Journal of Ophthalmology

Journal of Ophthalmology / 2019 / Article

Research Article | Open Access

Volume 2019 |Article ID 1691064 | 7 pages | https://doi.org/10.1155/2019/1691064

Prediction of Causative Genes in Inherited Retinal Disorders from Spectral-Domain Optical Coherence Tomography Utilizing Deep Learning Techniques

Academic Editor: Lawrence S. Morse
Received03 Dec 2018
Accepted11 Mar 2019
Published09 Apr 2019

Abstract

Purpose. To illustrate a data-driven deep learning approach to predicting the gene responsible for the inherited retinal disorder (IRD) in macular dystrophy caused by ABCA4 and RP1L1 gene aberration in comparison with retinitis pigmentosa caused by EYS gene aberration and normal subjects. Methods. Seventy-five subjects with IRD or no ocular diseases have been ascertained from the database of Japan Eye Genetics Consortium; 10 ABCA4 retinopathy, 20 RP1L1 retinopathy, 28 EYS retinopathy, and 17 normal patients/subjects. Horizontal/vertical cross-sectional scans of optical coherence tomography (SD-OCT) at the central fovea were cropped/adjusted to a resolution of 400 pixels/inch with a size of 750 × 500 pix2 for learning. Subjects were randomly split following a 3 : 1 ratio into training and test sets. The commercially available learning tool, Medic mind was applied to this four-class classification program. The classification accuracy, sensitivity, and specificity were calculated during the learning process. This process was repeated four times with random assignment to training and test sets to control for selection bias. For each training/testing process, the classification accuracy was calculated per gene category. Results. A total of 178 images from 75 subjects were included in this study. The mean training accuracy was 98.5%, ranging from 90.6 to 100.0. The mean overall test accuracy was 90.9% (82.0–97.6). The mean test accuracy per gene category was 100% for ABCA4, 78.0% for RP1L1, 89.8% for EYS, and 93.4% for Normal. Test accuracy of RP1L1 and EYS was not high relative to the training accuracy which suggests overfitting. Conclusion. This study highlighted a novel application of deep neural networks in the prediction of the causative gene in IRD retinopathies from SD-OCT, with a high prediction accuracy. It is anticipated that deep neural networks will be integrated into general screening to support clinical/genetic diagnosis, as well as enrich the clinical education.

1. Introduction

Inherited retinal disorder (IRD) that has been lacking effective treatment is a leading cause of blindness in developed countries [15], affecting around 1 in 3000 people worldwide. The successful identification of the causative genes for retinal dystrophies with next generation sequencing technologies has increased the number of associated genes and disease-causing variants; to date, over 250 genes and 300 genes and loci have been identified (RetNet: https://sph.uth.edu/retnet/sum-dis.htm). Emerging treatment approaches, such as gene replacement therapy, pharmacological agents, regenerative therapies, retina prosthesis, and others, have increased the therapeutic potential for IRDs [6, 7].

Recently, large cohort studies have provided powerful information to determine clinical manifestations of IRD, such as fundus appearance and morphological findings, and characteristic morphological features caused by specific genes in macular dystrophy are well established [814]. For instance, typical cases with ABCA4 retinopathy show disruption of photoreceptor layers with thinned sensory retina at the macula, and cases with RP1L1 retinopathy demonstrate blurring of photoreceptor ellipsoid zone (EZ) and loss of photoreceptor interdigitation zone (IZ) [8, 11, 1522].

On the contrary, it has been challenging to make a diagnosis at general ophthalmology clinics without IRD specialists since detailed and specific phenotypic assessment is unavailable because of the limited information due to the rarity of IRD. Recently, applications of artificial intelligence (AI) including deep neural network have been increasingly developed for screening/predicting several common retinal diseases from fundus images or spectral-domain optical coherence tomographic (SD-OCT) images [2328]. However, such characterization/classification using automatic image analysis methods has not been developed in orphan retinal diseases.

The purpose of our study was to illustrate a data-driven deep learning approach to predict the causative genes of IRD in macular dystrophy caused by ABCA4 and RP1L1 in comparison with retinitis pigmentosa caused by EYS gene aberration and normal subjects.

2. Materials and Methods

The protocol of this study adhered to the tenets of the Declaration of Helsinki and was approved by the Ethics Committee of the participating institutions: National Institute of Sensory Organs (NISO), National Hospital Organization, and Tokyo Medical Center (Reference number: R15-037). A signed informed consent was obtained from all patients.

For the purpose of this study, two most prevalent genes for macular dystrophies and one most prevalent gene for retinitis pigmentosa were selected from the dataset of Japan Eye Genetics Consortium (JEGC): proportion of ABCA4 retinopathy, RP1L1 retinopathy, and EYS retinopathy were 5.5%, 8.5%, and 15.9%, respectively.

In total, 75 patients/subjects with molecularly confirmed IRD or no ocular diseases have been ascertained from the JEGC database [11]: 10 patients with ABCA4 retinopathy, 20 patients with RP1L1 retinopathy, 28 with EYS retinopathy, and 17 normal subjects. SD-OCT images are obtained with three OCT devices (Cirrus HD-OCT; Carl Zeiss Meditec, Dublin, CA, USA, Spectralis OCT; Heidelberg Engineering, Franklin, MA, USA; Swept Source OCT DRI OCT-1 Atlantis, Topcon Corporation, Tokyo, Japan).

Horizontal and vertical cross-sectional scans of SD-OCT at the central fovea of both eyes were cropped and adjusted to a spatial resolution of 400 pixels/inch with a size of 750 × 500 pix2 for deep learning. Four gene categories were defined based on the clinical and genetic diagnosis; ABCA4, RP1L1, EYS, and Normal (Figure 1). After preparation of SD-OCT images for four gene categories, patients/subjects were randomly split following a 3 : 1 ratio into training and test sets.

The commercially available deep learning web tool, Medic Mind (https://www.medicmind.tech/), was applied to this four-class classification program [29]. The classification accuracy, sensitivity, and specificity were calculated during the learning process. The process was repeated four times with random assignment of subjects to training and test sets to control for selection bias, given the relatively small sample size. For each training/testing process, the classification accuracy was calculated per gene category.

3. Results and Discussion

A total of 178 images from 75 patients/subjects with IRD/Normal were included in this study. The detailed training and test results of deep learning performance in prediction of causative genes in IRD are presented in Table 1.


Experiment 1.Training resultsTest 1.Test resultsTotal number of images included in this study
Number of imagesSensitivity (%)Specificity (%)Accuracy (%)Original category of genetic diagnosisAccuracy (%)
ABCA4RP1L1EYSNormalTotal

Original classification of genetic diagnosisABCA415100100Predicted category of genetic diagnosisABCA4415100ABCA419
RP1L12985.7100RP1L17787.5RP1L137
EYS4310095EYS1414100EYS57
Normal49100100Normal1616100Normal65
Total13696.9Total4814164297.6Total178

Experiment 2.Training resultsTest 2.Test resultsTotal number of images included in this study
Number of imagesSensitivity (%)Specificity (%)Accuracy (%)Original category of genetic diagnosisAccuracy (%)
ABCA4RP1L1EYSNormalTotal

Original classification of genetic diagnosisABCA413100100Predicted category of genetic diagnosisABCA46219100ABCA419
RP1L126100100RP1L19121282RP1L137
EYS43100100EYS1231585.7EYS57
Normal46100100Normal——141473.7Normal65
Total128100Total61114195082Total178

Experiment 3.Training resultsTest 3.Test resultsTotal number of images included in this study
Number of imagesSensitivity (%)Specificity (%)Accuracy (%)Original category of genetic diagnosisAccuracy (%)
ABCA4RP1L1EYSNormalTotal

Original classification of genetic diagnosisABCA41510096.6Predicted category of genetic diagnosisABCA4415100ABCA419
RP1L13166.7100RP1L14466.7RP1L137
EYS4590.990.5EYS2101290.9EYS57
Normal49100100Normal1616100Normal65
Total14090.6Total4611163791.9Total178

Experiment 4.Training resultsTest 4.Test resultsAccuracy (%)Total number of images included in this study
Number of imagesSensitivity (%)Specificity (%)Accuracy (%)Original category of genetic diagnosis
ABCA4RP1L1EYSNormalTotal

Original classification of genetic diagnosisABCA414100100Predicted category of genetic diagnosisABCA455100ABCA419
RP1L125100100RP1L11011176.9RP1L137
EYS40100100EYS141482.4EYS57
Normal51100100Normal221418100Normal65
Total130100Total51317144989.8Total178

In total, 75 subjects with molecularly confirmed inherited retinal disorders or no ocular diseases have been ascertained: 10 with ABCA4 retinopathy, 20 patients with RP1L1 retinopathy, 28 with EYS retinopathy, and 17 normal subjects. After preparation of spectral-domain optical coherence tomographic (SD-OCT) images for four gene categories, subjects were randomly split following a 3 : 1 ratio into training and test sets. The commercially available deep learning web tool, Medic Mind, was applied to this four-class classification problem. The classification accuracy, sensitivity, and specificity were calculated during the learning process, and the process was repeated four times with randomly assigned training/test sets to control for selection bias. For each training/testing process, the classification accuracy was calculated per gene category.

The mean training accuracy of the four repeated experiments was 96.9% and ranged from 90.6% to 100.0%. The mean sensitivity was 100% for ABCA4, 88.1 (66.7–100%) for RP1L1, 97.7 % (90.9–100%) % for EYS, and 100% for Normal. Sufficient training accuracy over 66.7% was obtained. The mean test accuracy of the four repeated experiments was 90.3%, ranging from 82.0% to 97.6%. The test accuracy per gene category was 100% for ABCA4 and ranged from 66.7 to 87.5% for RP1L1, 82.4 to 100% for EYS, and 73.7 to 100% for Normal according to the four repeated experiments.

The test accuracy of ABCA4 and Normal was considerably higher than that of RP1L1 and EYS, which suggests overfitting. Four (10.5%) out of 38 with an original diagnosis of RP1L1 were classified as EYS, two (5.3%) with RP1L1 were classified as Normal, and one (2.6%) with original RP1L1 was classified as ABCA4 (Figure 2). Two (3.6%) out of 56 with an original diagnosis of EYS were classified as ABCA4, two (3.6%) with original EYS were classified as RP1L1, and two (3.6%) with EYS were classified as Normal.

The potential efficacy of the automatic screening/diagnostic system of IRD from SD-OCT has been illustrated. To the best of our knowledge, this is the first report of utilizing deep learning technology in retinal orphan disorders.

Since IRD exhibits strong gene-characteristic morphological features which are less influenced by environmental factors, these present an ideal application of AI to assist medical diagnosis. Furthermore, in general, deducing IRD is not hard at general ophthalmology clinics, as the family history, specific chief complains, and symmetric findings are unique; complex diagnosis, therefore, with such demographic information and AI-guided assessment for SD-OCT could assist properly identifying patients with IRD.

Application of AI in common diseases such as diabetic retinopathy and age-related macular degeneration manifests huge impacts/effects on improvement of medical care and saving the cost. This study further supports the extended utility of AI in relatively common IRD (covering 30% of total IRD), which has a huge impact not only on ocular diseases but also on other orphan disorders.

Interestingly, the predicted classification of ABCA4 and classification of Normal were highly accurate in this study, while there were still some difficulties in classifying RP1L1 and EYS retinopathies. This may be because the characteristic features of RP1L1 retinopathy with blurring of EZ and loss of IZ are hard to be detected from image analysis, and the disrupted zone of some EYS retinopathy might lie outside of the observed field in SD-OCT images.

There are several limitations in this study. The cohort size of this study was relatively small both in control and affected groups, and target ethnicity was only Japanese, in which larger cohort studies with data from various ethnicities could expand the potential utility of our approach. In our study, comparison analysis of the variation in diagnoses by human and AI prediction was not performed due to the limited data resources, which could delineate the usefulness and weakness of our approach. There are limitations to AI classifiers which are trained to only distinguish between a finite number of predefined classes for which we have sufficient training data. However, as we know, there are over 300 genes that can cause IRD; therefore, the four-class classifier we have developed here cannot be usefully applied in practice to predict a gene from an SD-OCT of a patient with a novel gene diagnosis. It would however be expected that our algorithm would return probabilities close to 25% (1/4) if presented with an SD-OCT which does not fit any of the four classes. Instead, another more clinically applicable approach may be to train several one-or-rest classifiers to distinguish one gene from all others. These could then be applied successively to an SD-OCT to assess which gene diagnosis is the best fitting. However, this would require training several classifiers and would require a sufficient number of cases for each gene classifier, which might not be possible for the rarer IRDs. There is also a current interpretability limitation to AI classifiers which use deep learning. This field remains an active research area, and we intend to explore approaches such as saliency and occlusion maps to highlight what parts of an SD-OCT are important to the classification decision.

Our approach could be extended to other retinopathy-associated genes/loci, of which they are over 300. The challenge lies in obtaining sufficiently large training datasets for the rarest forms of IRD and in distinguishing similar morphological changes within the same mechanism group/cascade (e.g., generalized rod dysfunction, generalized cone dysfunction, and confined macular dysfunction). Further comprehensive studies in larger cohorts, in combination with other phenotypic modalities such as fundus appearance, fluorescein angiography, visual fields, and electrophysiological findings, and imaging performed at different time points in the progression of disease should improve the performance of these prediction algorithms.

4. Conclusions

This study highlighted a novel application of deep neural networks in the prediction of the major causative genes (30%) in IRD retinopathies from SD-OCT, with a mean prediction accuracy of 90%. It is anticipated that deep neural networks will be integrated into general screening to support clinical diagnosis, suggest a causative gene to guide genetic screening, as well as enrich clinical education of orphan retinal disease.

Data Availability

Raw data were generated in the Japan Eye Genetics Consortium database and obtained with permission of the Japan Eye Genetics Consortium at https://niso.kankakuki.go.jp/opkarte/. Derived data supporting the findings of this study are available from the corresponding author on request.

Disclosure

Yu Fujinami-Yokokawa and Nikolas Pontikos are the co-first authors. The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Conflicts of Interest

All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Individual investigators who participated in the sponsored project(s) are not directly compensated by the sponsor but may receive salary or other support from the institution to support their effort on the project(s). Kaoru Fujinami is a paid consultant of Astellas Pharma Inc, Kubota Pharmaceutical Holdings Co., Ltd, Acucela Inc., and Novartis AG. Kaoru Fujinami reports personal fees from Astellas Pharma Inc, Kubota Pharmaceutical Holdings Co., Ltd., Acucela Inc., SANTEN Company Limited, Foundation Fighting Blindness, Foundation Fighting Blindness Clinical Research Institute, Japanese Ophthalmology Society, and Japan Retinitis Pigmentosa Society. Laboratory of Visual Physiology, Division for Vision Research, National Institute of Sensory Organs, National Hospital Organization, Tokyo Medical Center, Tokyo, is supported by grants from Astellas Pharma Inc (NCT03281005), outside the submitted work.

Authors’ Contributions

Yu Fujinami-Yokokawa and Nikolas Pontikos contributed equally to this work.

Acknowledgments

Yu Fujinami-Yokokawa was supported by grants from Grant-in-Aid for Young Scientists of the Ministry of Education, Culture, Sports, Science and Technology, Japan (18K16943). Nikolas Pontikos was funded by the NIHR Moorfields Biomedical Research Center. Kaoru Fujinami was supported by grants from Grant-in-Aid for Young Scientists (A) of the Ministry of Education, Culture, Sports, Science and Technology, Japan (16H06269), Grant-in-Aid for Scientists to support international collaborative studies of the Ministry of Education, Culture, Sports, Science and Technology, Japan (16KK01930002), National Hospital Organization Network Research Fund (H30-NHO-2-12), Foundation Fighting Blindness ALAN LATIES CAREER DEVELOPMENT PROGRAM (CF-CL-0416-0696-UCL), Health Labour Sciences Research Grant, the Ministry of Health Labour and Welfare. (201711107A), and Great Britain Sasakawa Foundation Butterfield Awards. We are grateful to Dr. Xiao Liu and Dr. Gavin Arno, National Institute of Sensory Organs, National Tokyo Medical Center, Japan, for their help in clinical and genetic data analysis. We thank Professor Andrew Webster and Professor Michel Michaelides, UCL Institute of Ophthalmology associated with Moorfields Eye Hospital, UK, and Professor Yozo Miyake, Aichi Medical University, Japan, for insightful comments. We also thank all the collaborators of the Japan Eye Genetics Consortium (http://www.jegc.org/) and the East Asia Inherited Retinal Disease Society (https://www.fujinamik.com/east-asia-inherited-retinal-disease) for data collection. We thank Medic Mind for supporting the analytic process of this study.

References

  1. N. Kumaran, M. E. Pennesi, P. Yang et al., “Leber congenital amaurosis/early-onset severe retinal dystrophy overview,” in GeneReviews((R)), M. P. Adam et al., Ed., Seattle, WA, USA, 2018. View at: Google Scholar
  2. N. Kumaran, A. T. Moore, R. G. Weleber, and M. Michaelides, “Leber congenital amaurosis/early-onset severe retinal dystrophy: clinical features, molecular genetics and therapeutic interventions,” British Journal of Ophthalmology, vol. 101, no. 9, pp. 1147–1154, 2017. View at: Publisher Site | Google Scholar
  3. N. Hirji, J. Aboshiha, M. Georgiou, J. Bainbridge, and M. Michaelides, “Achromatopsia: clinical features, molecular genetics, animal models and therapeutic options,” Ophthalmic Genetics, vol. 39, no. 2, pp. 149–157, 2018. View at: Publisher Site | Google Scholar
  4. L. Sheck, W. I. L. Davies, P. Moradi et al., “Leber congenital amaurosis associated with mutations in CEP290 , clinical phenotype, and natural history in preparation for trials of novel therapies,” Ophthalmology, vol. 125, no. 6, pp. 894–903, 2018. View at: Publisher Site | Google Scholar
  5. J. J. L. Tee, A. J. Smith, A. J. Hardcastle, and M. Michaelides, “RPGR-associated retinopathy: clinical features, molecular genetics, animal models and therapeutic options,” British Journal of Ophthalmology, vol. 100, no. 8, pp. 1022–1027, 2016. View at: Publisher Site | Google Scholar
  6. J. Smith, D. Ward, M. Michaelides, A. T. Moore, and S. Simpson, “New and emerging technologies for the treatment of inherited retinal diseases: a horizon scanning review,” Eye, vol. 29, no. 9, pp. 1131–1140, 2015. View at: Publisher Site | Google Scholar
  7. H. P. N. Scholl, R. W. Strauss, M. S. Singh et al., “Emerging therapies for inherited retinal degeneration,” Science Translational Medicine, vol. 8, no. 368, p. 368rv6, 2016. View at: Publisher Site | Google Scholar
  8. K. Fujinami, K. Tsunoda, G. Hanazono et al., “Fundus autofluorescence in autosomal dominant occult macular dystrophy,” Archives of Ophthalmology, vol. 129, no. 5, pp. 597–602, 2011. View at: Publisher Site | Google Scholar
  9. K. Fujinami, N. Lois, A. E. Davidson et al., “A longitudinal study of stargardt disease: clinical and electrophysiologic assessment, progression, and genotype correlations,” American Journal of Ophthalmology, vol. 155, no. 6, pp. 1075–1088, 2013. View at: Publisher Site | Google Scholar
  10. K. Fujinami, J. Zernant, R. K. Chana et al., “Clinical and molecular characteristics of childhood-onset Stargardt disease,” Ophthalmology, vol. 122, no. 2, pp. 326–334, 2015. View at: Publisher Site | Google Scholar
  11. K. Fujinami, S. Kameya, S. Kikuchi et al., “NovelRP1L1Variants and genotype-photoreceptor microstructural phenotype Associations in cohort of Japanese patients with occult macular dystrophy,” Investigative Opthalmology and Visual Science, vol. 57, no. 11, pp. 4837–46, 2016. View at: Publisher Site | Google Scholar
  12. K. Fujinami, N. Lois, R. Mukherjee et al., “A longitudinal study of Stargardt disease: quantitative assessment of fundus autofluorescence, progression, and genotype correlations,” Investigative Opthalmology and Visual Science, vol. 54, no. 13, pp. 8181–90, 2013. View at: Publisher Site | Google Scholar
  13. X. Kong, R. W Strauss, M Michaelides et al., “Visual acuity loss and associated risk factors in the retrospective progression of stargardt disease study (ProgStar report No. 2),” Ophthalmology, vol. 123, no. 9, pp. 1887–97, 2016. View at: Publisher Site | Google Scholar
  14. R. W. Strauss, A. Ho, B. Muñoz et al., “The natural history of the progression of atrophy secondary to stargardt disease (ProgStar) studies,” Ophthalmology, vol. 123, no. 4, pp. 817–828, 2016. View at: Publisher Site | Google Scholar
  15. K. Fujinami, P. I. Sergouniotis, A. E. Davidson et al., “The clinical effect of homozygous ABCA4 alleles in 18 patients,” Ophthalmology, vol. 120, no. 11, pp. 2324–2331, 2013. View at: Publisher Site | Google Scholar
  16. K. Fujinami, P. I. Sergouniotis, A. E. Davidson et al., “Clinical and molecular analysis of Stargardt disease with preserved foveal structure and function,” American Journal of Ophthalmology, vol. 156, no. 3, pp. 487–501, 2013. View at: Publisher Site | Google Scholar
  17. D. Zobor, G. Zobor, S. Hipp et al., “Phenotype variations caused by mutations in the RP1L1 gene in a large mainly German cohort,” Investigative Opthalmology and Visual Science, vol. 59, no. 7, pp. 3041–3052, 2018. View at: Publisher Site | Google Scholar
  18. S. Kikuchi, S. Kameya, K. Gocho et al., “Cone dystrophy in patient with homozygous RP1L1 mutation,” BioMed Research International, vol. 2015, p. 545243, 2015. View at: Publisher Site | Google Scholar
  19. A. E. Davidson, P. I. Sergouniotis, D. S. Mackay et al., “RP1L1Variants are associated with a spectrum of inherited retinal diseases including retinitis pigmentosa and occult macular dystrophy,” Human Mutation, vol. 34, no. 3, pp. 506–514, 2013. View at: Publisher Site | Google Scholar
  20. S. J. Ahn, J. Ahn, K. H. Park, and S. J. Woo, “Multimodal imaging of occult macular dystrophy,” JAMA Ophthalmology, vol. 131, no. 7, pp. 880–90, 2013. View at: Publisher Site | Google Scholar
  21. S. J. Ahn, S. I. Cho, J. Ahn, S. S. Park, K. H. Park, and S. J. Woo, “Clinical and genetic characteristics of Korean occult macular dystrophy patients,” Investigative Opthalmology & Visual Science, vol. 54, no. 7, pp. 4856–63, 2013. View at: Publisher Site | Google Scholar
  22. S. J. Park, S. J. Woo, K. H. Park, J.-M. Hwang, and H. Chung, “Morphologic photoreceptor abnormality in occult macular dystrophy on spectral-domain optical coherence tomography,” Investigative Opthalmology and Visual Science, vol. 51, no. 7, pp. 3673–9, 2010. View at: Publisher Site | Google Scholar
  23. J. De Fauw, J. R. Ledsam, B. Romera-Paredes et al., “Clinically applicable deep learning for diagnosis and referral in retinal disease,” Nature Medicine, vol. 24, no. 9, pp. 1342–1350, 2018. View at: Publisher Site | Google Scholar
  24. N. Ramachandran, S. C. Hong, M. J. Sime, and G. A. Wilson, “Diabetic retinopathy screening using deep neural network,” Clinical & Experimental Ophthalmology, vol. 46, no. 4, pp. 412–416, 2018. View at: Publisher Site | Google Scholar
  25. U. Schmidt-Erfurth, A. Sadeghipour, B. S. Gerendas et al., “Artificial intelligence in retina,” Progress in Retinal and Eye Research, vol. 67, pp. 1–29, 2018. View at: Publisher Site | Google Scholar
  26. D. S. W. Ting, L. R. Pasquale, L. Peng et al., “Artificial intelligence and deep learning in ophthalmology,” British Journal of Ophthalmology, vol. 103, no. 2, pp. 167–175, 2019. View at: Publisher Site | Google Scholar
  27. T. K. Redd, J. P Campbell, J. M Brown et al., “Evaluation of a deep learning image assessment system for detecting severe retinopathy of prematurity,” British Journal of Ophthalmology, 2018. View at: Publisher Site | Google Scholar
  28. M. Treder, J. L. Lauermann, and N. Eter, “Deep learning-based detection and classification of geographic atrophy using a deep convolutional neural network classifier,” Graefe’s Archive for Clinical and Experimental Ophthalmology, vol. 256, no. 11, pp. 2053–2060, 2018. View at: Publisher Site | Google Scholar
  29. C. H. Stevenson, S. C. Hong, and K. C. Ogbuehi, “Development of an artificial intelligence system to classify pathology and clinical features on retinal fundus images,” Clinical and Experimental Ophthalmology, 2018. View at: Publisher Site | Google Scholar

Copyright © 2019 Yu Fujinami-Yokokawa et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

664 Views | 244 Downloads | 0 Citations
 PDF  Download Citation  Citation
 Download other formatsMore
 Order printed copiesOrder
 Sign up for content alertsSign up