Fundoscopic Images Research Articles

PS-R04-8: DIFFERENTIAL RECOVERY OF IMPAIRED ENDOTHELIAL AND RENAL FUNCTION IN A CASE OF MALIGNANT HYPERTENSION

A 37-year-old man with no past history of hypertension presented with no symptoms but with acutely elevated blood pressure (230/140mmHg) and proteinuria (> 300 mg/dL). Fundoscopic appearance represented grade IV hypertensive retinopathy with papilledema, flamed-shaped hemorrhages, exudates and arteriovenous nipping, indicating typical features of the malignant hypertension. The laboratory tests at admission indicated the possible organ damages in kidney and heart (eGFR 21 mL/min/1.73m2, BNP 263pg/mL). The other examinations are normal. After 20 days of treatments with amlodipine, azilsartan and carvedilol, the average of multiple blood pressure measurements and BNP showed almost normal levels (146/87mmHg, BNP < 10pg/mL), although eGFR level was still under 30. In addition, at this time point, flow mediated dilation (FMD), a noninvasive index of vascular endothelial function measured in forearm, showed significant impairment of endothelial function (%FMD 2.6). At 60 days, fundoscopic images represented entire restoration of retinopathy with disappearance of papilledema, hemorrhage and exudates. In contrast, however, both endothelial and renal function were still under normal levels (%FMD 3.2, eGFR 28.2 mL/min/1.73m2). Notably, FMD showed improvement of sustained endothelial dysfunction at day 90 (% FMD 7.5), while a little recovery of renal function was observed at 360 days (eGFR 35.4 mL/min/1.73m2). This case represents the possibility that damaged endothelial and renal function in acute severe hypertension seem to be still reversal with appropriate anti-hypertensive treatment, while differential time-course recovery in each organ might be existed.

Prediction of White Matter Hyperintensity in Brain MRI Using Fundus Photographs via Deep Learning.

Purpose: We investigated whether a deep learning algorithm applied to retinal fundoscopic images could predict cerebral white matter hyperintensity (WMH), as represented by a modified Fazekas scale (FS), on brain magnetic resonance imaging (MRI). Methods: Participants who had undergone brain MRI and health-screening fundus photography at Hallym University Sacred Heart Hospital between 2010 and 2020 were consecutively included. The subjects were divided based on the presence of WMH, then classified into three groups according to the FS grade (0 vs. 1 vs. 2+) using age matching. Two pre-trained convolutional neural networks were fine-tuned and evaluated for prediction performance using 10-fold cross-validation. Results: A total of 3726 fundus photographs from 1892 subjects were included, of which 905 fundus photographs from 462 subjects were included in the age-matched balanced dataset. In predicting the presence of WMH, the mean area under the receiver operating characteristic curve was 0.736 ± 0.030 for DenseNet-201 and 0.724 ± 0.026 for EfficientNet-B7. For the prediction of FS grade, the mean accuracies reached 41.4 ± 5.7% with DenseNet-201 and 39.6 ± 5.6% with EfficientNet-B7. The deep learning models focused on the macula and retinal vasculature to detect an FS of 2+. Conclusions: Cerebral WMH might be partially predicted by non-invasive fundus photography via deep learning, which may suggest an eye–brain association.

Experimental Autoimmune Uveitis: An Intraocular Inflammatory Mouse Model.

Experimental Autoimmune Uveitis (EAU) is driven by immune cells responding to self-antigens. Many features of this non-infectious, intraocular inflammatory disease model recapitulate the clinical phenotype of posterior uveitis affecting humans. EAU has been used reliably to study the efficacy of novel inflammatory therapeutics, their mode of action and to further investigate the mechanisms that underpin disease progression of intraocular disorders. Here, we provide a detailed protocol on EAU induction in the C57BL/6J mouse - the most widely used model organism with susceptibility to this disease. Clinical assessment of disease severity and progression will be demonstrated using fundoscopy, histological examination and fluorescein angiography. The induction procedure involves subcutaneousinjection of an emulsion containing a peptide (IRBP1-20) from the ocular protein interphotoreceptor retinoid binding protein (also known as retinol binding protein 3), Complete Freund's Adjuvant (CFA) and supplemented with killed Mycobacterium tuberculosis. Injection of this viscous emulsion on the back of the neck is followed by a single intraperitoneal injection of Bordetella pertussis toxin. At the onset of symptoms (day 12-14) and under general anesthesia, fundoscopic images are taken to assess disease progression through clinical examination. These data can be directly compared with those at later timepoints and peak disease (day 20-22) with differences analyzed. At the same time, this protocol allows the investigator to assess potential differences in vessel permeability and damage using fluorescein angiography. EAU can be induced in other mouse strains - both wildtype or genetically modified - and combined with novel therapies offering flexibility for studying drug efficacy and/or disease mechanisms.

Leber's Congenital Amaurosis: Current Concepts of Genotype-Phenotype Correlations.

Leber’s congenital amaurosis (LCA), one of the most severe inherited retinal dystrophies, is typically associated with extremely early onset of visual loss, nystagmus, and amaurotic pupils, and is responsible for 20% of childhood blindness. With advances in molecular diagnostic technology, the knowledge about the genetic background of LCA has expanded widely, while disease-causing variants have been identified in 38 genes. Different pathogenetic mechanisms have been found among these varieties of genetic mutations, all of which result in the dysfunction or absence of their encoded proteins participating in the visual cycle. Hence, the clinical phenotypes also exhibit extensive heterogenicity, including the course of visual impairment, involvement of the macular area, alteration in retinal structure, and residual function of the diseased photoreceptor. By reviewing the clinical course, fundoscopic images, optical coherent tomography examination, and electroretinogram, genotype-phenotype correlations could be established for common genetic mutations in LCA, which would benefit the timing of the diagnosis and thus promote early intervention. Gene therapy is promising in the management of LCA, while several clinical trials are ongoing and preliminary success has been announced, focusing on RPE65 and other common disease-causing genes. This review provides an update on the genetics, clinical examination findings, and genotype-phenotype correlations in the most well-established causative genetic mutations of LCA.

More roles of IL-4 than being as an antagonist of IFN-γ in experimental autoimmune uveitis

Abstract IL-4 appears to be beneficial for some autoimmune diseases. Despite its role in antagonizing pathogenic Th1 response, our recent work has demonstrated IL-4 possesses functions affecting the regulatory T (Treg) cells. Here we aimed to identify the role of IL-4 in the development of experimental autoimmune uveitis (EAU). EAU was induced in WT and IL-4 KO mice via active interphotoreceptor retinoid binding protein (IRBP) immunization, and the disease development was monitored by funduscopic and histology analysis as well as the IRBP-specific immune responses by flow cytometric analysis and ELISA. It showed that the onset of disease, if the clinical score equals to 1, was day 14 and 20 in IL-4KO and WT mice, respectively. The fundoscopic images demonstrated the more and bigger size of lesions, increased leukocyte infiltration, vasculitis and neovascularization in IL-4KO mice. The histological analysis also confirmed the mononuclear cell infiltration. Moreover, Th1 effector is known as the dominant pathogenic mediators, whereas Treg cells play an important and indispensable role in control of EAU progression and disease remission. The splenic T cells from IL-4 KO EAU animals produced significantly elevated level of IFN-γ in responding to IRBP stimulation. However, there seems no significant but some alternation in the presence of Treg cells (Foxp3+) among CD4+CD25+ T cells in the spleens and lymph nodes of IL-4 KO and WT mice. These results demonstrate the lack of IL-4 disrupting the induction of EAU and leading to an increased disease severity through not only antagonizing IFN-γ but also involved mechanisms of immune-regulation. Therefore, IL-4 might play more roles than being just an antagonist of IFN-γ in autoimmune uveitis.

CPGAN: Conditional patch‐based generative adversarial network for retinal vesselsegmentation

Retinal blood vessels, the diagnostic bio-marker of ophthalmologic and diabetic retinopathy, utilise thick and thin vessels for diagnostic and monitoring purposes. The existing deep learning methods attempt to segment the retinal vessels using a unified loss function. However, a difference in spatial features of thick and thin vessels and a biased distribution creates an imbalanced thickness, rendering the unified loss function to be useful only for thick vessels. To address this challenge, a patch-based generative adversarial network-based technique is proposed which iteratively learns both thick and thin vessels in fundoscopic images. It introduces an additional loss function that allows the generator network to learn thin and thick vessels, while the discriminator network assists in segmenting out both vessels as a combined objective function. Compared with state-of-the-art techniques, the proposed model demonstrates the enhanced accuracy, sensitivity, specificity, and area under the receiver operating characteristic curves on STARE, DRIVE, and CHASEDB1 datasets.

Preliminary results of a semi-automated platform for pixel-based quantification of choroidal naevus progression

To evaluate a novel method for choroidal naevus progression monitoring using semiautomated software for pixel-based quantification. Retrospective analysis of patients with choroidal melanotic lesions. In total, 87 eyes fulfilled our inclusion criteria. The progression of the choroidal lesions was evaluated by 2 masked and independent investigators. Total naevus area was measured on fundoscopic images using ImageJ semiautomated segmentation at baseline examination and 5 years later. The efficacy of our method was also tested on a separate patient group applying test-retest analysis. Pixel-based analysis for objective quantification, displayed high repeatability, and reliability. We identified 4 cases of naevus size progression using the described semiautomated technique, while only 2 cases of progression were reported by simple observation. The novel method performance displayed 50% increased sensitivity over simple observation and 100% specificity. The presented method of displayed higher sensitivity and specificity compared to conventional observation and photograph qualification. Our results imply that the proposed methodology may be applicable as an accurate, easily employed diagnostic tool for the early detection of size progression in melanotic choroidal lesions.

30-LB: Diabetes Complications Including Nephropathy, Neuropathy, and Retinopathy Can Now Be Studied in a New Spontaneous Autoimmune Diabetes Transgenic Model

We have developed a spontaneous humanized transgenic mouse model of autoimmune diabetes chronically develops diabetic complications. Our humanized mouse develop diabetic nephropathy. The basal glomerular membrane thickness was doubled in size at 4 months of age as shown by electron microscopy. The manifestation of glomerulopathy subsequently correlates with microalbuminuria. At 6 month, mice show advanced features of diabetic nephropathy like focal segmental/global glomerular sclerosis and focal tubular atrophy. The typical features of Kimmelstiel-Wilson are seen when severe chronic kidney disease develops. Our humanized mice develop diabetic retinopathy. We have noted a significant increase in acellular capillaries of the retina at the age of 6 months. Ocular tomography revealed a significant reduction in total retinal thickening. Fundoscopic retinal images revealed a classic vascular leakage as observed in humans. Our humanized mice also show diabetic neuropathy. We have observed an impaired sciatic and optic nerve conduction. Specifically, optic nerves at 6-months old diabetic mice showed an increased refractory period of the second compound action potential (CAP). This increase in the refractory period of CAP in the diabetic optic nerves is consistent with the reduced conduction velocity. However, a limb vasculopathy has not been observed so far. In our mouse model, the mouse MHC-II was replaced with human DQ8 in all antigen-presenting cells. The mouse also expresses human GAD65 in pancreatic beta cells. The diabetic phenotype of our mouse mimics the human disease progression. Importantly, our mouse model has a genetic background that compromises beta-cell neogenesis and/or regeneration (1). Our humanized mice model has the closest known phenotype to human diabetes and is most suitable for addressing physiopathology and treatment of diabetes complications. (U.S. Provisional Patent #62/713,827). Disclosure S. Imam: None. J.C. Jaume: None. Funding University of Toledo

Deep Learning Applications in Chest Radiography and Computed Tomography: Current State of the Art.

Deep learning is a genre of machine learning that allows computational models to learn representations of data with multiple levels of abstraction using numerous processing layers. A distinctive feature of deep learning, compared with conventional machine learning methods, is that it can generate appropriate models for tasks directly from the raw data, removing the need for human-led feature extraction. Medical images are particularly suited for deep learning applications. Deep learning techniques have already demonstrated high performance in the detection of diabetic retinopathy on fundoscopic images and metastatic breast cancer cells on pathologic images. In radiology, deep learning has the opportunity to provide improved accuracy of image interpretation and diagnosis. Many groups are exploring the possibility of using deep learning-based applications to solve unmet clinical needs. In chest imaging, there has been a large effort to develop and apply computer-aided detection systems for the detection of lung nodules on chest radiographs and chest computed tomography. The essential limitation to computer-aided detection is an inability to learn from new information. To overcome these deficiencies, many groups have turned to deep learning approaches with promising results. In addition to nodule detection, interstitial lung disease recognition, lesion segmentation, diagnosis and patient outcomes have been addressed by deep learning approaches. The purpose of this review article was to cover the current state of the art for deep learning approaches and its limitations, and some of the potential impact on the field of radiology, with specific reference to chest imaging.

Agreement between retinal images obtained via smartphones and images obtained with retinal cameras or fundoscopic exams - systematic review and meta-analysis.

BackgroundSmartphone fundoscopy is a new option for visualizing the ocular fundus but must be validated before being included in population-based examinations. Our aim was to evaluate the quality of fundoscopic images obtained via smartphone and to compare their agreement with retinal camera images or clinical examination.MethodsThe database for this study included all observational studies with smartphone fundoscopy that have comparative analyses with the gold standard methods.ResultsOut of 121 potentially relevant studies, nine were included in this analysis, comprising a total of 4,219 eyes. Mean age was 56.6 years (SD±8.5). Combined kappa (κ) agreement statistics were equal to 77.77% (95% CI: 70.34%, 83.70%). No heterogeneity was measured by random effects (I2=zero).ConclusionFundoscopic images obtained by using smartphones have substantial agreement with gold standards for clinical or photographic exams.

Towards Accurate Segmentation of Retinal Vessels and the Optic Disc in Fundoscopic Images with Generative Adversarial Networks.

Automatic segmentation of the retinal vasculature and the optic disc is a crucial task for accurate geometric analysis and reliable automated diagnosis. In recent years, Convolutional Neural Networks (CNN) have shown outstanding performance compared to the conventional approaches in the segmentation tasks. In this paper, we experimentally measure the performance gain for Generative Adversarial Networks (GAN) framework when applied to the segmentation tasks. We show that GAN achieves statistically significant improvement in area under the receiver operating characteristic (AU-ROC) and area under the precision and recall curve (AU-PR) on two public datasets (DRIVE, STARE) by segmenting fine vessels. Also, we found a model that surpassed the current state-of-the-art method by 0.2 - 1.0% in AU-ROC and 0.8 - 1.2% in AU-PR and 0.5 - 0.7% in dice coefficient. In contrast, significant improvements were not observed in the optic disc segmentation task on DRIONS-DB, RIM-ONE (r3) and Drishti-GS datasets in AU-ROC and AU-PR.

Nonmydriatic Fundoscopic Imaging Using the Pan Optic iExaminer System in the Pediatric Emergency Department.

The primary objective of this study was to determine the percentage of clinically adequate (CA) fundoscopic images that could be obtained using the Pan Optic iExaminer system to perform nonmydriatic fundoscopic imaging in the pediatric emergency department (ED). Secondary objectives were to identify target age groups in which this technology is best utilized and evaluate the overall ease of use in this setting. Children 18 years of age or less who presented to the pediatric ED with a non-eye-related chief complaint were enrolled and stratified by age group (0-2, 2-6, and 6-18 years). Each enrolled patient underwent a bilateral eye examination using the Pan Optic iExaminer system. Images were submitted for review to a pediatric ophthalmologist and were graded based on clarity and field of view. Ease of use was defined as 80% of patients having at least one image of quality to be considered "clinically adequate" for obtaining a full view of the optic nerve, examination time for both eyes 15 minutes or less, and three attempts or less for each eye. Overall, 91.06% (95% confidence interval [CI] = 86.01% to 96.1%) of children ages 2-18 years had at least one CA image obtained. A total of 16% (95% CI = 7% to 26%) of children 0-2 years, 85% (95% CI = 76% to 94.15%) of children 2-6 years, and 9% (95% CI = 92% to 100%) of 6-18 years had at least one CA image. The median total examination time was 3 minutes 24 seconds (interquartile range = 2 minutes 27 seconds to 4 minutes 49 seconds). Fundoscopic images were consistently obtained using the Pan Optic iExaminer system in the pediatric ED particularly in children 2-18 years of age. CA images were obtained in children less than 2 years old, but less consistently.

Efficacy of Retrobulbar Hyaluronidase Injection for Vision Loss Resulting from Hyaluronic Acid Filler Embolization.

Vision loss is a rare but serious complication of facial hyaluronic acid (HA) filler injection, for which there is no proven rescue therapy. Retrobulbar hyaluronidase injection is advocated by many plastic surgeons as an emergency treatment, but has not been carefully assessed for its efficacy. To evaluate the efficacy of retrobulbar hyaluronidase injection as a rescue treatment for vision loss caused by HA filler embolization. Patients with vision loss caused by HA filler embolization were treated with retrobulbar hyaluronidase injection. Their visual acuity and fundoscopic images before and after treatment were analyzed for efficacy assessment. One patient with branch retinal artery occlusion (BRAO), one patient with posterior ischemic optic neuropathy (PION), one patient with ophthalmic artery occlusion, and one patient with both BRAO and PION were treated with one or two retrobulbar injections of 1500 or 3000 units hyaluronidase. No patients demonstrated substantial retinal artery recanalization or vision acuity improvement after treatment. One or two retrobulbar injections of 1500 to 3000 IU hyaluronidase are unable to recanalize retinal artery occlusion or improve the visual outcome of patients who presented with vision loss caused by HA filler embolization at least four hours after onset. 4.

Spatiotemporal Patterns of Tumor Occurrence in Children with Intraocular Retinoblastoma.

PurposeTo accurately map the retinal area covered by tumor in a prospectively enrolled cohort of children diagnosed with retinoblastoma.MethodsOrbital MRI in 106 consecutive retinoblastoma patients (44 bilateral) was analyzed. For MRI-visible tumors, the polar angle and angle of eccentricity of points defining tumor perimeter on the retina were determined by triangulation from images in three orthogonal planes. The centroid of the mapped area was calculated to approximate tumor origin, and the location and cumulative tumor burden were analyzed in relation to mutation type (germline vs. somatic), tumor area, and patient age at diagnosis. Location of small tumors undetected by MRI was approximated with fundoscopic images.ResultsMapping was successful for 129 tumors in 91 eyes from 67 patients (39 bilateral, 43 germline mutation). Cumulative tumor burden was highest within the macula and posterior pole and was asymmetrically higher within the inferonasal periphery. Tumor incidence was lowest in the superotemporal periphery. Tumor location varied with age at diagnosis in a complex pattern. Tumor location was concentrated in the macula and superonasal periphery in patients <5.6 months, in the inferotemporal quadrant of the posterior pole in patients 5.6-8.8 months, in the inferonasal quadrant in patients 8.8-13.2 months, and in the nasal and superotemporal periphery in patients >13.2 months. The distribution of MRI-invisible tumors was consistent with the asymmetry of mapped tumors.ConclusionsMRI-based mapping revealed a previously unrecognized pattern of retinoblastoma localization that evolves with age at diagnosis. The structured spatiotemporal distribution of tumors may provide valuable clues about cellular or molecular events associated with tumorigenesis in the developing retina.

Stanford University Network for Diagnosis of Retinopathy of Prematurity (SUNDROP): Four-years of Screening with Telemedicine

Purpose: To report the four-year experience of the Stanford University Network for Diagnosis of Retinopathy of Prematurity (SUNDROP) telemedicine initiative, which was developed to reduce the risk of blindness from retinopathy of prematurity (ROP).Materials and methods: A retrospective analysis of the SUNDROP archival data between 12/1/2005 and 11/30/2009. A total of 410 consecutively enrolled infants meeting ROP screening criteria had nurse-obtained fundoscopic images evaluated remotely by an ROP specialist. Every infant then received at least one dilated bedside binocular indirect ophthalmoscope (BIO) examination within one week of discharge. All infants were then followed with both telemedicine images and bedside evaluation in clinic according to recommended screening timelines. Primary outcomes were treatment-warranted ROP (TW-ROP), defined as Early Treatment of ROP Type 1, and adverse anatomical outcomes.Results: The SUNDROP telemedicine screening initiative has not missed any TW-ROP in its four-year evaluation period. A total of 410 infants (820 eyes) were imaged, resulting in 1486 examinations and 18,097 unique images. An average of 12.2 images were obtained per patient. Fourteen infants met TW-ROP criteria on telemedicine examination. After bedside evaluation, 13 infants required laser photocoagulation and one was followed until he spontaneously regressed. Infants with TW-ROP had a significantly lower gestational age (24.9 weeks), birth weight (658.7 grams), and were more likely to be male than the no TW-ROP cohort (all p values <0.00001). Telemedicine had a calculated sensitivity of 100%, specificity of 99.8%, positive predicative value of 92.9% and negative predictive value of 100% for the detection of TW-ROP. No patient progressed to retinal detachment or any adverse anatomical outcome.Conclusions: The SUNDROP initiative demonstrated a high degree of diagnostic reliability and was able to capture all infants with TW-ROP. Telemedicine offers a cost-effective, reliable and accurate screening methodology for identifying infants with TW-ROP without sacrificing quality of care.

Lokalisation von Bevacizumab in der Netzhaut von Affen nach einer intravitrealen Bevacizumab-Injektion

The aim was to investigate to what extent the vascular endothelial growth factor (VEGF) antibody bevacizumab is able to penetrate the retina in primates after an intravitreal injection of Avastin. Monkeys (Macaca fascicularis) were injected intravitreally with radioactively labelled and free Avastin. The animals were sacrificed 1, 4, 7, or 14 days after the injection, and the eyes were examined histologically and immunohistochemically. Blood samples were also taken on several days. In the fundoscopic images, no pathologic changes could be found during the experiment. Using immunohistochemistry, bevacizumab was found in the choroid and the inner layers of the retina 1 day after the injection. Bevacizumab penetrated more quickly in the fovea than in the rest of the retina. It was also encountered in the photoreceptors and blood vessels. When (125)I-labelled Avastin was used, radioactivity could be determined in the blood serum 1 day after the injection. The results show that the bevacizumab molecule can penetrate the retina after intravitreal injection of Avastin. However, there is an active uptake in the retinal cells.