Fundus Camera Images Research Articles

Unified deep learning model for predicting fundus fluorescein angiography image from fundus structure image

The prediction of fundus fluorescein angiography (FFA) images from fundus structural images is a cutting-edge research topic in ophthalmological image processing. Prediction comprises estimating FFA from fundus camera imaging, single-phase FFA from scanning laser ophthalmoscopy (SLO), and three-phase FFA also from SLO. Although many deep learning models are available, a single model can only perform one or two of these prediction tasks. To accomplish three prediction tasks using a unified method, we propose a unified deep learning model for predicting FFA images from fundus structure images using a supervised generative adversarial network. The three prediction tasks are processed as follows: data preparation, network training under FFA supervision, and FFA image prediction from fundus structure images on a test set. By comparing the FFA images predicted by our model, pix2pix, and CycleGAN, we demonstrate the remarkable progress achieved by our proposal. The high performance of our model is validated in terms of the peak signal-to-noise ratio, structural similarity index, and mean squared error.

Retinal Vessel Caliber Measurement Bias in Fundus Images in the Presence of the Central Light Reflex.

To investigate the agreement between a fundus camera and a scanning laser ophthalmoscope in retinal vessel caliber measurements and to identify whether the presence of the central light reflex (CLR) explains potential discrepancies. For this cross-sectional study, we obtained fundus camera and scanning laser ophthalmoscope images from 85 eyes of 85 healthy individuals (aged 50-65 years) with different blood pressure status. We measured the central retinal artery equivalent (CRAE) and central retinal artery vein equivalent (CRVE) with the Knudtson-Parr-Hubbard algorithm and assessed the CLR using a semiautomatic grading method. We used Bland-Altman plots, 95% limits of agreement, and the two-way mixed effects intraclass correlation coefficient for consistency [ICC(3,1)] to describe interdevice agreement. We used multivariable regression to identify factors associated with differences in between-device measurements. The between-device difference in CRAE (9.5 µm; 95% confidence interval, 8.0-11.1 µm) was larger than the between-device difference in CRVE (2.9 µm; 95% confidence interval, 1.3-4.5 µm), with the fundus camera yielding higher measurements (both P < 0.001). The 95% fundus camera-scanning laser ophthalmoscope limits of agreement were -4.8 to 23.9 µm for CRAE and -12.0 to 17.8 µm for CRVE. The corresponding ICCs(3,1) were 0.89 (95% confidence interval, 0.83-0.92) and 0.91 (95% confidence interval, 0.86-0.94). The between-device CRAE difference was positively associated with the presence of a CLR (P = 0.002). Fundus cameras and scanning laser ophthalmoscopes yield correlated but not interchangeable caliber measurements. The CLR induces bias in arteriolar caliber in fundus camera images, compared with scanning laser ophthalmoscope images. Refined measurements could yield better estimates of the association between retinal vessel caliber and ophthalmic or systemic disease.

Clinical validation of a smartphone-based retinal camera for diabetic retinopathy screening

AimsThis study aims to compare the performance of a handheld fundus camera (Eyer) and standard tabletop fundus cameras (Visucam 500, Visucam 540, and Canon CR-2) for diabetic retinopathy and diabetic macular edema screening.MethodsThis was a multicenter, cross-sectional study that included images from 327 individuals with diabetes. The participants underwent pharmacological mydriasis and fundus photography in two fields (macula and optic disk centered) with both strategies. All images were acquired by trained healthcare professionals, de-identified, and graded independently by two masked ophthalmologists, with a third senior ophthalmologist adjudicating in discordant cases. The International Classification of Diabetic Retinopathy was used for grading, and demographic data, diabetic retinopathy classification, artifacts, and image quality were compared between devices. The tabletop senior ophthalmologist adjudication label was used as the ground truth for comparative analysis. A univariate and stepwise multivariate logistic regression was performed to determine the relationship of each independent factor in referable diabetic retinopathy.ResultsThe mean age of participants was 57.03 years (SD 16.82, 9–90 years), and the mean duration of diabetes was 16.35 years (SD 9.69, 1–60 years). Age (P = .005), diabetes duration (P = .004), body mass index (P = .005), and hypertension (P < .001) were statistically different between referable and non-referable patients. Multivariate logistic regression analysis revealed a positive association between male sex (OR 1.687) and hypertension (OR 3.603) with referable diabetic retinopathy. The agreement between devices for diabetic retinopathy classification was 73.18%, with a weighted kappa of 0.808 (almost perfect). The agreement for macular edema was 88.48%, with a kappa of 0.809 (almost perfect). For referable diabetic retinopathy, the agreement was 85.88%, with a kappa of 0.716 (substantial), sensitivity of 0.906, and specificity of 0.808. As for image quality, 84.02% of tabletop fundus camera images were gradable and 85.31% of the Eyer images were gradable.ConclusionsOur study shows that the handheld retinal camera Eyer performed comparably to standard tabletop fundus cameras for diabetic retinopathy and macular edema screening. The high agreement with tabletop devices, portability, and low costs makes the handheld retinal camera a promising tool for increasing coverage of diabetic retinopathy screening programs, particularly in low-income countries. Early diagnosis and treatment have the potential to prevent avoidable blindness, and the present validation study brings evidence that supports its contribution to diabetic retinopathy early diagnosis and treatment.

Comparison of 50° handheld fundus camera versus ultra-widefield table-top fundus camera for diabetic retinopathy detection and grading.

To compare the performance of a handheld fundus camera with standard 50° visual field to ultra-widefield (UWF) table-top fundus camera in diabetic retinopathy (DR) detection and grading. Patients affected by diabetes mellitus and referred to our diabetic retinopathy clinic were enroled and underwent fundus photography in mydriasis. All photos were taken using the ultra-widefield table-top fundus camera Zeiss Clarus™ 500 (four fields per eye) and the Optomed Aurora® handheld fundus camera (3 fields per eye). The following parameters were analysed: the gradability of the images, the grade of DR, and diabetic maculopathy (DM), the presence of hypertensive retinopathy (HR) and the presence of other ocular diseases. We enroled 759 eyes of 384 diabetic patients and analysed 5313 fundus photos. The handheld fundus camera obtained a sensitivity of 84.2% and specificity of 95.4% for referable cases. Moreover, it obtained, compared to UWF, an almost perfect agreement with linear weighting for DR, DM and HR (k = 0.877, k = 0.854, and k = 0.961, respectively). The lowest sensitivity was achieved for proliferative DR (58.7% sensitivity, 100% specificity). Optomed Aurora® handheld fundus camera imaging showed a strong agreement compared to UWF in grading DR, considering all DR and DM grades, in mydriasis. However, the use of UWF imaging increases the detection of referable eyes.

Smartphone-enabled retinal arteriovenous imaging and correlation with coronary SYNTAX score

ObjectiveTo assess the feasibility of measurement of retinal arteriovenous (AV) ratio using a smartphone, we performed a comparative evaluation with fundus camera imaging and coronary SYNTAX score. MethodSuccessive coronary artery disease (CAD) patients who underwent coronary angiography were recruited for smartphone retinal imaging. Following pupillary dilatation, fundus camera images and smartphone photography were performed. Video images were captured with a smartphone, edited and analysed. Retinal artery and vein size at 0.5 and 1 disc diameter (DD) were measured using DICOM software by two independent observers. Another observer calculated SYNTAX score. ResultsAnalysable smartphone images were available in 91 (89.2%) of 102 patients. Tobacco use was found in 26%, hypertension in 54%, diabetes in 55%, and high LDL cholesterol in 50%. Median and 25–75 interquartile range (IQR) AV ratio at 0.5 and 1.0 DD, respectively, with smartphone were 0.48 (0.45–0.52) and 0.47 (0.45–0.52) and fundus camera were 0.48 (0.44–0.53) and 0.48 (0.45–0.53) (Spearman's correlation 0.80 and 0.79, p < 0.001). Coronary single vessel disease was in 21%, double vessel in 16%, triple vessel in 55%, normal angiogram in 8%, and median SYNTAX score was 18.0 (8.0–25.0). There was an inverse correlation of SYNTAX score with smartphone-derived AV ratio at 0.5 and 1.0 DD (rho −0.27,p = 0.007 and −0.26,p = 0.009) as well as with fundus camera (rho −0.37 and −0.38, p < 0.001). Trend-analysis showed an inverse association of smartphone AV ratio with increasing CAD (ptrend <0.001). ConclusionsSmartphone-based retinal AV imaging is feasible and comparable to fundus-camera imaging. There is a significant inverse correlation with coronary angiographic severity.

Diabetes Mellitus Associated Progressive Neurovascular Retinal Injury: Recommendations for Imaging and Functional Testing and Potential Role for Early Intervention with Modern Retinal Laser Therapy

Diabetes mellitus is now recognized as a systemic, autoimmune, microvascular disorder, which, in the retina results in severe, multifocal injury now recognized as a leading cause, world-wide, of progressive vision loss. To address this problem in the prediabetic and overt diabetic states, it must be realized that, although the injury processes may be system wide, there are varying responses, effector, and repair mechanisms that differ from organ to organ or within varying cell structures. Specifically, within the “neurovascular unit” of the retina, lesions occur of focal microvascular occlusions, inflammatory endothelial and pericyte injury, and small vessel leakage resulting in injury to astrocytes, Müller cells, microglia, and causing progressive neuronal apoptosis. Such lesions are now recognized to occur before the first microaneurysms are visible by fundus camera imaging or before detectable symptoms or signs recognizable to the patient or clinician. Treatments, therefore which currently are not initiated until edema develops or progressive vascular occlusion, are applied relatively late with some reduction in progressive cellular injury but with minimal vision improvement.

An Evaluation of the Autofluorescence Changes with Wide-Angle Digital Fundus Camera in Chronic Central Serous Chorioretinopathy Patients

To evaluate the autofluorescence variations and localisation in wide-angle digital fundus camera images of patients with chronic central serous chorioretinopathy (CSCR). A retrospective scan was made of the images of patients diagnosed with chronic CSCR with wide-angle digital fundus angiography and applied at the same time with autofluorescence imaging. The ultra wide area autofluorescence images of 46 patients were examined. The retina was separated as zone 1, zone 2, and zone 3 in respect of disease involvement, and zone 3 represented the peripheral retina. The images of the patients were recorded as hyperautofluorescence (punctate, diffuse) or hypo-autofluorescence (granular, confluent) in respect of the type of autofluorescence involvement. After the exclusion of 2 patients because peripheral images could not be clearly selected, the study evaluations were made of 44 eyes of 44 patients. The mean duration of the disease was found to be 2.7 years. In the result of the examination with wide-angle digital fundus autofluorescence imaging, there was seen to be zone 2 involvement in the 4 quadrants of inferior, nasal, temporal and superior. Peripheral retinal involvement in zone 3 was seen in a total of 7 (15.9%) patients in the form of inferior gravitational defect, in 6 (13.6%) patients together with zones 1 and 2, and in 1 (2.3%) patient together with zone 2. Hyperautofluorescence was determined in 39 (88.6%) patients and hypo-autofluorescence in 5 (11.4%). The involvement frequency was determined as zone 1+2 in 19 (43.2%) patients, followed by zone 1 involvement alone in 16 (36.4%) patients. Common autofluorescence type was detected as hyperautofluorescence in chronic CSCR patients with a high rate of approximately 90%. It has also been shown that in some patients, the disease may also affect the inferior peripheral retina due to the effect of gravity.

Prevalence of Ocular Diseases in Human Donor Eyes in New Zealand: A Study Based on Clinical and Histological Imaging.

Fundus pathology in donor eyes was correlated with cross-sectional Optical Coherence Tomography (OCT) images and histological assessment was performed to determine the prevalence of retinal diseases without the constraints imposed during in vivo clinical imaging. A fundus camera and OCT imaging system was adapted to enable posterior segment imaging of the entire post-mortem human eye. Retinas from 59 donors (57 retina pairs and two single globes) were imaged in a seven-field imaging format and cross-sectional analysis was done using OCT. To confirm that the signs observed represented true disease incidence analysis of disease markers including gliosis (Glial Fibrillary Acidic Protein), hemichannel expression (Connexin43), Müller cell activation (vimentin) and choroidal endothelial cells (CD-31) and macrophages (CD-68 marker) was performed. Pathological signs were correlated with clinical diagnoses in eyes from 25 donors (donor ages 45-87 years) but lesions were also found in 23 eyes (donor ages 39- 83 years) with no previously reported clinical diagnosis. Retinas from six donors aged 21-89 years of age were unremarkable. Of all donors, five donors had signs of age related macular degeneration (AMD) and 14 had signs of diabetic retinopathy (DR). Their lesions correlated with OCT and histopathology showed signs of activated microglia, Müller cell hyper-reactivity, increased Cx43 expression and choroidal inflammation. These data indicate that with over 8% of donors showing signs of AMD and 24% of donors showing signs of DR the incidence of AMD may be 1.7 times higher and DR up to 1.6 times higher than clinically reported. The detection of pathological signs characteristic of AMD and DR in donors suggests a higher prevalence of posterior segment abnormalities amongst New Zealanders donors than previously reported. A more detailed evaluation protocol of the posterior segment in patients will aid detection of lesions that are none the less pathological signs.

Handheld fundus camera performance, image quality and outcomes of diabetic retinopathy grading in a pilot screening study

To compare the performance and image quality of the handheld fundus camera to standard table-top fundus cameras in diabetic retinopathy (DR) screening. The reliability and diagnostic accuracy of DR grading performed by an ophthalmologist and a photographer reader were evaluated. 157 patients with diabetes, attending screening or follow-up of DR, were evaluated by fundus photographs taken in mydriasis by Optomed Aurora and Canon or Zeiss Visucam fundus cameras. The image quality and the severity of DR were evaluated independently by an ophthalmologist and experienced photographer. The sensitivity, specificity and reliability of the assessments were determined. 1884 fundus images from 314 eyes were analysed. In 53% of all eyes, DR was not present. 10% had mild non-proliferative diabetic retinopathy (NPDR), 16% moderate NPDR, 6% severe NPDR and 16% proliferative diabetic retinopathy (PDR). The DR grading outcomes by Aurora highly equalled to those of Canon or Zeiss (κ = 0.93, 95% CI 0.91 to 0.94), and there was almost perfect agreement in grading between the ophthalmologist and photographer (κ = 0.96, 95% CI 0.95 to 0.97). The image quality of Aurora was sufficient for reliable assessment according to both graders in 84-88% of the cases. The Optomed Aurora fundus camera seems appropriate for DR screening. The sufficient image quality and high diagnostic accuracy for DR grading are supportive for a less expensive and easily transportable screening system for DR. Immediate image grading carried out by a photographer would further improve and speed up the screening process in all settings.

Diabetic Retinopathy Screening in Patients with Diabetes Using a Handheld Fundus Camera: The Experience from the South-Eastern Region in Hungary.

Purpose Diabetic retinopathy (DR) is the leading cause of vision loss among active adults in industrialized countries. We aimed to investigate the prevalence of diabetes mellitus (DM), DR and its different grades, in patients with DM in the Csongrád County, South-Eastern region, Hungary. Furthermore, we aimed to detect the risk factors for developing DR and the diabetology/ophthalmology screening patterns and frequencies, as well as the effect of socioeconomic status- (SES-) related factors on the health and behavior of DM patients. Methods A cross-sectional study was conducted on adults (>18 years) involving handheld fundus camera screening (Smartscope Pro Optomed, Finland) and image assessment using the Spectra DR software (Health Intelligence, England). Self-completed questionnaires on self-perceived health status (SPHS) and health behavior, as well as visual acuity, HbA1c level, type of DM, and attendance at healthcare services were also recorded. Results 787 participants with fundus camera images and full self-administered questionnaires were included in the study; 46.2% of the images were unassessable. T1D and T2D were present in 13.5% and 86.5% of the participants, respectively. Among the T1D and T2D patients, 25.0% and 33.5% had DR, respectively. The SES showed significant proportion differences in the T1D group. Lower education was associated with a lower DR rate compared to non-DR (7.7% vs. 40.5%), while bad/very bad perceived financial status was associated with significantly higher DR proportion compared to non-DR (63.6% vs. 22.2%). Neither the SPHS nor the health behavior showed a significant relationship with the disease for both DM groups. Mild nonproliferative retinopathy without maculopathy (R1M0) was detected in 6% and 23% of the T1D and T2D patients having DR, respectively; R1 with maculopathy (R1M1) was present in 82% and 66% of the T1D and T2D groups, respectively. Both moderate nonproliferative retinopathy with maculopathy (R2M1) and active proliferative retinopathy with maculopathy (R3M1) were detected in 6% and 7% of the T1D and T2D patients having DR, respectively. The level of HbA1c affected the attendance at the diabetology screening (HbA1c > 7% associated with >50% of all quarter-yearly attendance in DM patients, and with 10% of the diabetology screening nonattendance). Conclusion The prevalence of DM and DR in the studied population in Hungary followed the country trend, with a slightly higher sight-threatening DR than the previously reported national average. SES appears to affect the DR rate, in particular, for T1D. Although DR screening using handheld cameras seems to be simple and dynamic, much training and experience, as well as overcoming the issue of decreased optic clarity is needed to achieve a proper level of image assessability, and in particular, for use in future telemedicine or artificial intelligence screening programs.

On the quantitative effects of compression of retinal fundus images on morphometric vascular measurements in VAMPIRE

Background and Objectives: This paper reports a quantitative analysis of the effects of joint photographic experts group (JPEG) image compression of retinal fundus camera images on automatic vessel segmentation and on morphometric vascular measurements derived from it, including vessel width, tortuosity and fractal dimension.Methods: Measurements are computed with vascular assessment and measurement platform for images of the retina (VAMPIRE), a specialized software application adopted in many international studies on retinal biomarkers. For reproducibility, we use three public archives of fundus images (digital retinal images for vessel extraction (DRIVE), automated retinal image analyzer (ARIA), high-resolution fundus (HRF)). We generate compressed versions of original images in a range of representative levels.Results: We compare the resulting vessel segmentations with ground truth maps and morphological measurements of the vascular network with those obtained from the original (uncompressed) images. We assess the segmentation quality with sensitivity, specificity, accuracy, area under the curve and Dice coefficient. We assess the agreement between VAMPIRE measurements from compressed and uncompressed images with correlation, intra-class correlation and Bland-Altman analysis.Conclusions: Results suggest that VAMPIRE width-related measurements (central retinal artery equivalent (CRAE), central retinal vein equivalent (CRVE), arteriolar-venular width ratio (AVR)), the fractal dimension (FD) and arteriolar tortuosity have excellent agreement with those from the original images, remaining substantially stable even for strong loss of quality (20% of the original), suggesting the suitability of VAMPIRE in association studies with compressed images.

Retinal venular tortuosity and fractal dimension predict incident retinopathy in adults with type 2 diabetes: the Edinburgh Type 2 Diabetes Study

Aims/hypothesisOur aim was to determine whether a range of prespecified retinal vessel traits were associated with incident diabetic retinopathy in adults with type 2 diabetes.MethodsIn the prospective observational cohort Edinburgh Type 2 Diabetes Study of 1066 adults with type 2 diabetes, aged 60–75 years at recruitment, 718 were free from diabetic retinopathy at baseline. Baseline retinal traits including vessel widths, tortuosity (curvature) and fractal dimensions (network complexity), were quantified using fundus camera images and semiautomated software, and analysed using logistic regression for their association with incident diabetic retinopathy over 10 years.ResultsThe incidence of diabetic retinopathy was 11.4% (n = 82) over 10 years. After adjustment for a range of vascular and diabetes-related risk factors, both increased venular tortuosity (OR 1.51; 95% CI 1.15, 1.98; p = 0.003) and decreased fractal dimension (OR 0.75; 95% CI 0.58, 0.96; p = 0.025) were associated with incident retinopathy. There was no evidence of an association with arterial tortuosity, and associations between measurements of vessel widths and retinopathy lost statistical significance after adjustment for diabetes-related factors and vascular disease. Adding venular tortuosity to a model including established risk factors for diabetic retinopathy (HbA1c, BP and kidney function) improved the discriminative ability (C statistic increased from 0.624 to 0.640, p = 0.013), but no such benefit was found with fractal dimension.Conclusions/interpretationIncreased retinal venular tortuosity and decreased fractal dimension are associated with incident diabetic retinopathy, independent of classical risk factors. There is some evidence that venular tortuosity may be a useful biomarker to improve the predictive ability of models based on established retinopathy risk factors, and its inclusion in further risk prediction modelling is warranted.Graphical abstract

Light‐field imaging for glaucoma diagnosis—reproducibility on one patient

PurposeA suspicious optic nerve head appearance with abnormal cupping or increased cup‐to‐disc ratio is considered as a strong risk factor for glaucoma. Typical systems to measure this deformation are optical coherence tomography or scanning laser ophthalmoscopy. We show that the needed depth information for glaucoma diagnosis can be obtained reliably with a commercial fundus camera using an adapted light‐field camera providing a one‐shot 3d retinal image.MethodsWe examined the left eye of a glaucoma suspect (male, 39 years) 11 times in compliance with the Declaration of Helsinki. For each examination, the glaucoma suspect repositioned himself and the device was readjusted. The images were taken with a fundus camera (FF450, Carl Zeiss Meditec AG) combined with a modified light‐field camera (Raytrix R12, Raytrix GmbH). The eye was illuminated with a wavelength of 520 nm (±20 nm). The absolute virtual depth (VD) values were exported from the Raytrix software (RxLive 5.0, Raytrix GmbH) and further processed in Matlab R2018. All data were aligned to each other to measure at similar locations. The maximum VD on a vessel segment above the rim and the minimum above the root of the vessel inside the optic nerve head was determined to calculate the depth difference.ResultsAll images showed a clearly identifiable optic nerve head structure and allowed a successful and reliable determination of the depth information. The average relative depth difference over all examinations was 6.1 ± 0.34VD. The 95% confidence interval was between 5.86 and 6.34VD. Besides the depth values also the lateral values are virtual because of an aperture mismatch between fundus camera and light‐field camera. A successful depth estimations of the Raytrix software require high contrast images with textures. Caused by the objective of the fundus camera image distortions existed.ConclusionsThe presented results from our preliminary system show the potential of the light‐field technique to be used for glaucoma diagnostics and progression analysis based on the optic nerve head in 3D. Reliable and reproducible measurements of this technique can enhance the informative value of the classical fundus photography.

A significant progress of computer aided diagnosis system using automated glaucoma screening

The automatic retinal image examination will be developing and significant screening device for initial recognition of eye diseases. This research proposes a computer aided diagnosis framework for initial recognition of glaucoma through Cup to Disc Ratio (CDR) measurement utilizing 2D fundus images. The system uses computer based analytical methods to procedure the patient data. The Glaucoma is chronic &amp; progressive eye disease which damages optic nerve (ON) caused by improved intraocular pressure (IOP) of eye. The Glaucoma mostly affects on optic disc (OD) by enhancing cup size. It might lead to blindness whether not recognized initially. The glaucoma detection through “Heidelberg Retinal Tomography (HRT) optical Coherence Tomography (OCT)” have been more costly. This system proposes an efficient technique to recognize glaucoma utilizing 2D fundus image. The physical analysis of OD is a normal process utilized for identifying glaucoma. In this manuscript, we suggest automatic OD parameterization method is based on segmented OD and optic cup (OC) region attained from fundus retinal images. To automatically extract OD and optic cup, we used K-means clustering technique, SLIC (Simple linear iterative clustering) method, Gabor filter and thresholding. To reshape the attained disc and cup boundary ellipse fitting (EF) is applied to obtained image. We also propose a novel method which automatically calculates CDR from non-stereographic fundus camera images. The CDR is initial clinical indicator for glaucoma assessment. Also, we have calculated OD and cup area. These features are validated by classifying image either normal or glaucomatous for given patient.

Unsupervised sorting of retinal vessels using locally consistent Gaussian mixtures

Background and Objectives: Retinal blood vessels classification into arterioles and venules is a major task for biomarker identification. Especially, clustering of retinal blood vessels is a challenging task due to factors affecting the images such as contrast variability, non-uniform illumination etc. Hence, a high performance automatic retinal vessel classification system is highly prized. In this paper, we propose a novel unsupervised methodology to classify retinal vessels extracted from fundus camera images into arterioles and venules.Methods: The proposed method utilises the homomorphic filtering (HF) to preprocess the input image for non-uniform illumination and denoising. In the next step, an unsupervised multiscale line operator segmentation technique is used to segment the retinal vasculature before extracting the discriminating features. Finally, the Locally Consistent Gaussian Mixture Model (LCGMM) is utilised for unsupervised sorting of retinal vessels.Results: The performance of the proposed unsupervised method was assessed using three publicly accessible databases: INSPIRE-AVR, VICAVR, and MESSIDOR. The proposed framework achieved 90.14%,90.3% and 93.8% classification rate in zone B for the three datasets respectively.Conclusions: The proposed clustering framework provided high classification rate as compared to conventional Gaussian mixture model using Expectation-Maximisation (GMM-EM) approach, thus have a great capability to enhance computer assisted diagnosis and research in field of biomarker discovery.

Automatic Identification of Referral-Warranted Diabetic Retinopathy Using Deep Learning on Mobile Phone Images.

PurposeTo evaluate the performance of a deep learning algorithm in the detection of referral-warranted diabetic retinopathy (RDR) on low-resolution fundus images acquired with a smartphone and indirect ophthalmoscope lens adapter.MethodsAn automated deep learning algorithm trained on 92,364 traditional fundus camera images was tested on a dataset of smartphone fundus images from 103 eyes acquired from two previously published studies. Images were extracted from live video screenshots from fundus examinations using a commercially available lens adapter and exported as a screenshot from live video clips filmed at 1080p resolution. Each image was graded twice by a board-certified ophthalmologist and compared to the output of the algorithm, which classified each image as having RDR (moderate nonproliferative DR or worse) or no RDR.ResultsIn spite of the presence of multiple artifacts (lens glare, lens particulates/smudging, user hands over the objective lens) and low-resolution images achieved by users of various levels of medical training, the algorithm achieved a 0.89 (95% confidence interval [CI] 0.83–0.95) area under the curve with an 89% sensitivity (95% CI 81%–100%) and 83% specificity (95% CI 77%–89%) for detecting RDR on mobile phone acquired fundus photos.ConclusionsThe fully data-driven artificial intelligence-based grading algorithm herein can be used to screen fundus photos taken from mobile devices and identify with high reliability which cases should be referred to an ophthalmologist for further evaluation and treatment.Translational RelevanceThe implementation of this algorithm on a global basis could drastically reduce the rate of vision loss attributed to DR.

A review of machine learning methods for retinal blood vessel segmentation and artery/vein classification.

The eye affords a unique opportunity to inspect a rich part of the human microvasculature non-invasively via retinal imaging. Retinal blood vessel segmentation and classification are prime steps for the diagnosis and risk assessment of microvascular and systemic diseases. A high volume of techniques based on deep learning have been published in recent years. In this context, we review 158 papers published between 2012 and 2020, focussing on methods based on machine and deep learning (DL) for automatic vessel segmentation and classification for fundus camera images. We divide the methods into various classes by task (segmentation or artery-vein classification), technique (supervised or unsupervised, deep and non-deep learning, hand-crafted methods) and more specific algorithms (e.g. multiscale, morphology). We discuss advantages and limitations, and include tables summarising results at-a-glance. Finally, we attempt to assess the quantitative merit of DL methods in terms of accuracy improvement compared to other methods. The results allow us to offer our views on the outlook for vessel segmentation and classification for fundus camera images.

Artificial Intelligence for Automated Overlay of Fundus Camera and Scanning Laser Ophthalmoscope Images.

PurposeThe purpose of this study was to evaluate the ability to align two types of retinal images taken on different platforms; color fundus (CF) photographs and infrared scanning laser ophthalmoscope (IR SLO) images using mathematical warping and artificial intelligence (AI).MethodsWe collected 109 matched pairs of CF and IR SLO images. An AI algorithm utilizing two separate networks was developed. A style transfer network (STN) was used to segment vessel structures. A registration network was used to align the segmented images to each. Neither network used a ground truth dataset. A conventional image warping algorithm was used as a control. Software displayed image pairs as a 5 × 5 checkerboard grid composed of alternating subimages. This technique permitted vessel alignment determination by human observers and 5 masked graders evaluated alignment by the AI and conventional warping in 25 fields for each image.ResultsOur new AI method was superior to conventional warping at generating vessel alignment as judged by masked human graders (P < 0.0001). The average number of good/excellent matches increased from 90.5% to 94.4% with AI method.ConclusionsAI permitted a more accurate overlay of CF and IR SLO images than conventional mathematical warping. This is a first step toward developing an AI that could allow overlay of all types of fundus images by utilizing vascular landmarks.Translational RelevanceThe ability to align and overlay imaging data from multiple instruments and manufacturers will permit better analysis of this complex data helping understand disease and predict treatment.

Verification of 3D-printed universal smartphone retinal imaging adapter against conventional fundus camera imaging for diabetic retinopathy screening

Introduction: Screening for diabetic retinopathy (DR) is critical in preventing visual loss. However, current tools are expensive, bulky and sensitive, thus limiting screening coverage, especially in developing areas such as the interior of Borneo. Smartphone-assisted devices may provide an alternative and this study seeks to determine the level of agreement between a smartphone retinal imaging adapter (SRIA) against conventional ones.&#x0D; Materials and methods: This was a cross-sectional study with Institutional Review Board approval from the Medical Ethics Board of University of Malaya Medical Centre. A total of 284 eyes from 142 patients included underwent retinal imaging using a conventional fundus camera and the SRIA. The images were graded according to Early Treatment of Diabetic Retinopathy Study (ETDRS) classification. Agreement between both modalities was calculated using Cohen’s Kappa statistics.&#x0D; Results: The Kappa agreement between SRIA and conventional fundus imaging in grading individual ETDRS stages stood at 0.648 (p &lt; 0.001), achieving up to 0.752 (p &lt; 0.001) when differentiating between no DR, non-proliferative DR, and proliferative DR.&#x0D; Conclusion: DR grading SRIA and conventional fundus camera imaging were comparable. SRIA can be useful in eye screenings but still needs improvement.

PREVALENCE AND RISK FACTORS FOR DIABETIC RETINOPATHY IN TURKEY: A SCREENING PROGRAMME USING NON MYDRIATIC CAMERA

Purpose: Early identification of patients at risk for diabetes mellitus (DM) and associated morbidities such as diabetic retinopathy (DR) is essential for effective lifestyle intervention and treatment. Large-scale, cost-effective and minimally invasive screening programs are critical for this purpose. This study evaluated the prevalence and history-based risk factors for DR and whether the non-mydriatic fundus camera is an effective screening method in Turkey at local health centers. Methods: A total of 5182 adults 40 years and older were invited to local health centers in rural Eskisehir, Turkey, for DR risk factor assessment and screening by fundus examination. Fundal images were graded on-site by trained ophthalmology assistants for DR and diabetic macular edema (DME). Patients with DR and DME were referred to the tertiary center for follow-up. Results: The severity of DR was associated with female sex, older age, longer duration of DM, insulin usage, lower body mass index (BMI), lower educational level, higher systolic and diastolic blood pressure and poor control of blood glucose. DR was not associated with alcohol intake or smoking. DME was associated with DM duration and age. Conclusions: Identifying DR prevalence and risk factors are essential for disease control. Non-mydriatic fundus camera imaging proved effective for large-scale DR screening