RESUMO
PURPOSE: Age-related macular degeneration (AMD) is a common threat to vision. While classification of disease stages is critical to understanding disease risk and progression, several systems based on color fundus photographs are known. Most of these require in-depth and time-consuming analysis of fundus images. Herein, we present an automated computer-based classification algorithm. DESIGN: Algorithm development for AMD classification based on a large collection of color fundus images. Validation is performed on a cross-sectional, population-based study. PARTICIPANTS: We included 120 656 manually graded color fundus images from 3654 Age-Related Eye Disease Study (AREDS) participants. AREDS participants were >55 years of age, and non-AMD sight-threatening diseases were excluded at recruitment. In addition, performance of our algorithm was evaluated in 5555 fundus images from the population-based Kooperative Gesundheitsforschung in der Region Augsburg (KORA; Cooperative Health Research in the Region of Augsburg) study. METHODS: We defined 13 classes (9 AREDS steps, 3 late AMD stages, and 1 for ungradable images) and trained several convolution deep learning architectures. An ensemble of network architectures improved prediction accuracy. An independent dataset was used to evaluate the performance of our algorithm in a population-based study. MAIN OUTCOME MEASURES: κ Statistics and accuracy to evaluate the concordance between predicted and expert human grader classification. RESULTS: A network ensemble of 6 different neural net architectures predicted the 13 classes in the AREDS test set with a quadratic weighted κ of 92% (95% confidence interval, 89%-92%) and an overall accuracy of 63.3%. In the independent KORA dataset, images wrongly classified as AMD were mainly the result of a macular reflex observed in young individuals. By restricting the KORA analysis to individuals >55 years of age and prior exclusion of other retinopathies, the weighted and unweighted κ increased to 50% and 63%, respectively. Importantly, the algorithm detected 84.2% of all fundus images with definite signs of early or late AMD. Overall, 94.3% of healthy fundus images were classified correctly. CONCLUSIONS: Our deep learning algoritm revealed a weighted κ outperforming human graders in the AREDS study and is suitable to classify AMD fundus images in other datasets using individuals >55 years of age.
Assuntos
Algoritmos , Aprendizado Profundo , Técnicas de Diagnóstico Oftalmológico , Macula Lutea/patologia , Degeneração Macular/diagnóstico , Idoso , Estudos Transversais , Feminino , Fundo de Olho , Humanos , Masculino , Pessoa de Meia-Idade , Fotografação , Reprodutibilidade dos Testes , Índice de Gravidade de DoençaRESUMO
IMPORTANCE: Age-related macular degeneration (AMD) is a common threat to vision loss in individuals older than 50 years. While neovascular complications in AMD are treatable, there is currently no therapy for geographic atrophy secondary to AMD. Geographic atrophy lesion progression over time shows considerable interindividual variability, but little is known about prognostic factors. OBJECTIVE: To elucidate the contribution of common genetic variants to geographic atrophy lesion growth. DESIGN, SETTING, AND PARTICIPANTS: This pooled analysis combined 4 independent studies: the Fundus Autofluorescence Imaging in Age-Related Macular Degeneration (FAM) study, the Directional Spread in Geographic Atrophy (DSGA) study, the Age-Related Eye Disease Study (AREDS), and the Geographic Atrophy Treatment Evaluation (GATE) study. Each provided data for geographic atrophy lesion growth in specific designs. Patients with geographic atrophy secondary to AMD were recruited to these studies. Genotypes were retrieved through the database of Genotypes and Phenotypes (for AREDS) or generated at the Cologne Center for Genomics (for FAM, DSGA, and GATE). MAIN OUTCOMES: The correlation between square root-transformed geographic atrophy growth rate and 7â¯596â¯219 genetic variants passing quality control was estimated using linear regression. The calculations were adjusted for known factors influencing geographic atrophy growth, such as the presence of bilateral geographic atrophy as well as the number of lesion spots and follow-up times. MAIN OUTCOMES AND MEASURES: Slopes per allele, 95% CIs, and P values of genetic variants correlated with geographic atrophy lesion growth. RESULTS: A total of 935 patients (mean [SD] age, 74.7 [7.8] years; 547 female participants [59.0%]) were included. Two gene loci with conservative genome-wide significance were identified. Each minor allele of the genome-wide associated variants increased the geographic atrophy growth rate by a mean of about 15% or 0.05 mm per year. Gene prioritization within each locus suggests the protein arginine methyltransferase 6 gene (PRMT6; chromosome 1; slope, 0.046 [95% CI, 0.026-0.066]; P = 4.09 × 10-8) and the lanosterol synthase gene (LSS; chromosome 21; slope, 0.105 [95% CI, 0.068-0.143]; P = 4.07 × 10-7) as the most likely progression-associated genes. CONCLUSIONS AND RELEVANCE: These data provide further insight into the genetic architecture of geographic atrophy lesion growth. Geographic atrophy is a clinical outcome with a high medical need for effective therapy. The genes PRMT6 and LSS are promising candidates for future studies aimed at understanding functional aspects of geographic atrophy progression and also for designing novel and targeted treatment options.