DeepSeeNet: A Deep Learning Model for Automated Classification of Patient-based Age-related Macular Degeneration Severity from Color Fundus Photographs

PURPOSE

In assessing the severity of age-related macular degeneration (AMD), the Age-Related Eye Disease Study (AREDS) Simplified Severity Scale predicts the risk of progression to late AMD. However, its manual use requires the time-consuming participation of expert practitioners. Although several automated deep learning systems have been developed for classifying color fundus photographs (CFP) of individual eyes by AREDS severity score, none to date has used a patient-based scoring system that uses images from both eyes to assign a severity score.

DESIGN

DeepSeeNet, a deep learning model, was developed to classify patients automatically by the AREDS Simplified Severity Scale (score 0-5) using bilateral CFP.

PARTICIPANTS

DeepSeeNet was trained on 58 402 and tested on 900 images from the longitudinal follow-up of 4549 participants from AREDS. Gold standard labels were obtained using reading center grades.

METHODS

DeepSeeNet simulates the human grading process by first detecting individual AMD risk factors (drusen size, pigmentary abnormalities) for each eye and then calculating a patient-based AMD severity score using the AREDS Simplified Severity Scale.

MAIN OUTCOME MEASURES

Overall accuracy, specificity, sensitivity, Cohen's kappa, and area under the curve (AUC). The performance of DeepSeeNet was compared with that of retinal specialists.

RESULTS

DeepSeeNet performed better on patient-based classification (accuracy = 0.671; kappa = 0.558) than retinal specialists (accuracy = 0.599; kappa = 0.467) with high AUC in the detection of large drusen (0.94), pigmentary abnormalities (0.93), and late AMD (0.97). DeepSeeNet also outperformed retinal specialists in the detection of large drusen (accuracy 0.742 vs. 0.696; kappa 0.601 vs. 0.517) and pigmentary abnormalities (accuracy 0.890 vs. 0.813; kappa 0.723 vs. 0.535) but showed lower performance in the detection of late AMD (accuracy 0.967 vs. 0.973; kappa 0.663 vs. 0.754).

CONCLUSIONS

By simulating the human grading process, DeepSeeNet demonstrated high accuracy with increased transparency in the automated assignment of individual patients to AMD risk categories based on the AREDS Simplified Severity Scale. These results highlight the potential of deep learning to assist and enhance clinical decision-making in patients with AMD, such as early AMD detection and risk prediction for developing late AMD. DeepSeeNet is publicly available on https://github.com/ncbi-nlp/DeepSeeNet.

Prevalence and Genetic Characteristics of Geographic Atrophy among Elderly Japanese with Age-Related Macular Degeneration

Objective

To investigate the prevalence and genetic characteristics of geographic atrophy (GA) among elderly Japanese with advanced age-related macular degeneration (AMD) in a clinic-based study.

Methods

Two-hundred and ninety consecutive patients with advanced AMD were classified into typical neovascular AMD, polypoidal choroidal vasculopathy (PCV), retinal angiomatous proliferation (RAP) or geographic atrophy (GA). Genetic variants of ARMS2 A69S (rs10490924) and CFH I62V (rs800292) were genotyped using TaqMan Genotyping Assays. The clinical and genetic characteristics were compared between patients with and without GA.

Results

The number of patients diagnosed as having typical neovascular AMD, PCV, RAP and GA were 98 (33.8%), 151 (52.1%), 22 (7.5%) and 19 (6.6%), respectively. Of 19 patients with GA, 13 patients (68.4%) had unilateral GA with exudative AMD in the contralateral eye. Patients with GA were significantly older, with a higher prevalence of reticular pseudodrusen, bilateral involvement of advanced AMD and T-allele frequency of ARMS2 A69S compared with those with typical AMD and PCV; although there were no differences in the genetic and clinical characteristics among patients with GA and RAP.

Conclusions

The prevalence of GA was 6.6% among elderly Japanese with AMD. Patients with GA and RAP exhibited genetic and clinical similarities.

The Dry Form of Age-Related Macular Degeneration (AMD): The Current Concepts of Pathogenesis and Prospects for Treatment

Age-related macular degeneration (AMD) is a disease that causes varying degrees of blindness, which afflicts millions of adults in their later years. Preliminary changes occur during normal aging, but in some individuals the pathology leads to the development of AMD. The pathology seems to be a mixture of biochemical, cellular, and molecular events. Lipofuscinogenesis and early drusen genesis are in the early stages of AMD and their inhibition or reversal would dramatically increase the quality of vision in elderly people. The disease is characterized by abnormal extracellular deposits, known as drusen, which accumulate along the basal surface of the retinal pigmented epithelium RPE. Widespread drusen deposition is associated with retinal pigmented epithelial cell dysfunction and degeneration of the photoreceptors. Recent studies have shown that drusen contain a variety of immunomodulatory molecules, suggesting that the process of drusen formation involves local inflammatory events, including activation of the complement cascade. Molecular pathways involved in the etiology of this disease and the potential prospects of its treatment will be presented on the basis of the results of the current studies.

Macular ganglion cell complex and retinal nerve fiber layer comparison in different stages of age-related macular degeneration

PURPOSE

To employ optical coherence tomography (OCT) to analyze the morphologic changes in the inner retina in different categories of age-related macular degeneration (AMD).

DESIGN

Observational cross-sectional study.

METHODS

Single-center study. Inclusion criteria were age over 50, diagnosis of Age-Related Eye Disease Study (AREDS) category 2 and 3, naïve neovascular AMD, and atrophic AMD. Healthy patients of similar age acted as a control group. Primary outcome measures were the changes in ganglion cell complex (GCC) and retinal nerve fiber layer (RNFL). Secondary outcomes included modifications of rim area and cup-to-disc ratio.

RESULTS

One hundred and thirty eyes of 130 patients were recruited: 26 eyes for AREDS category 2, 26 for AREDS category 3, 26 for neovascular AMD, 26 with atrophic AMD, and 26 controls. Mean peripapillary RNFL thickness was significantly lower in neovascular AMD, compared to controls (P = .004); peripapillary RNFL did not significantly vary among AREDS category 2 and 3 and atrophic AMD groups, compared to controls. Mean GCC thickness was higher in the control group, becoming progressively thinner up to neovascular and atrophic AMD groups (P < .0001). Rim area was significantly thinner in the neovascular AMD group compared with controls (P = .047); cup-to-disc ratio was higher in the neovascular AMD group compared with the control group (P = .047).

CONCLUSIONS

This study demonstrates that eyes with neovascular AMD display reduced RNFL and GCC thickness. RNFL is partially spared in atrophic advanced AMD. The identification of alteration in RNFL and GCC thickness may reveal useful for future therapeutic implications.

Intra and interobserver agreement in the classification of fundus autofluorescence patterns in geographic atrophy secondary to age-related macular degeneration

BACKGROUND

To describe the intra and interobserver agreement in the evaluation of fundus autofluorescence patterns in geographic atrophy secondary to age-related macular degeneration.

METHODS

A consecutive series of patients with geographic atrophy and fundus autofluorescence images of a minimum acceptable quality for grading were included. Four observers with experience in the evaluation of autofluorescence images independently classified the randomly presented series of images on two occasions, at least 1 month apart from each other (intraobserver analysis). The second determination of each observer was used for evaluation of interobserver agreement. The kappa statistic and 95% confidence intervals, together with percentages of agreement, were used to analyze the results.

RESULTS

The final sample included 69 eyes of 49 patients. Intraobserver agreement ranged from substantial to almost perfect (kappa between 0.51 and 0.83), while interobserver results ranged from poor to substantial (corresponding to kappa between 0.30 and 0.62). The use of more simple classifications improved reproducibility.

CONCLUSIONS

Although intraobserver reproducibility was high, interobserver agreement was variable. Clear descriptions and a uniform set of criteria to classify these patients are needed if fundus autofluorescence imaging is used to evaluate patients with geographic atrophy.

Progression of geographic atrophy in age-related macular degeneration imaged with spectral domain optical coherence tomography

PURPOSE

To determine the area and enlargement rate (ER) of geographic atrophy (GA) in patients with age-related macular degeneration (AMD) using the spectral domain optical coherence tomography (SD-OCT) fundus image.

DESIGN

Prospective, longitudinal, natural history study.

PARTICIPANTS

Eighty-six eyes of 64 patients with ≥6 months of follow-up.

METHODS

Patients with GA secondary to AMD were enrolled in this study. Macular scans were performed using the Cirrus SD-OCT (Carl Zeiss Meditec, Dublin, CA). The areas of GA identified on the SD-OCT fundus images were quantified using a digitizing tablet. Reproducibility of these measurements was assessed and the ER of GA was calculated. The usefulness of performing square root transformations of the lesion area measurements was explored.

MAIN OUTCOME MEASURES

Enlargement rate of GA.

RESULTS

At baseline, 27% of eyes had a single area of GA. The mean total area at baseline was 4.59 mm(2) (1.8 disc areas [DA]). The mean follow-up time was 1.24 years. Reproducibility, as assessed with the intraclass correlation coefficient (ICC), was excellent on both the original area scale (ICC = 0.995) and the square root scale (ICC = 0.996). Intergrader differences were not an important source of variability in lesion size measurement (ICC = 0.999, 0.997). On average, the ER of GA per year was 1.2 mm(2) (0.47 DA; range, 0.01-3.62 mm(2)/year). The ER correlated with the initial area of GA (r = 0.45; P<0.001), but there were variable growth rates for any given baseline area. When the square root transformation of the lesion area measurements was used as a measure of lesion size, the ER (0.28 mm/yr) was not correlated with baseline size (r = -0.09; P = 0.40). In this cohort of lesions, no correlation was found between ER and length of follow-up. Square root transformation of the data helped to facilitate sample size estimates for controlled clinical trials involving GA.

CONCLUSIONS

The SD-OCT fundus image can be used to visualize and quantify GA. Advantages of this approach include the convenience and assurance of using a single imaging technique that permits simultaneous visualization of GA along with the loss of photoreceptors and the retinal pigment epithelium that should correlate with the loss of visual function.