|
| |
 |
|
| ORIGINAL ARTICLE |
|
| Year : 2022 | Volume
: 4
| Issue : 1 | Page : 14 |
|
Screening strategy in ocular diabetes with automatic detection system in the Chilean public health system
Fernando Barría von-Bischhosffshausen1, G Francisco Andrighetti2, G Nicole Rivera3, F Vezna Sabando4
1 Director of the Ophthalmology Unit of the Digital Hospital, Chilean Ministry of Health; President of the Chilean Society of Ophthalmology; Ophthalmology Service of the regional clinical hospital of Concepción Chile; Committee for the prevention of blindness of the Pan American Association of Ophthalmology
2 Director of the Ophthalmology Unit of the Digital Hospital, Chilean Ministry of Health; Ophthalmology Service of the Pasteur Clinic
3 Project Manager, Department of Digital Health, Chilean Ministry of Health
4 Head of Digital Unit, Department of Digital Health Undersecretary of Assistance Networks, Chilean Ministry of Health
| Date of Submission | 06-Nov-2021 |
| Date of Acceptance | 10-Jan-2022 |
| Date of Web Publication | 23-Mar-2022 |
Correspondence Address:
Dr. Fernando Barría von-Bischhosffshausen
Calle San Martín 1350 Concepción, Santiago
 Source of Support: None, Conflict of Interest: None
DOI: 10.4103/pajo.pajo_117_21
Summary: Diabetic retinopathy (DR) is the leading cause of blindness among working-age persons in high-income countries. A public system strategy was developed to improve screening, using telemedicine, automatic detection using artificial intelligence (A/I) and medical reporting. In the current work, we evaluated program efficiency.
Material: We conducted a cross-sectional study using information from an institutional database of retinographies submitted to the A/I platform in 2019. With a positive test, a medical report was made using the international scale.
Results: In 2019, 220,994 retinographies were reported, corresponding to 24.0% of diabetic patients. Around half (53.0%) of cases were discarded by A/I, being different in each regional health service. The medical analysis discarded diabetic retinopathy in 30.2% of exams, 11.5% had diabetic retinopathy, including 2.3% with risk of blindness, while 3.7% could not be evaluated.
Discussion: The use of A/I allowed optimizing the medical resources, discarded 53% of cases, which helped in the screening of diabetic retinopathy. Coverage is still insufficient, and detection of macular edema must be improved.
Keywords: Automatic detection, diabetic retinopathy, screening
How to cite this article:
von-Bischhosffshausen FB, Andrighetti G F, Rivera G N, Sabando F V. Screening strategy in ocular diabetes with automatic detection system in the Chilean public health system. Pan Am J Ophthalmol 2022;4:14 |
How to cite this URL:
von-Bischhosffshausen FB, Andrighetti G F, Rivera G N, Sabando F V. Screening strategy in ocular diabetes with automatic detection system in the Chilean public health system. Pan Am J Ophthalmol [serial online] 2022 [cited 2023 Mar 21];4:14. Available from: https://www.thepajo.org/text.asp?2022/4/1/14/340384 |
| Introduction | |  |
There is a worldwide diabetic epidemic. According to the International Diabetes Federation, it is estimated that in 2021, 537 million people had diabetes mellitus (DM), which could increase to 783 million in 2045.[1] Diabetic retinopathy (DR) is a microvascular complication of DM that affects the retina. It is associated with chronic hyperglycemia and the duration of diabetes and is the main cause of blindness among working-age persons in high-income countries.[2],[3] The World Health Organization reports a global prevalence of any retinopathy of 34.6%, estimating that 146 million are affected by retinopathy.[4] In 2010, 0.8 million people were blind and 3.7 million had a visual impairment associated with retinopathy.[5] As has been shown in community studies[4],[6],[7],[8] cases of diabetes, retinopathy and risk of blindness are increasing. A systematic review and meta-analysis of population-based surveys of eye disease from 1980 to 2018 was done[9] were cataract followed by other conditions as DR (0.86 million cases [0.59–1.23]).
Evidence has demonstrated the effectiveness of screening for the early detection of DR and preventing vision loss.[2],[3] DR-related vision loss occurs in the late stages of the disease and is irreversible, thus it is important to screen the population with diabetes and establish timely management, including patient education. The fundus examination, with good sensitivity and specificity,[10] is the method of choice to detect the presence of DR, and although the fundus examination is not a substitute for ophthalmological one, it serves as an initial tool in the detection of DR.[11] To meet the demand for care for the increasing population of persons living with diabetes, it is necessary to implement strategies that allow improved coverage and are supported by the use of technology.
Screening programs in high-income countries provide suggestions for examination frequency and define procedures that facilitate the reduction in vision loss.[12] Developing countries must adapt strategies considering economic limitations and access to equipment while ensuring quality control in program implementation. The ICO guidelines[13] address the needs and requirements for different levels of service on high-resource settings, where the screening and management of DR are based on evidence on clinical trials or in low-intermediate resource settings with consideration for availability and access to care in different settings. Since 2004, Chile guarantees access to treatment for retinopathy.[14] To achieve this objective, it is necessary to have a strategy to detect retinopathy, and in 2013, a telemedicine platform was implemented and implemented a DR detection program using digital photo and telemedicine that was implementing by the Chilean Ministry of Health. The strategy provides diagnostic and therapeutic orientation and seeks to guarantee equity and opportunity of care for communities with access gaps by delivering remote services within the framework of an Integrated National Health Services Network.[15] Since 2018, there is a Digital Hospital,[16] dependent on the Ministry of Health, that developed a health-care model uses available technologies. This web platform is universally accessible and allows you to access a medical consultation or start a treatment online. These efforts have increased the coverage of retinopathy screening, reaching 36.5% of the patient population living with diabetes in 2019.
The development and implementation of artificial intelligence (A/I) in Digital Hospital in 2018 has improved the efficiency of the screening programs, reducing the gap between increasing demand and limited health-care resources.[17],[18] With A/I, normal images are separated from altered ones, the latter being referred to a medical report, optimizing the specialized medical resource. Since 2006, neural network analyses have been developed to recognize photos from algorithms with high precision. The automatic detection system can identify cases of DR that are then referred to the ophthalmologist to classify them and guide their management, reducing specialist workload.[19]
The objective of this work is to describe the telemedicine program of the Digital Hospital of Chile with respect to the DR screening program that uses an automatic detection strategy compared to fundus examination and the distribution of the levels of RD to determine cases with risk of blindness and evaluate barriers to prevent blindness.
| Materials and Methods | | |
We conducted a cross-sectional study of 220,944 fundus photos analyzed by the A/I system in 2019, from 165 establishments, mainly Primary Ophthalmology Care Units, belonging to 27 regional health services in Chile. Another two regional health services have adopted their own strategy and do not participate in the platform. All patients with a diagnosis of diabetes who are controlled in the cardiovascular program under the national ministry of health are included, and there are no exclusion criteria. All patients with diabetes who were referred for DR screening were considered through a platform based on A/I with automated detection of DR (DARTMR, TeleDx, Santiago, Chile)[20],[21],[22] associated with a medical report in case of altered retinographies.
For data analysis, an anonymized database was obtained from a digital platform that complies with cybersecurity regulations of the Ministry of Health on confidentiality and protection of sensitive data. The institutional ethics committee determined that informed consent was not required due to the anonymous nature of study data and the fact that patient care would not be affected. An internal resolution authorizes the use of these data. The study was carried out in accordance with the Declaration of Helsinki.
Retinography was taken by an ophthalmic assistant, according to the EURODIAB.[23] This protocol includes taking a photo centered on the macula and a nasal photo centered on the papilla that allows the emergence of the vessels to be seen. The assistant was free to take more photos if deemed necessary, using a nonmydriatic fundus camera available at the health center. The photos were uploaded to the A/I platform for centralized evaluation.
The platform analyzed the photos as the first line of screening. Photos were considered as a negative examination if DR was not suspected and photos were determined to be a positive examination for DR if there was a suspicion of retinopathy or could not be evaluated. In a negative examination, the patient attended their regular annual checkup and was educated in diabetes care and did not require any additional reporting by an ophthalmologist. In the event of a positive examination, patients were referred for evaluation by an ophthalmologist, hired by the different regional health services, to generate a report that classified retinopathy or defined it as not evaluable. If DR was detected, the patients were referred for a face-to-face appointment that included education on diabetes control. For grading, the International Clinical DR (ICDR) scale[24] was considered. To classify DR severity, the patient's worst eye was considered, a derivable retinopathy being the presence of a nonproliferative DR greater than mild.
In a validation study, the platform was found to have a sensitivity of 94.6% (95% confidence interval [CI]: 90.9%–96.9%) and specificity of 74.3% (95% CI: 73.3%–75%),[25] similar to an internal study carried out by the Ministry of Health.
| Results | | |
The eye fundus among diabetic patients of the national cardiovascular program was analyzed, which is carried out in the public health network from 2011 to 2020 [Figure 1], and the coverage increased until 2019 (36%), although coverage decreased the following year due to the effects of the pandemic. In 2019, there were 920,704 diabetic patients in the Chilean public health system, and 336,140 examinations were performed, achieving coverage of 36.5% of patients. Of these, 220,994 (65.7%) were submitted to the A/I platform to obtain a report, corresponding to 24.0% of diabetic patients. The majority of the patients were female (n = 138,090, 62.5%) and 50.4% were over 65 years of age, with a wide distribution in age and sex [Figure 2]. | Figure 1: Distribution in number of eye funds and percentage of national coverage in Chile according to REM P4, 2011–2020. Source: Data obtained from DIPRECE Chilean Ministry of Health
Click here to view |
 | Figure 2: Distribution by sex and age of the examinations carried out in the population with diabetes in 2019. Source: Data obtained from Digital Hospital, Ministry of Health
Click here to view |
During 2019, there was a monthly increase in submitted examinations until August with a reduction in November [Figure 3]. Considering the coverage of this strategy by the National Health Service, we found that some regional health services (e.g., Reloncaví, Valdivia, and O'Higgins) had >50% coverage of their diabetic population, while other services (e.g., Maule, Ñuble, and Arauco) reached 10% [Figure 4]. | Figure 3: Tele-ophthalmology report with distribution of fundus examinations monthly. Source: Data obtained from Digital Hospital, Ministry of Health
Click here to view |
 | Figure 4: Distribution of regional health services according to fundus coverage rate per 1000 patients under surveillance for the control of diabetes in 2019. Source: Data obtained from Digital Hospital, Ministry of Health
Click here to view |
Of the total number of cases submitted to A/I, 117,273 (53.0%) examinations were discarded as no retinopathy was detected. In the case of the O'Higgins Regional Health Service, the platform discarded 24.6% of retinographies, thus the number of cases that required an ophthalmologist report was greater than regional health services in the rest of the country [Figure 5]. | Figure 5: Percentage of cases without derivable retinopathy reported by A/I by regional health service. Source: Data obtained from Digital Hospital, Ministry of Health
Click here to view |
Of the 103,771 remaining examinations, a medical report was completed for 100,041 (45.3%), leaving 3666 examinations (1.7%) pending and the results of 64 (0.03%) examinations not reported because of a lack of photos. Considering the 100,041 examinations reported, in 66,618 cases (30.2% of the total examinations performed), retinopathy was not detected, in 25,426 (11.5% of the total), various levels of DR were described, and 7,997 examinations (3.6%) could not be evaluated. Of the 25,426 cases where the medical report described a retinopathy, 20,337 (9.4% of total tests performed) had mild or moderate nonproliferative retinopathy, 5089 (2.3% of total) had proliferative (0.4%) or severe nonproliferative (1.9%) retinopathy with risk of blindness that required a referral to tertiary level care, and for 7997 cases (3.7% of the total), the examination could not be evaluated.
The 5089 (2.3%) examinations with severe nonproliferative or proliferative retinopathy with risk of blindness were analyzed by regional health services. The Antofagasta Regional Health Service had 262 patients at risk of blindness (4.4% of the cases evaluated in that service), while the Valparaíso-San Antonio Service had only 6 cases (0.1%) [Figure 6]. | Figure 6: Distribution of health services according to the percentage of cases with nonproliferative severe or proliferative retinopathy. Source: Data obtained from Digital Hospital, Ministry of Health
Click here to view |
| Discussion | | |
The program described in the current work is associated with the public health system in Chile. A nonmydriatic fundus camera is used and appears to achieve good quality retinographies after the initial learning, considering that the telemedicine program started in 2013. Good results have been reported using a wide-field camera that allows evaluating up to 82% of the retinal surface with photos centered on the macula and papilla, however, this technology is not available in Chile.[26],[27] The effectiveness, cost, and benefit of the change and the importance of the visual prognosis associated with peripheral lesions should be evaluated to predict the progression to advanced disease. The use of photos obtained by cell phones for their automatic analysis has also been described, which could be evaluated in specific strategies.[28] Current programs, with two-dimensional (2D) photos, find it difficult to detect macular edema, and it is advisable to integrate the use of optical coherence tomography (OCT) in systems considering the best-corrected vision. In addition, photos do not visualize the peripheral retina, and to avoid dilation, ultra-wide-field images and confocal laser ophthalmoscopy can be evaluated.
In Chile, the Digital Hospital was created with the aim of developing strategies with technological tools to reduce gaps in care for various subspecialties. Since April 2018, the hospital has incorporated the automatic detection of DR with an A/I engine[20],[22] that allowed for the analysis of 65.7% of the examinations carried out in the country. We observed different coverage by regional health services that may reflect various barriers: scheduling, ability to take retinographies, and missed appointments. These barriers should be explored to improve screening coverage, starting with increasing the taking of retinographies and completion of medical reports to timely reach cases at risk of blindness.
The first line of screening is an automatic detection of retinopathy, which a validation study[25] showed had a sensitivity of 94.6% and specificity of 74.3%. One study[ 17] analyzed different systems for automatic detection of DR, based on 24 studies (235,235 cases) that found a combined sensitivity of 91.9% (95% CI: 89.6% to 94.3%) and a specificity of 91.3% (95% CI: 89.0% to 93.5%). These neural networks have been validated to detect DR; in order to improve diagnosis precision, new algorithms are required rather than increasing the sample size.
Of the total cases evaluated in 2019, 53.0% were discarded by A/I and did not require a medical report, which allowed for the optimization of the medical resource by only analyzing the altered reports to classify a suspected case of retinopathy. Another study evaluated the quality of retinography and the efficacy of A/I to rule out DR and suggested that retinographies with poor quality evaluations were taken with portable equipment. Other studies describe that high-resolution photos generate more details of injuries and help with classification,[29] but other reports do not demonstrate this relationship and only raise the threshold beyond which diagnostic precision does not improve.[30],[31],[32] To improve performance, it would be advisable to train algorithms and find a minimum requirement for image resolution and sample size associated with staff training.[33],[34]
Regarding level of damage, a referable case was considered when there was a DR worse than mild according to the ICDR scale[3],[24],[24] and corresponded to level 35 or higher of the early treatment study of DR. Another condition is a macular edema where hard exudate is found within a disc diameter of the macula.[24] The diagnosis is not associated with the criteria used for the diagnosis, considering that the ICDR was developed based on the Early Treatment DR Study (ETDRS), so both scales are related.[35] However, an analysis of the results of the medical report, considering the cases with risk of blindness demonstrated large variability in the different services, which could be associated with a lack of training or local strategies such as one regional health service that immediately refers cases with advanced damage, effectively bypassing the A/I platform. There is also no systematic or continuous monitoring of the graders to detect deficiencies in reports, which must be corrected.
A limitation of automatic detection is that it underestimates mild cases, considering that the cutoff point for a referable case of DR is moderate, but the important thing is to be able to detect 100% of severe cases, which is demonstrated in validation studies.[25] An internal study found 2.3% false negatives associated with mild cases but no serious cases with risk of blindness. A challenge is the diagnosis of macular edema, and although it is related to the severity of DR, it can be found at any stage.[36] Although neural networks have been shown to be useful in DR using photos (2D) of the retina as an examination, OCT is a better examination to detect macular edema.[37] Therefore, it could be recommended to train algorithms to interpret OCT images together with fundus photos to improve results in the screening of DR. Another concern is the 1.7% of pending cases that do not have medical reports that depend on each regional health service in the process of being solved. Finally, it is important to note that two regional health services operating in Chile have their own strategies, and it would be desirable to generate a joint long-term strategy at the national level.
A/I is promising, but the WHO has posed ethical challenges for systems of care, professionals, and beneficiaries of medical services in public health.[38] A/I increases the capacity to improve care with more accurate diagnoses and optimize treatment, as well as supporting the formulation of health policies or the allocation of resources in health systems. This allows countries with scarce resources to bridge the gaps in access to health services, generating more efficient health-care systems. In addition, health workers must guarantee proper use and have mechanisms to audit system performance, as well as promote the interests of patients, implementing ethically based policies and ensuring that their rights will not be subordinated to commercial interests of technology companies or the interests of governments. According to the WHO, there are six ethical principles for the use of A/I: protecting human autonomy, promoting human well-being and safety, ensuring transparency, fostering responsibility and accountability, promoting A/I that is responsive and sustainable, and ensuring inclusion and equity.
Finally, we must consider that DR is a vascular complication of the retina that cannot be reversed or cured, for which we must increase screening coverage considering good practices such as the United Kingdom NHS diabetic eye screening program.[39],[40] These national program stars in 2003 and invited all individuals with diabetes aged 12 years and over for an annual diabetic eye screening appointment. Since 2008, the program has achieved near comprehensive population coverage (>80% annual coverage).[41] The retinal photography is transferred to a grading center and Individuals with sight-threatening diabetic retinopathy are referred for timely ophthalmology assessment and management. A review of the blindness registry revealed that the condition was no longer the most common cause of blindness in the working age population[42] and provides evidence that systematic diabetic retinopathy screening can reduce vision impairment and blindness.
The evaluation of fundus photos to detect changes such as microaneurysms, hemorrhages and exudates, abnormal blood vessels and macular involvement takes a long time but is made more efficient using an A/I system that can detect these changes for the diagnosis of retinopathy.[43] Automated detection of DR with the analysis of digital photos in the background is justified by the reduction of health care costs and product development and there is increased demand for automated detection due to the current diabetes epidemic.[44],[45] However, it is still necessary to develop new, more efficient algorithms to identify retinal changes associated with DR and then to validate these systems to reduce the cost and inconsistencies associated with a manual evaluation of retinographies.
Financial support and sponsorship
Nil.
Conflicts of interest
There are no conflicts of interest.
| References | | |
| 1. | International Diabetes Federation. IDF Diabetes Atlas 10th Edition 2021. Brussels, Belgium: International Diabetes Federation, 2021 Available from: https://diabetesatlas.org/atlas/tenth-edition/. (last accessed on 2022 February 10).  |
| 2. | |
| 3. | |
| 4. | |
| 5. | Leasher JL, Bourne RR, Flaxman SR, Jonas JB, Keeffe J, Naidoo K, et al. Global estimates on the number of people blind or visually impaired by diabetic retinopathy: a meta-analysis from 1990 to 2010. Diabetes Care 2016 Sept 39 1643–1649. ( https://doi.org/10.2337/dc15-2171). |
| 6. | Flaxman SR, Bourne RRA, Resnikoff S, Ackland P, Braithwaite T, Cicinelli MV, et al. Vision Loss Expert Group of the Global Burden of Disease Study. Global causes of blindness and distance vision impairment 1990-2020: a systematic review and meta-analysis. Lancet Glob Health. 2017 Dec;5(12):e1221-e1234. doi: 10.1016/S2214-109X(17)30393-5. Epub 2017 Oct 11. PMID: 29032195 Review. |
| 7. | |
| 8. | |
| 9. | GBD 2019 Blindness and Vision Impairment Collaborators; Vision Loss Expert Group of the Global Burden of Disease Study: Causes of blindness and vision impairment in 2020 and trends over 30 years, and prevalence of avoidable blindness in relation to VISION 2020: the Right to Sight: an analysis for the Global Burden of Disease Study. Lancet Glob Health. 2021 9(2):e144-e160. doi: 10.1016/S2214-109X(20)30489-7. Epub 2020 Dec 1. |
| 10. | Gibbins, RL, Owens, DR, Allen, J.C, Eastman, L. Practical application of European Field Guide in screening for diabetic retinopathy by using ophthalmoscope and 35 mm retinal slides Diabetología 1998 Jan 41(1): 59-64. |
| 11. | Williams GA, Scott IU, Haller JA, Maguire AM, Marcus D, McDonald HR. Single-field fundus photography for diabetic retinopathy screening: a report by the American Academy of Ophthalmology. Ophthalmology. 2004 May;111(5):1055-62. |
| 12. | |
| 13. | Wong TY, Sun J, Kawasaki R, Ruamviboonsuk P, Gupta N, Lansingh VCh, et al. Guidelines on Diabetic Eye Care The International Council of Ophthalmology Recommendations for Screening, Follow-up, Referral, and Treatment Based on Resource Settings Ophthalmology 2018 Oct;125 (10): 1608-1622. doi:10.1016/j.ophtha.2018.04.007. Epub 2018 May 24. |
| 14. | |
| 15. | |
| 16. | |
| 17. | Wang S, Zhang Y, Lei S, Zhu H, Li J, Wang Q, et al. Performance of deep neural network-based artificial intelligence method in diabetic retinopathy screening: a systematic review and meta-analysis of diagnostic test accuracy Eur J Endocrinol. 2020 Jun 1;183(1)41-49. https://doi.org/10.1530/EJE-19-0968. |
| 18. | |
| 19. | |
| 20. | Arenas-Cavalli JT. Automated diabetic retinopathy detection based on remote computational intelligence. IEEE E-Health Tech Comm Newsletter. 2013;2(4): 2–3. |
| 21. | Donoso R, Arenas-Cavalli JT, Pola M. Clinical validation of an automated screening system for diabetic retinopathy in a public health reference center in Chile. XXI Brazilian Congress of Ophthalmology, Blindness Prevention and Visual Rehabilitation; Recife, Brazil, 2014. |
| 22. | Arenas-Cavalli JT, Ríos SA, Pola M, Donoso R. A web-based platform for automated diabetic retinopathy screening. Procedia Comp Science. 2015; 60(1): 557–563. |
| 23. | Aldington S, Kohner E, Meuer S, Klein R, Sjølie A, Group EICS. Methodology for retinal photography and assessment of diabetic retinopathy: the EURODIAB IDDM complications study. Diabetologia. 1995; Apr38(4):437–444. |
| 24. | Wilkinson C, Ferris FL III, Klein RE, Lee PP, Agardh CD, Davis M, et al. Proposed international clinical diabetic retinopathy and diabetic macular edema disease severity scales. Ophthalmology. 2003;110 (9):1677–1682. |
| 25. | Arenas-Cavalli, J.T., Abarca, I., Rojas-Contreras, M. et al. Correction: Clinical validation of an artificial intelligence-based diabetic retinopathy screening tool for a national health system. Eye (lond) 2021. Oct 35 (10): 2910 https://doi.org/10.1038/s41433-021-01690-z. |
| 26. | Wang K, Jayadev C, Nittala M, Velaga SB, Ramachandra CA, Bhaskaranand M, et al. Automated detection of diabetic retinopathy lesions on ultrawidefield pseudocolour images Acta Ophthalmol. 2018 Mar;96(2):e168-e173. |
| 27. | Oh K, Min Kang H, Leem D, Lee H, Yul Seo K y Yoon S; Early detection of diabetic retinopathy based on deep learning and ultra-wide-field fundus images. Sci Rep. 2021 Jan21;11(1):1897. |
| 28. | Hacisoftaoglu RE, Mahmut Karakaya M y Sallam A; Deep Learning Frameworks for Diabetic Retinopathy Detection with Smartphone-based Retinal Imaging Systems Pattern Recognit Lett. 2020 Jul;135:409-417. |
| 29. | Raman B, Bursell ES, Wilson M, Zamora G, Benche I, Nemeth SC & Soliz P. The effects of spatial resolution on an automated diabetic retinopathy screening system's performance in detecting microaneurysms for diabetic retinopathy. Eds Long R, Antani S, Lee DJ, Nutter B & Zhang M, pp. 128–133, 2004. Proceedings 17 th IEEE Symposiumon Computer-Based Medical Systems; year 2004 page 128 DOI Bookmark: 10.1109/CBMS.2004.1311703 https://www.computer.org/csdl/proceedings-article/cbms/2004/21040128/12OmNCykm7s (last accessed on 2022 February 10). |
| 30. | |
| 31. | |
| 32. | Sim DA, Keane PA, Tufail A, Egan CA, Aiello LP, Silva PS. Automated retinal image analysis for diabetic retinopathy in telemedicine. Curr Diabet Rep 2015 1;15(3)14. |
| 33. | Panwar N, Huang P, Lee J, Keane PA, Chuan TS, Richhariya A, et al. Fundus Photography in the 21 st Century – a review of recent technological advances and their implications for worldwide healthcare. Telemed JE Health 2016 Mar22(3) 198–208. |
| 34. | Tajbakhsh N, Shin JY, Gurudu SR, Hurst RT, Kendall CB, Gotway MB, et al. Convolutional neural networks for medical image analysis: full training or fine tuning? IEEE Trans Med Imag 2016 may35(5) 1299–1312. |
| 35. | Wu L, Fernandez-Loaiza P, Sauma J, Hernandez-Bogantes E y Masis M. Classification of diabetic retinopathy and diabetic macular edema. World Jl of Diabetes 2013 DEc4(6) 290–294. |
| 36. | Browning DJ, Fraser CM, Clark S. The relationship of macular thickness to clinically graded diabetic retinopathy severity in eyes without clinically detected diabetic macular edema. Ophthalmology 2008;115:533–9. |
| 37. | Hassan B, Hassan T, Li B, Ahmed R & Hassan O. Deep ensemble learning based objective grading of macular edema by extracting clinically significant findings from fused retinal imaging modalities. Sensors (Basel) 2019 Jul;19(3): 2970. |
| 38. | |
| 39. | |
| 40. | |
| 41. | Scanlon PH. The English National Screening Programme for diabetic retinopathy 2003-2016. Acta diabetologica. 2017;54(6):515–25. |
| 42. | Liew G, Michaelides M, Bunce C. A comparison of the causes of blindness certifications in England and Wales in working age adults (16-64 years), 1999-2000 with 2009-2010. BMJ open. 2014;4(2):e004015. |
| 43. | Amin J, Sharif M y Yasmin M; A Review on Recent Developments for Detection of Diabetic Retinopathy Scientifica (Cairo). Scientifica (Cairo). 2016; Volume 2016, Article ID 6838976, Published online 2016 Sep 29. https://doi.org/10.1155/2016/6838976. |
| 44. | |
| 45. | Barria F, Scanlom P: Personal work: telemedicine in diabetes: how to improve its implementation? Revista de Salud Comunitaria 2017, vol 7 N 18, 14-16. Community Eye Health Journal. Published on 2018, January 22. |
[Figure 1], [Figure 2], [Figure 3], [Figure 4], [Figure 5], [Figure 6], [Figure 1], [Figure 2], [Figure 3], [Figure 4], [Figure 5], [Figure 6]
|
Search Pubmed for