed model is used to develop an automated transformer network mechanism for classifying DR disease in earlier stages from the given inputs. The most crucial areas, including blood vessels, hemorrhage, soft exudates, and hard exudates from the pre-processed images are segmented using D2_UNet approach to make accurate classification. Finally, a new SE-ResCA-GTNet model is proposed to classify the varied severity of retinopathy disease from the given input images. Whereas, the Gazelle Optimization (GO) algorithm is used to fine-tune the proposed classifier to attain improved accuracy results. Using the fundus image dataset, the proposed SE-ResCA-GTNet classification model obtains accuracy, recall, and f1-measures of 99.8 %, 99.4 %, and 99.3 %. When compared to state-of-the-art methods, the experimental results demonstrate that the proposed method outperforms them. [ABSTRACT FROM AUTHOR]

The structure of the retinal vascular tree can reflect many indicators of ophthalmic health status. Retinal vessel segmentation, an important basis for the quantitative analysis of vascular structure, has been developed over decades. However, in the areas of imaging quality and contrast, there are still vascular ruptures and missed detection by existing methods. Moreover, the problem of false positives caused by existing lesions has not been properly addressed. Inspired by the primary visual cortex of the biological visual system, this paper proposes Anisotropic Perceptive Convolution (APC) and Anisotropic Enhancement Module (AEM) to model visual cortex cells and their orientation selection mechanism, respectively, as well as a novel network named W-shaped Deep Matched Filtering (WS-DMF) having a W-shaped overall framework. In the feature extraction backbone, a DMF based on a multilayer aggregation of APC is designed to enhance vascular features and inhibit the expression of pathological features. In addition, AEMs are embedded into DMF to enhance the orientation and position information of high-dimensional features. Furthermore, in order to enhance the ability of APC for perceiving linear textures (such as blood vessels), an Orientation Anisotropic Loss (OAL) is introduced. Based on experiments on several widely used datasets and several professional vessel segmentation metrics, the proposed method exhibits strong abilities of vessel extraction and lesion inhibition, surpassing many state-of-the-art algorithms. The results indicate the potential value of WS-DMF in clinical practice. The source code is available at https://www.github.com/tyb311/WS-DMF. [Display omitted] • In order to reduce the missed detection of vessels under poor imaging quality, we propose an anisotropic perceptual convolution (APC) layer to highlight the vascular structure by extracting locally linear textures and suppressing isotropic regions. • To reduce the phenomenon of vessel rupture, we propose a deep matched filtering (DMF) backbone based on the aggregation of a multilayer orientation perception convolution, enabling the model to deduce and connect the broken vessels. • Moreover, in order to enhance the anisotropic features (i.e., vessels), we design the Anisotropic Enhancement Modules (AEM) to emphasize the orientation and position of vessels in high-dimensional space. • Qualitative and quantitative experiments on several widely used datasets show that the proposed method has excellent performance. [ABSTRACT FROM AUTHOR]

Diabetic retinopathy and diabetic macular edema are the two leading causes for blindness in working-age people, and the quantitative and qualitative diagnosis of these two diseases usually depends on the presence and areas of lesions in fundus images. The main related lesions include soft exudates, hard exudates, microaneurysms, and haemorrhages. However, segmentation of these four kinds of lesions is difficult due to their uncertainty in size, contrast, and high interclass similarity. Therefore, we aim to design a multi-lesion segmentation model. We have designed the first small object segmentation network (L-Seg) that can segment the four kinds of lesions simultaneously. Taking into account that small lesion regions could not response at high level of network, we propose a multi-scale feature fusion method to handle this problem. In addition, when considering the cases of both class-imbalance and loss-imbalance problems, we propose a multi-channel bin loss. We have evaluated L-Seg on three fundus datasets including two publicly available datasets - IDRiD and e-ophtha and one private dataset - DDR. Extensive experiments have demonstrated that L-Seg achieves better performance in small lesion segmentation than other deep learning models and traditional methods. Specially, the mAUC score of L-Seg is over 16.8%, 1.51% and 3.11% higher than that of DeepLab v3+ on IDRiD, e-ophtha and DDR datasets, respectively. Moreover, our framework shows competitive performance compared with top-3 teams in IDRiD challenge. The source code of L-Seg is available at: https://github.com/guomugong/L-Seg. [ABSTRACT FROM AUTHOR]

• For the first time, the visibility graph is applied in the image processing field. • The Radon transform applied globally to fundus image. • Graph topology analysis measures are used for grading of diabetic retinopathy. • High accuracy, sensitivity and specificity are obtained using proposed method. Diabetic retinopathy (DR) is one of the problems caused due to the diabetes disease in which the retina is damaged because fluid leaks into the retina from the blood vessels. In extreme cases, the patient may loss vision. Therefore, determination of DR grades has an important role in the treatment process of the disease and preventing vision loss. Different image processing based methods have been proposed to detect the different stages of DR automatically. In this paper, a method based on Radon transform (RT) and visibility graph (VG) was proposed to automatically discriminate grades 0 (normal), 1, 2 and 3 of the DR from fundus images. The proposed method is summarized in two stages: feature extraction and classification. In this study, for the first time, the VG method was employed in the image processing field for feature extraction. Then, these features were given to error-correcting output codes (ECOC) method for classification purposes. The proposed method was easy enjoying an accuracy of 97.92%, a sensitivity of 95.83% and a specificity of 98.61%. The VG based method can be a very easy, cheap, and effective test for the automatic grading of DR stages and it can apply in other image processing application. [ABSTRACT FROM AUTHOR]

Background and Objective: The condition of vessel of the human eye is an important factor for the diagnosis of ophthalmological diseases. Vessel segmentation in fundus images is a challenging task due to complex vessel structure, the presence of similar structures such as microaneurysms and hemorrhages, micro-vessel with only one to several pixels wide, and requirements for finer results.Methods: In this paper, we present a multi-scale deeply supervised network with short connections (BTS-DSN) for vessel segmentation. We used short connections to transfer semantic information between side-output layers. Bottom-top short connections pass low level semantic information to high level for refining results in high-level side-outputs, and top-bottom short connection passes much structural information to low level for reducing noises in low-level side-outputs. In addition, we employ cross-training to show that our model is suitable for real world fundus images.Results: The proposed BTS-DSN has been verified on DRIVE, STARE and CHASE_DB1 datasets, and showed competitive performance over other state-of-the-art methods. Specially, with patch level input, the network achieved 0.7891/0.8212 sensitivity, 0.9804/0.9843 specificity, 0.9806/0.9859 AUC, and 0.8249/0.8421 F1-score on DRIVE and STARE, respectively. Moreover, our model behaves better than other methods in cross-training experiments.Conclusions: BTS-DSN achieves competitive performance in vessel segmentation task on three public datasets. It is suitable for vessel segmentation. The source code of our method is available at: https://github.com/guomugong/BTS-DSN. [ABSTRACT FROM AUTHOR]

Highlights • We propose a novel and robust cascade classification framework for retinal vessel segmentation. This classification model is trained by a one-pass feed forward process. Thus, the degree of nonlinearity of the proposed classifier is not predefined, but determined by the complexity of the data structure. • The proposed cascade classification framework produces a definite training and testing result that can be exactly reproduced. This greatly reduces the uncertainty of its usage in the both training and testing and hence it is user friendly. • We show that the proposed cascade classification framework can handle typically challenging retinal vessel structures that are difficult segmentation issues for some state-of-the art methods. • As an adaptive and effective solution to the difficult classification problems, the proposed technique can be flexibly extended to other image recognition tasks. Abstract Accurate segmentation of retinal vessel from fundus image is a prerequisite for the computer-aided diagnosis of ophthalmology diseases. In this paper, we propose a novel and robust cascade classification framework for retinal vessel segmentation. Our classification model envelops a set of computationally efficient Mahalanobis distance classifiers to form a highly nonlinear decision. Different from other nonlinear classifiers that need a predefined nonlinear kernel or need iterative training, the proposed cascade classification framework is trained by a one-pass feed forward process. Thus, the degree of nonlinearity of the proposed classifier is not predefined, but determined by the complexity of the data structure. Experimental evaluations on three diverse publicly available databases show that the proposed cascade classification framework achieves 95.41–96.40% vessel segmentation accuracy, and outperforms the state-of-the-art methods in terms of F1-score and Matthew correlation coefficient consistently on all three diverse databases. A qualitative comparison between different segmentation approaches demonstrates the superiority of the proposed method in dealing with typically challenging retinal structures. The proposed cascade classification framework consistently yields a high performance for retinal vessel segmentation, and delineates a more complete and accurate vessel tree. As an adaptive and effective solution to the difficult classification problems, the proposed technique can be flexibly extended to other image recognition tasks. [ABSTRACT FROM AUTHOR]

Highlights • To the best of our knowledge, the problem of resolution enhancement of the retinal images has not been addressed. • The proposed method takes into consideration the diagnostic information in the retinal fundus image during super-resolution (SR). • The method performs SR only on the zone of interest rather than the entire image. This leads to computational time efficiency. • A set of novel fundus image specific features are extracted to classify the diagnostically significant and non-significant zones. • A trade-off between the learning based method and bicubic interpolation is performed during SR which leads to achievement in computational time without loss in reconstruction accuracy. Abstract Advancements in tele-medicine have led to the development of portable and cheap hand-held retinal imaging devices. However, the images obtained from these devices have low resolution (LR) and poor quality that may not be suitable for retinal disease diagnosis. Therefore, this paper proposes a novel framework for the super-resolution (SR) of the LR fundus images. The method takes into consideration the diagnostic information in the fundus images during the SR process. In this work, SR is performed on the zone of interest of the fundus images. Clinical information of the selected zone is captured using the Shannon entropy, the contrast sensitivity index (CSI), the multi-resolution (MR) intra-band energy and the MR inter-band eigen features. The support vector machine (SVM) classifier is used to decide the clinical significance of the zone. Highly accurate learning based SR method or the bicubic interpolation is applied to the selected zone based on the classification output. The method is tested on the Standard Diabetic Retinopathy Database Calibration level 1 (DIARETDB1) and the Digital Retinal Images for Vessel Extraction (DRIVE) databases. Classification accuracy of 85.22% and 85.77% is achieved for the DIARETDB1 and DRIVE databases respectively. The SR performance of the algorithm is quantified in terms of peak signal to noise ratio (PSNR), structural similarity index measure (SSIM) and computational time. The proposed diagnostic information based SR achieves computational time efficiency without compromising with the high resolution (HR) reconstruction accuracy of the fundus image zones. [ABSTRACT FROM AUTHOR]

Highlights • A Deep Belief Net (DBN) is trained to detect vessel interior, centreline and edges. • Particle filtering is used to quantify vasculature by using the output of the DBN. • A vessel profile is represented with probability profiles of centreline and edges. • The appearance of vessels in fundus images is considered in a vessel geometry model. • The lack of labelled data for segmentation was tackled using a probabilistic approach. Graphical abstract Abstract Deep networks have recently seen significant application to the analysis of medical image data, particularly for segmentation and disease classification. However, there are many situations in which the purpose of analysing a medical image is to perform parameter estimation, assess connectivity or determine geometric relationships. Some of these tasks are well served by probabilistic trackers, including Kalman and particle filters. In this work, we explore how the probabilistic outputs of a single-architecture deep network may be coupled to a probabilistic tracker, taking the form of a particle filter. The tracker provides information not easily available with current deep networks, such as a unique ordering of points along vessel centrelines and edges, whilst the construction of observation models for the tracker is simplified by the use of a deep network. We use the analysis of retinal images in several datasets as the problem domain, and compare estimates of vessel width in a standard dataset (REVIEW) with manually determined measurements. [ABSTRACT FROM AUTHOR]

Highlights • Hemorrhage is one of the early symptoms of diabetic retinopathy. • Accurate detection of hemorrhage has important significance in building automatic screening system of diabetic retinopathy. • A method based on two dimensional (2D) Gaussian fitting and human visual characteristics is proposed. • This method can realize automatic detection of hemorrhages in fundus image with high accuracy. Abstract Hemorrhage is one of the early symptoms of diabetic retinopathy, so accurate detection of hemorrhage has important significance in building automatic screening system of diabetic retinopathy. In this paper, a method based on two dimensional (2D) Gaussian fitting and human visual characteristics is proposed. Firstly, brightness correction and contrast limited adaptive histogram equalization are used to preprocess original color fundus image. Secondly, candidate hemorrhages are extracted based on background estimation and watershed segmentation. Thirdly, 2D Gaussian fitting and human visual characteristics are used to extract visual features of candidate hemorrhages. Finally, hemorrhages are obtained from candidate hemorrhages based on visual features. The proposed method is evaluated on 219 fundus images of DIARETDB database. Experimental results show that overall average sensitivity, specificity and accuracy for hemorrhage in image level are 100%, 82% and 95.42% respectively, and overall average sensitivity and positive predictive value for hemorrhage in lesion level are 94.01% and 90.30% respectively. The results show that this method can realize automatic detection of hemorrhages in fundus image with high accuracy. [ABSTRACT FROM AUTHOR]

Highlights • Proposed a noise removal and contrast enhancement algorithm for color fundus image. • Our method uses integration of filters and contrast limited adaptive histogram equalization (CLAHE) technique. • Efficacy of the method is evaluated through different performance parameters. • The performance of our method is better than other state-of-the-art technique. Abstract Now-a-days medical fundus images are widely used in clinical diagnosis for the detection of retinal disorders. Fundus images are generally degraded by noise and suffer from low contrast issues. These issues make it difficult for ophthalmologist to detect and interpret diseases in fundus images. This paper presents a noise removal and contrast enhancement algorithm for fundus image. Integration of filters and contrast limited adaptive histogram equalization (CLAHE) technique is applied for solving the issues of de-noising and enhancement of color fundus image. The efficacy of the proposed method is evaluated through different performance parameters like Peak Signal to Noise Ratio (PSNR), Structural Similarity Index (SSIM), Correlation coefficient (CoC) and Edge preservation index (EPI). The proposed method achieved 7.85% improvement in PSNR, 1.19% improvement in SSIM, 0.12% improvement in CoC and 1.28% improvement in EPI when compared to the state of the art method. [ABSTRACT FROM AUTHOR]

Highlights • Contrast Limited Adaptive Histogram Equalization (CLAHE) eradicates false photographic artifacts and illuminations in fundus images. • Tandem Pulse Coupled Neural Network (TPCNN) model operates on inter and intra channel linking of the input neurons and generates the automatic feature vectors. • The cross channel linking enables the visibility of vessels at the cross over points. • Classification and extraction of retinal blood vessels via Deep Learning Based Support Vector Machine (DLBSVM) helps the network to attain better data expression to improve the final classification result. Abstract The most prominent ophthalmic cause of blindness is Diabetic Retinopathy (DR). This retinal disease is characterized by variation in diameter of the retinal blood vessel and the new blood vessel growth inside the retina. A system to enhance the quality of the segmentation result over the pathological retinal images has been proposed. The proposed method uses Contrast Limited Adaptive Histogram Equalization (CLAHE) for preprocessing and Tandem Pulse Coupled Neural Network (TPCNN) model for automatic feature vectors generation then classification and extraction of the retinal blood vessels via Deep Learning Based Support Vector Machine (DLBSVM). The proposed approach is assessed over the standard public fundus image databases to evaluate the performance. The results render that these techniques improve the segmentation results with an average value of 74.45% sensitivity, 99.40% specificity, and 99.16% accuracy. The results evoke that the proposed method is a suitable alternative for supervised techniques. [ABSTRACT FROM AUTHOR]

Unsupervised anomaly detection (UAD) methods are trained with normal (or healthy) images only, but during testing, they are able to classify normal and abnormal (or disease) images. UAD is an important medical image analysis (MIA) method to be applied in disease screening problems because the training sets available for those problems usually contain only normal images. However, the exclusive reliance on normal images may result in the learning of ineffective low-dimensional image representations that are not sensitive enough to detect and segment unseen abnormal lesions of varying size, appearance, and shape. Pre-training UAD methods with self-supervised learning, based on computer vision techniques, can mitigate this challenge, but they are sub-optimal because they do not explore domain knowledge for designing the pretext tasks, and their contrastive learning losses do not try to cluster the normal training images, which may result in a sparse distribution of normal images that is ineffective for anomaly detection. In this paper, we propose a new self-supervised pre-training method for MIA UAD applications, named P seudo M ulti-class S trong A ugmentation via C ontrastive L earning (PMSACL). PMSACL consists of a novel optimisation method that contrasts a normal image class from multiple pseudo classes of synthesised abnormal images, with each class enforced to form a dense cluster in the feature space. In the experiments, we show that our PMSACL pre-training improves the accuracy of SOTA UAD methods on many MIA benchmarks using colonoscopy, fundus screening and Covid-19 Chest X-ray datasets. • A new self-supervised pre-training method for unsupervised anomaly detection algorithms. • PMSACL re-formulates the one-class classification into a pseudo multi-class problem. • A new contrastive optimisation to learn better fine-grained features for downstream anomaly detectors. • This method can adapt well to different types of downstream anomaly detection methods. • PMSACL can achieve state-of-the-art results on datasets from many medical domains. [ABSTRACT FROM AUTHOR]

• The review presents the comprehensive evolution of Computer Aided Diagnosis (CAD) system for automatic diagnosis of Hypertensive Retinopathy (HR). • Detailed analysis of the current research methodologies on conventional, machine learning, and deep learning techniques for HR diagnosis. • It provides the detailed summaries and analysis for each task and overall. • It discusses the pathophysiology, brief details of datasets, and, performance evaluation metrics for HR. • This study also highlights the challenges in the existing studies and proposes the future research directions. Hypertensive Retinopathy (HR) is a retinal disease caused by elevated blood pressure for a prolonged period. There are no obvious signs in the early stages of high blood pressure, but it affects various body parts over time, including the eyes. HR is a biomarker for several illnesses, including retinal diseases, atherosclerosis, strokes, kidney disease, and cardiovascular risks. Early microcirculation abnormalities in chronic diseases can be diagnosed through retinal examination prior to the onset of major clinical consequences. Computer-aided diagnosis (CAD) plays a vital role in the early identification of HR with improved diagnostic accuracy, which is time-efficient and demands fewer resources. Recently, numerous studies have been reported on the automatic identification of HR. This paper provides a comprehensive review of the automated tasks of Artery-Vein (A/V) classification, Arteriovenous ratio (AVR) computation, HR detection (Binary classification), and HR severity grading. The review is conducted using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) protocol. The paper discusses the clinical features of HR, the availability of datasets, existing methods used for A/V classification, AVR computation, HR detection, and severity grading, and performance evaluation metrics. The reviewed articles are summarized with classifiers details, adoption of different kinds of methodologies, performance comparisons, datasets details, their pros and cons, and computational platform. For each task, a summary and critical in-depth analysis are provided, as well as common research issues and challenges in the existing studies. Finally, the paper proposes future research directions to overcome challenges associated with data set availability, HR detection, and severity grading. [ABSTRACT FROM AUTHOR]

Abstract Optic disc (OD) detection is a basic procedure for the image processing algorithms which intend to diagnose and track retinal disorders. In this study, a new OD localization approach is proposed, based on color and shape properties of OD as well as the convergence point of the main vessels. This study is comprised of two successive fundamental steps. At the first step, an algorithm finding the approximate convergent point of the vessels is used in order to roughly localize OD. At the second step, three new features are suggested and a fuzzy logic controller (FLC) whose input membership functions are designed based on these features is proposed. The proposed method is applied to the DRIVE, STARE, DIARETDB0 and DIRETDB1 datasets and the obtained results validate the improvement in the performance by attaining success rate of 100%, 91,35%, 90% and 100% respectively and detecting OD centers and contours precisely in a reasonable execution time. [ABSTRACT FROM AUTHOR]

The presence of peripapillary atrophy (PPA) is associated with two kinds of diseases, namely glaucoma and myopia. PPA is one of the characteristics of these diseases that can be observed through retinal fundus images. We propose an automatic detection method of PPA in retinal fundus images using statistical features and Backpropagation Neural Network. In this research, those images are classified into two classes: no-PPA and PPA. The features are extracted from the focal areas, which capture the areas where PPA may occur in each sector. There are three features used in this method namely, standard deviation, smoothness and third moment; they are selected using gain ratio method. The performance of the proposed method achieves the accuracy of 0.95, 0.96, and 0.96 for three different datasets. These are obtained using 155 retinal fundus images, from which training and testing data of 47 images and 108 images, respectively, are randomly selected. [ABSTRACT FROM AUTHOR]

Background and objectives: The automatic classification of retinal blood vessels into artery and vein (A/V) is still a challenging task in retinal image analysis. Recent works on A/V classification mainly focus on the graph analysis of the retinal vasculature, which exploits the connectivity of vessels to improve the classification performance. While they have overlooked the importance of pixel-wise classification to the final classification results. This paper shows that a complicated feature set is efficient for vessel centerline pixels classification. Methods: We extract enormous amount of features for vessel centerline pixels, and apply a genetic-search based feature selection technique to obtain the optimal feature subset for A/V classification. Results: The proposed method achieves an accuracy of 90.2%, the sensitivity of 89.6%, the specificity of 91.3% on the INSPIRE dataset. It shows that our method, using only the information of centerline pixels, gives a comparable performance as the techniques which use complicated graph analysis. In addition, the results on the images acquired by different fundus cameras show that our framework is capable for discriminating vessels independent of the imaging device characteristics, image resolution and image quality. Conclusion: The complicated feature set is essential for A/V classification, especially on the individual vessels where graph-based methods receive limitations. And it could provide a higher entry to the graph-analysis to achieve a better A/V labeling. [ABSTRACT FROM AUTHOR]

Background and objective Macula segmentation and fovea localization is one of the primary tasks in retinal analysis as they are responsible for detailed vision. Existing approaches required segmentation of retinal structures viz. optic disc and blood vessels for this purpose. Method This work avoids knowledge of other retinal structures and attempts data mining techniques to segment macula. Unsupervised clustering algorithm is exploited for this purpose. Selection of initial cluster centres has a great impact on performance of clustering algorithms. A heuristic based clustering in which initial centres are selected based on measures defining statistical distribution of data is incorporated in the proposed methodology. The initial phase of proposed framework includes image cropping, green channel extraction, contrast enhancement and application of mathematical closing. Then, the pre-processed image is subjected to heuristic based clustering yielding a binary map. The binary image is post-processed to eliminate unwanted components. Finally, the component which possessed the minimum intensity is finalized as macula and its centre constitutes the fovea. Results The proposed approach outperforms existing works by reporting that 100%,of HRF, 100% of DRIVE, 96.92% of DIARETDB0, 97.75% of DIARETDB1, 98.81% of HEI-MED, 90% of STARE and 99.33% of MESSIDOR images satisfy the 1R criterion, a standard adopted for evaluating performance of macula and fovea identification. Conclusion The proposed system thus helps the ophthalmologists in identifying the macula thereby facilitating to identify if any abnormality is present within the macula region. [ABSTRACT FROM AUTHOR]

Retinal nerve fiber layer defect (RNFLD) provides an early objective evidence of structural changes in glaucoma. RNFLD detection is currently carried out using imaging modalities like OCT and GDx which are expensive for routine practice. In this regard, we propose a novel automatic method for RNFLD detection and angular width quantification using cost effective redfree fundus images to be practically useful for computer-assisted glaucoma risk assessment. After blood vessel inpainting and CLAHE based contrast enhancement, the initial boundary pixels are identified by local minima analysis of the 1-D intensity profiles on concentric circles. The true boundary pixels are classified using random forest trained by newly proposed cumulative zero count local binary pattern ( CZC-LBP ) and directional differential energy ( DDE ) along with Shannon, Tsallis entropy and intensity features. Finally, the RNFLD angular width is obtained by random sample consensus (RANSAC) line fitting on the detected set of boundary pixels. The proposed method is found to achieve high RNFLD detection performance on a newly created dataset with sensitivity (SN) of 0.7821 at 0.2727 false positives per image (FPI) and the area under curve (AUC) value is obtained as 0.8733. [ABSTRACT FROM AUTHOR]

Glaucoma is a neuro-degenerative disorder of the eye and it leads to permanent blindness when untreated or detected in the later stage. The main cause of glaucoma is the damage of the optic nerve, which occurs due to the increase of eye pressure. Hence the early detection of this disease is critical in time and which can help to prevent further vision loss. The assessment of optic nerve head using fundus images is more beneficial than the raised intra ocular pressure assessment in population-based glaucoma screening. This work proposed a novel method for glaucoma identification based on time-invariant feature cup to disk ratio and anisotropic dual-tree complex wavelet transform features. Optic disk segmentation is done by using Fuzzy C-Means clustering method and Otsu's thresholding is used for optic cup segmentation. The results show the proposed method achieved an accuracy rate of 97.67% with 98% sensitivity using a multilayer perceptron model that is considered as clinically significant when compared to the existing works. [ABSTRACT FROM AUTHOR]

• There is a lack of computational tools for quality estimation of fundus images. • We introduce a novel fundus image quality scale (range 1–10). • We develop a new deep learning model, FundusQ-Net, that can estimate fundus image quality relative to this new scale. • FundusQ-Net is trained on 89,947 images and demonstrates high precision and to generalize to an external dataset. Objective: Ophthalmological pathologies such as glaucoma, diabetic retinopathy and age-related macular degeneration are major causes of blindness and vision impairment. There is a need for novel decision support tools that can simplify and speed up the diagnosis of these pathologies. A key step in this process is to automatically estimate the quality of the fundus images to make sure these are interpretable by a human operator or a machine learning model. We present a novel fundus image quality scale and deep learning (DL) model that can estimate fundus image quality relative to this new scale. Methods: A total of 1245 images were graded for quality by two ophthalmologists within the range 1–10, with a resolution of 0.5. A DL regression model was trained for fundus image quality assessment. The architecture used was Inception-V3. The model was developed using a total of 89,947 images from 6 databases, of which 1245 were labeled by the specialists and the remaining 88,702 images were used for pre-training and semi-supervised learning. The final DL model was evaluated on an internal test set (n = 209) as well as an external test set (n = 194). Results: The final DL model, denoted FundusQ-Net, achieved a mean absolute error of 0.61 (0.54–0.68) on the internal test set. When evaluated as a binary classification model on the public DRIMDB database as an external test set the model obtained an accuracy of 99%. Significance: the proposed algorithm provides a new robust tool for automated quality grading of fundus images. [ABSTRACT FROM AUTHOR]

• A multi-preprocessing fusion method is used to extract features. • There is no need to remove blood vessels or other structures. • The statistical features can describe the cross-section well. • The extracted features can achieve a better classification result. Diabetic retinopathy (DR) is the main cause of new cases of blindness, and retinal microaneurysm (MA) is one of the early clinical manifestations of DR. However, it is difficult to directly observe MA with naked eyes because of their small size and poor contrast with the surrounding background. This paper presents a new method for automatic detection of MAs from fundus images. It mainly includes four steps: image preprocessing, candidate region extraction, feature extraction and classification. At the stage of feature extraction, different from traditional methods, we propose a feature extraction method based on multi-preprocessing fusion. In this paper, features are extracted from the original green channel image and the original gray image respectively, and then different processing methods are used to extract the same features from multiple pre-processed images. Finally, the extracted features are classified by multi-layer perceptron (MLP) classifier to determine whether the candidate object is microaneurysm (MA). The method is evaluated on e-ophtha-MA database and achieved high sensitivity. Using the FROC (free-response ROC) indicator, the detection achieves the F-score, accuracy, sensitivity, specificity and AUC (area under the curve) of 0.507, 0.940, 0.870, 0.938 and 0.9821, respectively, compared to the most advanced methods available. The proposed method of multi-preprocessing and fusion feature extraction is beneficial to classification and improves the detection performance. [ABSTRACT FROM AUTHOR]

• A double imaging system was designed to capture cattle eye surface and fundus separately. • Chromatic features of fundus images are more reflective of retina information than eye surface images. • The new system outperformed previous system in classification of Vitamin A deficiency. Blood vitamin A levels during fattening have a crucial influence on beef marbling. As an alternative of conventional blood assays, eye imaging method has been raised as a potential non-invasive way to monitor vitamin A levels in cattle. However, current image capture systems merely acquire cattle eye surface images, which cannot adequately reflect the chromatic information of eye retina that indicates vitamin A level changes. Based on previous research, we propose a novel, automatic double imaging system that combined two illuminating system as well as the installation of polarising filters to eliminate eye surface reflection and consequently, obtain clear fundus images. This greatly facilitated subsequent image processing and feature extraction for vitamin A prediction. The proposed imaging system demonstrated a stronger correlation between vitamin A levels and chromatic features in fundus images than previous systems. An improved classification (accuracy 82%) of vitamin A deficiency level than previous study (56.6%) strongly supports this system's advantage. Being non-invasive cost-effective and rapid, this double imaging system is expected to play an important role in precision livestock farming as well as clinical diagnosis in veterinarian. [ABSTRACT FROM AUTHOR]

Background and Objectives: The authentication and integrity verification of medical images is a critical and growing issue for patients in e-health services. Accurate identification of medical images and patient verification is an essential requirement to prevent error in medical diagnosis. The proposed work presents an imperceptible watermarking system to address the security issue of medical fundus images for tele-ophthalmology applications and computer aided automated diagnosis of retinal diseases.Methods: In the proposed work, patient identity is embedded in fundus image in singular value decomposition domain with adaptive quantization parameter to maintain perceptual transparency for variety of fundus images like healthy fundus or disease affected image. In the proposed method insertion of watermark in fundus image does not affect the automatic image processing diagnosis of retinal objects & pathologies which ensure uncompromised computer-based diagnosis associated with fundus image. Patient ID is correctly recovered from watermarked fundus image for integrity verification of fundus image at the diagnosis centre.Results: The proposed watermarking system is tested in a comprehensive database of fundus images and results are convincing.Experimental: results indicate that proposed watermarking method is imperceptible and it does not affect computer vision based automated diagnosis of retinal diseases.Conclusions: Correct recovery of patient ID from watermarked fundus image makes the proposed watermarking system applicable for authentication of fundus images for computer aided diagnosis and Tele-ophthalmology applications. [ABSTRACT FROM AUTHOR]

• The use of artificial intelligence in diabetic retinopathy has become a popular research focus in the past decade. • Conducted a bibliometric approach to identify and analyse the academic literature on artificial intelligence in diabetic retinopathy. • This study summarizes the recent advances in artificial intelligence technology on diabetic retinopathy research and sheds light on the emerging trends, sources, and leading institutions. The use of artificial intelligence in diabetic retinopathy has become a popular research focus in the past decade. However, no scientometric report has provided a systematic overview of this scientific area. We utilized a bibliometric approach to identify and analyse the academic literature on artificial intelligence in diabetic retinopathy and explore emerging research trends, key authors, co-authorship networks, institutions, countries, and journals. We further captured the diabetic retinopathy conditions and technology commonly used within this area. Web of Science was used to collect relevant articles on artificial intelligence use in diabetic retinopathy published between January 1, 2012, and December 31, 2022. All the retrieved titles were screened for eligibility, with one criterion that they must be in English. All the bibliographic information was extracted and used to perform a descriptive analysis. Bibliometrix (R tool) and VOSviewer (Leiden University) were used to construct and visualize the annual numbers of publications, journals, authors, countries, institutions, collaboration networks, keywords, and references. In total, 931 articles that met the criteria were collected. The number of annual publications showed an increasing trend over the last ten years. Investigative Ophthalmology & Visual Science (58/931), IEEE Access (54/931), and Computers in Biology and Medicine (23/931) were the most journals with most publications. China (211/931), India (143/931, USA (133/931), and South Korea (44/931) were the most productive countries of origin. The National University of Singapore (40/931), Singapore Eye Research Institute (35/931), and Johns Hopkins University (34/931) were the most productive institutions. Ting D. (34/931), Wong T. (28/931), and Tan G. (17/931) were the most productive researchers. This study summarizes the recent advances in artificial intelligence technology on diabetic retinopathy research and sheds light on the emerging trends, sources, leading institutions, and hot topics through bibliometric analysis and network visualization. Although this field has already shown great potential in health care, our findings will provide valuable clues relevant to future research directions and clinical practice. [ABSTRACT FROM AUTHOR]

Recently, academic and clinical communities have paid increasing attention to design computer-aided diagnosis methods for automatic and accurate diabetic retinopathy (DR) grading from fundus images. However, most existing methods either possess limited capability in extracting lesion-aware information or require manual lesion annotations, resulting in ordinary grading performance or additional undesirable labor. To address these issues, this paper proposes an end-to-end Attention-Driven Cascaded Network (ADCNet) for DR grading. Specifically, we first propose a hybrid attention module at the shallow layer by incorporating a multi-branch spatial attention and a loss-based attention to extract rich lesion-aware information without any manual lesion annotations. Then, we orderly cascade the lesion-aware information from shallow to high layers through an attention-driven aggregation strategy to obtain and integrate plentiful DR-related features. Finally, the grading score is generated by fusing DR-related features of all layers. Experimental results on two publicly available datasets demonstrate that the proposed ADCNet is competent for accurate DR grading, and outperforms the state-of-the-art methods on seven widely used evaluation criteria. • A DR grading network is proposed by training the model without lesion annotations. • A module is proposed to provide initial guidance for lesion-aware feature extraction. • A module is proposed to promote transmission and interaction of lesion-aware information. • The proposed network achieves better performance over competing methods. [ABSTRACT FROM AUTHOR]

The Optic disc (OD) nerve head region in general and OD center coordinates in particular form basis for study and analysis of various eye pathologies. The shape, contour and size of OD is vital in classification and grading of retinal diseases like glaucoma. There is a need to develop fast and efficient algorithms for large scale retinal disease screening. With this in mind, this paper present a novel framework for fast and fully automatic detection of OD and its accurate segmentation in digital fundus images. The methodology involves optic disc center localization followed by removal of vascular structure by accurate inpainting of blood vessels in the optic disc region. An adaptive threshold based Region Growing technique is then employed for reliable segmentation of fundus images. The proposed technique achieved significant results when tested on standard test databases like MESSIDOR and DRIVE with average overlapping ratio of 89% and 87%, respectively. Validation experiments were done on a labeled dataset containing healthy and pathological images obtained from a local eye hospital achieving an appreciable 91% average OD segmentation accuracy. [ABSTRACT FROM AUTHOR]

Glaucoma is a disease of the retina which is one of the most common causes of permanent blindness worldwide. This paper presents an automatic image processing based method for glaucoma diagnosis from the digital fundus image. In this paper wavelet feature extraction has been followed by optimized genetic feature selection combined with several learning algorithms and various parameter settings. Unlike the existing research works where the features are considered from the complete fundus or a sub image of the fundus, this work is based on feature extraction from the segmented and blood vessel removed optic disc to improve the accuracy of identification. The experimental results presented in this paper indicate that the wavelet features of the segmented optic disc image are clinically more significant in comparison to features of the whole or sub fundus image in the detection of glaucoma from fundus image. Accuracy of glaucoma identification achieved in this work is 94.7% and a comparison with existing methods of glaucoma detection from fundus image indicates that the proposed approach has improved accuracy of classification. [ABSTRACT FROM AUTHOR]

Glaucoma is an eye disease that results in irreversible loss of vision. The manual examination of optic disk (OD) is a standard procedure used for detecting glaucoma. This paper presents a glaucoma expert system based on the segmentations of OD and optic cup attained from color fundus images. A novel implicit region based active contour model is proposed for OD segmentation which incorporates the image information at the point of interest from multiple image channels to have robustness against the variations found in and around the OD region. A novel optic cup segmentation method is also proposed based on the structural and gray level properties of cup. Based on the precise information about the contours of OD and cup different parameters are calculated for glaucoma assessment. The proposed system is evaluated on 59 retinal images comprising 17 normal and 42 glaucomatous images against the groundtruths given by an experienced ophthalmologist. The proposed OD segmentation method achieved an average F -score of 0.975, average boundary distance of 10.112 pixel and average correlation coefficient of 0.916. The cup segmentation method attained an average F -score of 0.89, average boundary distance of 18.927 pixel and average correlation coefficient of 0.835. The mean error and standard deviation of the error σ for all the parameters are much smaller in glaucomatous images compared to normal images. This indicates high sensitivity of the proposed method in glaucoma assessment. [ABSTRACT FROM AUTHOR]