Search

Ophthalmology extensively relies on the application of Artificial Intelligence and Deep Learning techniques to address various challenges and issues. Macula is the central part of the retina responsible for sharp, detailed vision. Macular edema is a condition that affects the macula. It occurs when fluid accumulates in the macula, leading to swelling and distortion of the central vision system. Macular Edema can be detected using computer vision and Artificial Intelligence techniques. Large datasets having high-resolution images are required for building robust Deep Learning medical image classification models. Access to large, annotated, and readily available datasets is often limited, as the preparation of such dataset is a costly and tedious job. Leveraging a mixture of publicly accessible datasets for retinal fundus images is one strategy that can be employed. However, this solution is accompanied by the challenges pertaining to the quality, reliability, and lack of uniformity across the datasets. The complex architectures of Deep Learning models are not well suited for medical image classification using smaller datasets. This is because the model over fits during training. This work proposes a robust ResNet-50 classifier model that is trained on synthetic retinal fundus images generated by the Progressive Growing of Generative Adversarial Network. The model classifies retinal fundus images based on the severity grade of macular edema. The ResNet-50 model trained on the original IDRiD dataset achieved an accuracy of 85% while it is improved to 90.82% using Progressive Growing of Generative Adversarial Network based approach. [ABSTRACT FROM AUTHOR]

Ocular biometrics and retina are essential in creating vision. Damage to this area can cause irreversible visual field loss. Accurate early diagnosis is important to prevent these structures' abnormalities. Deep Learning (DL) technology has been widely used to improve diagnosis quality lately. This technology has the potential to be used in the diagnosis of eye disorders. This systematic review was written to highlight the DL implementation of ocular biometric abnormalities and retinopathy diagnosis. The articles were selected and assessed by Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) protocols. Reported works have been analyzing DL-based diagnosis of glaucoma, diabetic macular edema, corneal disease, and retinopathy. These studies obtained significant results due to a high accuracy approaching human-level performance. Several convolutional models, which are either built from scratch or using pre-trained networks, were used in these studies. The DL diagnosis model achieved an average accuracy, specificity, and sensitivity of over 90%. DL has the potential to help medical workers determine ocular biometric and retinal image parameters–that happened to detect various eye diseases. [ABSTRACT FROM AUTHOR]

The applications of Deep Learning (DL) are huge, but particularly, healthcare produces a lot of data that can be handled by DL models easily. Diabetic patients often suffer from eye diseases occurring from their high sugar levels. These diseases, in milder cases, lead to poor vision, and in critical cases, there is a permanent loss of eyesight. Therefore, the detection and treatment of these eye disorders is necessary to prevent complete sight loss. Therefore, efficient automatic detection systems for these diseases are essential since manual detection is not only tedious but requires a high level of expertise. The paper presents a review of the current systems for the detection and classification of Diabetic Eye Diseases (DEDs), namely, Diabetic Macular Edema (DME), Diabetic Retinopathy (DR), Glaucoma, and Cataracts. It also draws a comparison of the various techniques (DL techniques in particular) used by the researchers' community in order to detect these DEDs. The DL techniques are utilized due to the fact that these are capable of handling huge datasets occurring from the fundus images of the eye, and in addition, the accuracy score of these techniques is good, although there is much scope for improvement. The DL algorithms can be fine-tuned by augmenting the datasets and using ensemble models to improve accuracy. [ABSTRACT FROM AUTHOR]

In optometry, fundus imaging is used to detect medical disorders like heart problems, and other chronic disorders like high blood pressure, etc. Manual evaluation of these fundus images requires practical experience and pragmatic knowledge. As a result, the computerized ocular model can help in the prediction of diabetes, cornea capillaryregion detection, peripheral vascular narrowing, classification of early macular edema, Cardiovascular illnesses, and valve size assessment. If the prediction of the possible retinal disorders can be done correctly, it would be very supportive for the ophthalmologists to provide quick treatment and prevent loss of vision. It would also benefit the telemedicine trials. It is necessary to develop efficient models and improved learning methodologies The purpose of retinal image quality is to change the image's dynamic range. so that the exact location of the abnormality can be pointed out. The retinal image quality is optimized to boost the clarity of the image. If somehow the preprocessing is done correctly, the operations such as landmark detection and localization, vessel segmentation, and identification of diseases will be done correctly. This survey presents a brief review of the existing methods for image enhancement and databases used in retinal image analysis. [ABSTRACT FROM AUTHOR]

Macular Edema is an eye disease which is caused by the swelling of the macula and results in reduced central vision. Nowadays, so many techniques are there to detect macular diseases. And the latest non-invasive imaging technique is the OCT imaging technique, in which disorders can be detected at a very early stage. Many algorithms were implemented by the researchers for the detection of ME from OCT images. However, this paper proposes a computer-aided detection for the classification of ME from OCT images using Naive Bayes classifier. This classifier’s main advantage is that classification from minimum features. Here the novelty is presented in the layer detection and texture feature extraction steps by developing a computer-aided detection algorithm for accurate ME identification. This paper provides a way to develop texture based machine learning algorithm for all biomedical imaging devices. Five distinct features (three based on the thickness profiles of the sub-retinal layers, one based on cyst fluid within the sub-retinal layers and one based on the texture feature) are extracted from the labeled images, and Naïve Bayes is trained on these. The algorithm correctly classified 196 out of 200 OCT scan images (100 ME and 100 healthy). This algorithm achieves an accuracy, sensitivity, and specificity of 98%, 97%, 99%. [ABSTRACT FROM AUTHOR]

Content-based image retrieval methods aid physicians in diagnosing the early detection of diabetic retinopathy for preventing blindness. In this work, the combination of two dimensional dual-tree complex wavelet transform (DT-CWT) and generalized Gaussian density (GGD) model is used for feature extraction. Kull-back Leibler Divergence (KLD) computes the similarity measure between two feature set. Experimental dataset includes 1200 images from MESSIDOR database. The mean precision rates at five retrieved images are obtained as 78.23% and 53.70% for Macular Edema and Retinopathy classes, respectively. Results are promising and give an indication that the dual-tree complex wavelet transform is efficient for content-based retinal image retrieval problem. [ABSTRACT FROM PUBLISHER]