The purpose of this study was to compare optical coherence tomography (OCT) angiography to standard fluorescein angiography (FA) in the grading of diabetic macular ischemia.
In our study, OCT ...angiography and traditional FA images were acquired from 24 diabetic patients. The level of diabetic macular ischemia in the superficial capillary plexus was graded with standard Early Treatment Diabetic Retinopathy Study (ETDRS) protocols and a comparison between conventional FA and OCT angiography was performed. The deep vascular plexus and choriocapillaris were also graded for macular ischemia. Additionally, flow indices were analyzed for all OCT angiography images.
We identified moderate agreement between diabetic macular ischemia grades for conventional FA and OCT angiography (weighted κ of 0.53 and 0.41). In addition, the intergrader agreement for the superficial, deep, and choriocapillaris scores was substantial (weighted κ of 0.65, 0.61, and 0.65, respectively). Finally, the parafoveal flow indices were shown to have a statistically significant relationship with diabetic macular ischemia grades for the superficial capillary plexus (P = 0.04) and choriocapillaris (P = 0.036), with a trend toward significance for the deep capillary plexus (P = 0.13).
We demonstrated moderate agreement between diabetic macular ischemia grading results for OCT angiography and conventional FA using standard ETDRS protocols. We also showed that OCT angiography images could be graded for diabetic macular ischemia with substantial intergrader agreement.
For nAMD, patients often require in excess of 50 injections, sometimes for more than 10 years, or the remainder of their lives.3 For DMO, patients typically require 12–15 injections in the first 3 ...years, at which time loss to follow-up is close to one-third.4 Treatment burden is thought to be a key reason the benefits of anti-VEGF in clinical trials have never fully translated to clinical practice.2 Given the ageing population and growing obesity epidemic, both nAMD and DMO are projected to increase significantly in the future.5,6 A key question for researchers has therefore evolved; namely, how can we improve the durability of drug therapy? In addition to binding VEGF-A, faricimab targets angiopoietin-2, a ligand operating distinct from the VEGF pathway that regulates vascular destabilisation and inflammation.7 These multicentre, randomised, double-masked, non-inferiority trials compare the safety and efficacy of faricimab with aflibercept, an existing anti-VEGF drug that is widely used.9 Jeffrey Heier and colleagues7 report week 48 results of the identical TENAYA and LUCERNE trials for nAMD. 1329 patients with nAMD (793 60% female, 536 40% male; 1153 87% White, 133 10% Hispanic or Latino, 126 9% Asian, four <1% American Indian or Alaska Native, and ten 1% Black or African American) were randomly assigned to receive faricimab at an interval between 8 and 16 weeks, or aflibercept every 8 weeks (in both cases preceded by loading doses every 4 weeks). PAK declares fees from DeepMind for consultancy in automated analysis of retinal imaging with deep learning, for Roche in personalised health care, for Novartis in personalised health care, for Apellis in automated analysis of geographic atrophy, and for BitFount in privacy protecting machine learning.
We evaluate how deep learning can be applied to extract novel information such as refractive error from retinal fundus imaging.
Retinal fundus images used in this study were 45- and 30-degree field ...of view images from the UK Biobank and Age-Related Eye Disease Study (AREDS) clinical trials, respectively. Refractive error was measured by autorefraction in UK Biobank and subjective refraction in AREDS. We trained a deep learning algorithm to predict refractive error from a total of 226,870 images and validated it on 24,007 UK Biobank and 15,750 AREDS images. Our model used the "attention" method to identify features that are correlated with refractive error.
The resulting algorithm had a mean absolute error (MAE) of 0.56 diopters (95% confidence interval CI: 0.55-0.56) for estimating spherical equivalent on the UK Biobank data set and 0.91 diopters (95% CI: 0.89-0.93) for the AREDS data set. The baseline expected MAE (obtained by simply predicting the mean of this population) was 1.81 diopters (95% CI: 1.79-1.84) for UK Biobank and 1.63 (95% CI: 1.60-1.67) for AREDS. Attention maps suggested that the foveal region was one of the most important areas used by the algorithm to make this prediction, though other regions also contribute to the prediction.
To our knowledge, the ability to estimate refractive error with high accuracy from retinal fundus photos has not been previously known and demonstrates that deep learning can be applied to make novel predictions from medical images.
Abstract Age-related macular degeneration (AMD) is the leading cause of severe visual loss in people aged 50 years or older in the developed world. In recent years, major advances have been made in ...the treatment of AMD, with the introduction of anti-angiogenic agents, offering the first hope of significant visual recovery for patients with neovascular AMD. In line with these advances, a new imaging modality—optical coherence tomography (OCT)—has emerged as an essential adjunct for the diagnosis and monitoring of patients with AMD. The ability to accurately interpret OCT images is thus a prerequisite for both retina specialists and comprehensive ophthalmologists. Despite this, the relatively recent introduction of OCT and the absence of formal training, coupled with rapid evolution of the technology, may make such interpretation difficult. These problems are compounded by the phenotypically heterogeneous nature of AMD and its complex morphology as visualized using OCT. We address these issues by summarizing the current understanding of OCT image interpretation in patients with AMD and describe how OCT can best be applied in clinical practice.
The purpose of this review is to describe the current status of automated deep learning in healthcare and to explore and detail the development of these models using commercially available platforms. ...We highlight key studies demonstrating the effectiveness of this technique and discuss current challenges and future directions of automated deep learning.
There are several commercially available automated deep learning platforms. Although specific features differ between platforms, they utilise the common approach of supervised learning. Ophthalmology is an exemplar speciality in the area, with a number of recent proof-of-concept studies exploring classification of retinal fundus photographs, optical coherence tomography images and indocyanine green angiography images. Automated deep learning has also demonstrated impressive results in other specialities such as dermatology, radiology and histopathology.
Automated deep learning allows users without coding expertise to develop deep learning algorithms. It is rapidly establishing itself as a valuable tool for those with limited technical experience. Despite residual challenges, it offers considerable potential in the future of patient management, clinical research and medical education.
http://links.lww.com/COOP/A44.
To determine the effects of age, sex, and race on the retinal nerve fiber layer (RNFL) in the normal human eye as measured by the spectral domain optical coherence tomography (SD-OCT) Spectralis ...machine (Heidelberg Engineering).
Peripapillary SD-OCT RNFL thickness measurements were determined in normal subjects seen at a university-based clinic. One randomly selected eye per subject was used for analysis in this cross-sectional study. Multiple regression analysis was applied to assess the effects of age, sex, ethnicity, and mean refractive error on peripapillary RNFL thickness. Results are expressed as means±SD wherever applicable.
The study population consisted of 190 healthy participants from 9 to 86 years of age. Of the 190 participants, 62 (33%) were men, 125 (66%) Caucasians, 26 (14%) African Americans, 14 (7%) Hispanics, 16 (8%) Asians, and 9 (5%) other races. The mean RNFL thickness for the normal population studied was 97.3 ± 9.6 µm. Normal RNFL thickness values follow the ISNT rule with decreasing RNFL thickness values starting from the thickest quadrant inferiorly to the thinnest quadrant temporally: inferior quadrant (126 ± 15.8), superior quadrant (117.2±16.13), nasal quadrant (75 ± 13.9), and temporal quadrant (70.6 ± 10.8 µm). Thinner RNFL measurements were associated with older age (P<0.001); being Caucasian, versus being either Hispanic or Asian (P=0.02 and 0.009, respectively); or being more myopic (P<0.001). For every decade of increased age, mean RNFL thickness measured thinner by approximately 1.5 µm (95% confidence interval, 0.24-0.07). Comparisons between ethnic groups revealed that Caucasians had mean RNFL values (96 ± 9.2 µm) slightly thinner than those of Hispanics (102.9 ± 11 µm; P=0.02) or Asians (100.7 ± 8.5 µm; P=0.009). African Americans RNFL values (99.2 ± 10.2 µm) were not significantly different when compared with Caucasians. There was no relationship between RNFL thickness and sex.
The thickest RNFL measurements were found in the inferior quadrant, followed by the superior, nasal, and temporal quadrants (ISNT rule applied to the RNFL). Thinner RNFL measurements were associated with older age and increasing myopia. Caucasians tend to have thinner RNFL values when compared with Hispanics and Asians. SD-OCT analysis of the normal RNFL showed results similar to time domain OCT studies.
Purpose To investigate the association between peripheral and central ischemia in diabetic retinopathy. Design Retrospective, cross-sectional. Methods Consecutive ultra-widefield fluorescein ...angiography images were collected from patients with diabetes over a 12-month period. Parameters quantified include the foveal avascular zone (FAZ) area, peripheral ischemic index, peripheral leakage index, and central retinal thickness measurements, as well as visual acuity. The peripheral ischemia or leakage index was calculated as the area of capillary nonperfusion or leakage, expressed as a percentage of the total retinal area. Results Forty-seven eyes of 47 patients were included. A moderate correlation was observed between the peripheral ischemia index and FAZ area (r = 0.49, P = .0001). A moderate correlation was also observed between the peripheral leakage index and FAZ area, but only in eyes that were laser naïve (r = 0.44, P = .02). A thinner retina was observed in eyes with macular ischemia (217 ± 81.8 μm vs 272 ± 36.0 μm) ( P = .02), but not peripheral ischemia (258 ± 76.3 μm vs 276 ± 68.0 μm) ( P = .24). The relationships between different patterns of peripheral and central macular pathology and visual acuity were evaluated in a step-wise multivariable regression model, and the variables that remained independently associated were age (r = 0.33, P = .03), FAZ area (r = 0.45, P = .02), and central retinal thickness (r = 0.38, P = .01), (R2 -adjusted = 0.36). Conclusions Ultra-widefield fluorescein angiography provides an insight into the relationships between diabetic vascular complications in the retinal periphery and central macula. Although we observed relationships between ischemia and vascular leakage in the macula and periphery, it was only macular ischemia and retinal thinning that was independently associated with a reduced visual function.
Artificial intelligence (AI) has great potential in ophthalmology. We investigated how ambiguous outputs from an AI diagnostic support system (AI-DSS) affected diagnostic responses from optometrists ...when assessing cases of suspected retinal disease. Thirty optometrists (15 more experienced, 15 less) assessed 30 clinical cases. For ten, participants saw an optical coherence tomography (OCT) scan, basic clinical information and retinal photography ('no AI'). For another ten, they were also given AI-generated OCT-based probabilistic diagnoses ('AI diagnosis'); and for ten, both AI-diagnosis and AI-generated OCT segmentations ('AI diagnosis + segmentation') were provided. Cases were matched across the three types of presentation and were selected to include 40% ambiguous and 20% incorrect AI outputs. Optometrist diagnostic agreement with the predefined reference standard was lowest for 'AI diagnosis + segmentation' (204/300, 68%) compared to 'AI diagnosis' (224/300, 75% p = 0.010), and 'no Al' (242/300, 81%, p = < 0.001). Agreement with AI diagnosis consistent with the reference standard decreased (174/210 vs 199/210, p = 0.003), but participants trusted the AI more (p = 0.029) with segmentations. Practitioner experience did not affect diagnostic responses (p = 0.24). More experienced participants were more confident (p = 0.012) and trusted the AI less (p = 0.038). Our findings also highlight issues around reference standard definition.
The introduction of fundus photography has impacted retinal imaging and retinal screening programs significantly.
Fundus cameras play a vital role in addressing the cause of preventive blindness. ...More attention is being turned to developing countries, where infrastructure and access to healthcare are limited. One of the major limitations for tele-ophthalmology is restricted access to the office-based fundus camera.
Recent advances in access to telecommunications coupled with introduction of portable cameras and smartphone-based fundus imaging systems have resulted in an exponential surge in available technologies for portable fundus photography. Retinal cameras in the near future would have to cater to these needs by featuring a low-cost, portable design with automated controls and digitalized images with Web-based transfer.
In this review, we aim to highlight the advances of fundus photography for retinal screening as well as discuss the advantages, disadvantages, and implications of the various technologies that are currently available.
Identifing potential screening tests for future cognitive decline is a priority for developing treatments for and the prevention of dementia.
To examine the potential of retinal nerve fiber layer ...(RNFL) thickness measurement in identifying those at greater risk of cognitive decline in a large community cohort of healthy people.
UK Biobank is a prospective, multicenter, community-based study of UK residents aged 40 to 69 years at enrollment who underwent baseline retinal optical coherence tomography imaging, a physical examination, and a questionnaire. The pilot study phase was conducted from March 2006 to June 2006, and the main cohort underwent examination for baseline measures from April 2007 to October 2010. Four basic cognitive tests were performed at baseline, which were then repeated in a subset of participants approximately 3 years later. We analyzed eyes with high-quality optical coherence tomography images, excluding those with eye disease or vision loss, a history of ocular or neurological disease, or diabetes. We explored associations between RNFL thickness and cognitive function using multivariable logistic regression modeling to control for demographic as well as physiologic and ocular variation.
Odds ratios (ORs) for cognitive performance in the lowest fifth percentile in at least 2 of 4 cognitive tests at baseline, or worsening results on at least 1 cognitive test at follow-up. These analyses were adjusted for age, sex, race/ethnicity, height, refraction, intraocular pressure, education, and socioeconomic status.
A total of 32 038 people were included at baseline testing, for whom the mean age was 56.0 years and of whom 17 172 (53.6%) were women. A thinner RNFL was associated with worse cognitive performance on baseline assessment. A multivariable regression controlling for potential confounders showed that those in the thinnest quintile of RNFL were 11% more likely to fail at least 1 cognitive test (95% CI, 2.0%-2.1%; P = .01). Follow-up cognitive tests were performed for 1251 participants (3.9%). Participants with an RNFL thickness in the 2 thinnest quintiles were almost twice as likely to have at least 1 test score be worse at follow-up cognitive testing (quintile 1: OR, 1.92; 95% CI, 1.29-2.85; P < .001; quintile 2: OR, 2.08; 95% CI, 1.40-3.08; P < .001).
A thinner RNFL is associated with worse cognitive function in individuals without a neurodegenerative disease as well as greater likelihood of future cognitive decline. This preclinical observation has implications for future research, prevention, and treatment of dementia.