Deep learning for diabetic retinopathy detection and classification based on fundus images: A review

Tsiknakis, Nikos; Theodoropoulos, Dimitris; Manikis, Georgios C.; Ktistakis, Emmanouil; Boutsora, Ourania; Bertó, Alexa; Scarpa, Fábio; Scarpa, Annalisa; Fotiadis, Dimitrios I.; Marias, Kostas

doi:10.1016/j.compbiomed.2021.104599

Cited by 153 publications

(57 citation statements)

References 129 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Since medical and biological images usually have a low data volume, non-uniform input scales, and large differences in target segmentation regions, it is difficult to achieve high-accuracy results by using traditional image segmentation methods ( 9 , 10 ). With the rapid development of deep learning techniques, convolutional neural networks have significantly improved the ability to extract features and can more easily extract and analyze the image features, thus significantly improving the image segmentation results ( 11 ). In this paper, the microscopic scale of the retinal HE images of rats and mice is the micron scale, the retina has ten layers of dense structure, and the layers are interlaced with each other with unclear boundaries.…”

Section: Discussionmentioning

confidence: 99%

Monosodium Glutamate-Induced Mouse Model With Unique Diabetic Retinal Neuropathy Features and Artificial Intelligence Techniques for Quantitative Evaluation

Liu

Huang²,

Sun

et al. 2022

Front. Immunol.

View full text Add to dashboard Cite

ObjectiveTo establish an artificial intelligence-based method to quantitatively evaluate subtle pathological changes in retinal nerve cells and synapses in monosodium glutamate (MSG) mice and provide an effective animal model and technique for quantitative evaluation of retinal neurocytopathies.MethodsICR mice were subcutaneously injected with MSG to establish a model of metabolic syndrome. We then established a mouse model of type 1 diabetes, type 2 diabetes, and KKAy mouse model as control. The HE sections of the retina were visualized using an optical microscope. AI technology was used for quantitative evaluation of the retinal lesions in each group of rats. The surface area custom parameters of the retinal nerve fiber layer (RNFL), inner plexiform layer (IPL), inner nuclear layer (INL), and outer plexiform layer (OPL) were defined as SR, SIPL, SINL, and SOPL, respectively. Their heights were defined as HR, HIPL, HINL, and HOPL, and the number of ganglion cells was defined as A. Then, the attention-augmented fully convolutional Unet network was used to segment the retinal HE images, and AI technology to identify retinal neurocytopathies quantitatively.ResultsThe attention-augmented fully convolutional Unet network increased PA and IOU parameters for INL, OPL, RNFL, and ganglion cells and was superior in recognizing fine structures. A quantitative AI identification of the height of each layer of the retina showed that the heights of the IPL and INL of the MSG model were significantly less than those of the control groups; the retinas of the other diabetic models did not exhibit this pathological feature. The RNFLs of type 2 diabetes were thinner, and the characteristics of retinopathy were not obvious in the other animal models. The pathological changes seen on HE images were consistent with the results of the quantitative AI evaluation. Immunohistochemistry results showed that NMDAR2A, GluR2, and NRG1 were significantly downregulated in the retina of MSG mice.ConclusionsThe MSG retinopathy model is closely associated with neurotransmitter abnormalities and exhibits important characteristics of retinal neurodegeneration, making it suitable for studying retinal neurocytopathies. The AI recognition technology for retinal images established in the present study can be used for the quantitative and objective evaluation of drug efficacy.

show abstract

Section: Discussionmentioning

confidence: 99%

Monosodium Glutamate-Induced Mouse Model With Unique Diabetic Retinal Neuropathy Features and Artificial Intelligence Techniques for Quantitative Evaluation

Liu

Huang²,

Sun

et al. 2022

Front. Immunol.

View full text Add to dashboard Cite

show abstract

“…Nevertheless, it should be noted that for a proper assessment, correct segmentation is necessary, and often there is improvement in NVC recognition with manual segmentation [79,81,86,93,103,104]. In the future, the accuracy of NVC detection in vitreoretinal slabs might increase with improvement of auto segmentation, deep learning and artificial intelligence [81,86,[107][108][109][110][111]. IRMAs seem to generate higher false positive rates due to the retinal slab image, despite the vitreoretinal slab image not showing extension into the vitreous cavity [82].…”

Section: Discussionmentioning

confidence: 99%

Update on Optical Coherence Tomography and Optical Coherence Tomography Angiography Imaging in Proliferative Diabetic Retinopathy

2021

View full text Add to dashboard Cite

Proliferative diabetic retinopathy (PDR) is a major cause of blindness in diabetic individuals. Optical coherence tomography (OCT) and OCT-angiography (OCTA) are noninvasive imaging techniques useful for the diagnosis and assessment of PDR. We aim to review several recent developments using OCT and discuss their present and potential future applications in the clinical setting. An electronic database search was performed so as to include all studies assessing OCT and/or OCTA findings in PDR patients published from 1 January 2020 to 31 May 2021. Thirty studies were included, and the most recently published data essentially focused on the higher detection rate of neovascularization obtained with widefield-OCT and/or OCTA (WF-OCT/OCTA) and on the increasing quality of retinal imaging with quality levels non-inferior to widefield-fluorescein angiography (WF-FA). There were also significant developments in the study of retinal nonperfusion areas (NPAs) using these techniques and research on the impact of PDR treatment on NPAs and on vascular density. It is becoming increasingly clear that it is critical to use adequate imaging protocols focused on optimized segmentation and maximized imaged retinal area, with ongoing technological development through artificial intelligence and deep learning. These latest findings emphasize the growing applicability and role of noninvasive imaging in managing PDR with the added benefit of avoiding the repetition of invasive conventional FA.

show abstract

“…Since a comprehensive list of ophthalmology image datasets has been provided in previous studies [ 3 , 11 ], we found no reason to discuss ocular image datasets in this work. Traditional data augmentation refers to an increase in the number of training examples through the rotation, flipping, cropping, translation, and scaling of existing images to improve the performance of deep learning models, and can also be used to train GAN models.…”

Section: Reviewmentioning

confidence: 99%

Application of generative adversarial networks (GAN) for ophthalmology image domains: a survey

et al. 2022

View full text Add to dashboard Cite

Background Recent advances in deep learning techniques have led to improved diagnostic abilities in ophthalmology. A generative adversarial network (GAN), which consists of two competing types of deep neural networks, including a generator and a discriminator, has demonstrated remarkable performance in image synthesis and image-to-image translation. The adoption of GAN for medical imaging is increasing for image generation and translation, but it is not familiar to researchers in the field of ophthalmology. In this work, we present a literature review on the application of GAN in ophthalmology image domains to discuss important contributions and to identify potential future research directions. Methods We performed a survey on studies using GAN published before June 2021 only, and we introduced various applications of GAN in ophthalmology image domains. The search identified 48 peer-reviewed papers in the final review. The type of GAN used in the analysis, task, imaging domain, and the outcome were collected to verify the usefulness of the GAN. Results In ophthalmology image domains, GAN can perform segmentation, data augmentation, denoising, domain transfer, super-resolution, post-intervention prediction, and feature extraction. GAN techniques have established an extension of datasets and modalities in ophthalmology. GAN has several limitations, such as mode collapse, spatial deformities, unintended changes, and the generation of high-frequency noises and artifacts of checkerboard patterns. Conclusions The use of GAN has benefited the various tasks in ophthalmology image domains. Based on our observations, the adoption of GAN in ophthalmology is still in a very early stage of clinical validation compared with deep learning classification techniques because several problems need to be overcome for practical use. However, the proper selection of the GAN technique and statistical modeling of ocular imaging will greatly improve the performance of each image analysis. Finally, this survey would enable researchers to access the appropriate GAN technique to maximize the potential of ophthalmology datasets for deep learning research.

show abstract

Deep learning for diabetic retinopathy detection and classification based on fundus images: A review

Cited by 153 publications

References 129 publications

Monosodium Glutamate-Induced Mouse Model With Unique Diabetic Retinal Neuropathy Features and Artificial Intelligence Techniques for Quantitative Evaluation

Monosodium Glutamate-Induced Mouse Model With Unique Diabetic Retinal Neuropathy Features and Artificial Intelligence Techniques for Quantitative Evaluation

Update on Optical Coherence Tomography and Optical Coherence Tomography Angiography Imaging in Proliferative Diabetic Retinopathy

Application of generative adversarial networks (GAN) for ophthalmology image domains: a survey

Contact Info

Product

Resources

About