Diabetic retinopathy retinal imaging -
International Journal of Retina and Vitreous volume 9 , Article number: 41 Cite this article. Metrics details. Diabetic retinopathy DR is a leading cause of blindness. Our objective was to evaluate the performance of an artificial intelligence AI system integrated into a handheld smartphone-based retinal camera for DR screening using a single retinal image per eye.
Images were obtained from individuals with diabetes during a mass screening program for DR in Blumenau, Southern Brazil, conducted by trained operators. The results were compared to the assessment by a retinal specialist, considered as the ground truth, using two images per eye.
Patients with ungradable images were excluded from the analysis. A total of individuals average age The rates of insulin use, daily glycemic monitoring, and systemic hypertension treatment were Although The majority Approximately DR classification based on the ground truth was as follows: absent or nonproliferative mild DR The area under the ROC curve was The portable retinal camera combined with AI demonstrated high sensitivity for DR screening using only one image per eye, offering a simpler protocol compared to the traditional approach of two images per eye.
Simplifying the DR screening process could enhance adherence rates and overall program coverage. The screening of diabetic retinopathy DR is a milestone for the prevention of blindness and is recommended by many countries as well as the World Health Organization [ 1 ]. Successful screening strategies worldwide are usually based on color fundus photographs CFPs , such as the English program [ 1 ].
However, blindness secondary to diabetes is still an unmet need in most low- and middle-income countries [ 2 ] and also in some high-income countries: in the USA, rates of screening as low as have been reported [ 3 ]. Solutions for increasing screening rates include public health policies, health education [ 2 ] and technological breakthroughs which may render the process simpler and more cost-effective.
In that sense, the incorporation of telemedicine protocols, handheld devices, and artificial intelligence AI have all shown to increase the efficiency of screening [ 4 ]. Recently, autonomous AI systems have been granted regulatory approval for the detection of DR based on the analysis of two retinal images per eye [ 5 , 6 ].
The imaging protocol for DR screening has gone through an evolution over the last decades, from the original ETDRS protocol of 7 fields until the widely accepted protocol of two retinal images per eye [ 7 ]. Simpler protocols have been associated with increased adherence, ultimately contributing to a program´s efficiency [ 7 ].
The challenge is to balance a simpler protocol without losing image quality and diagnostic accuracy. A protocol based on a single image per eye may save significant examination time in high-burden settings, such as mass screening campaigns, where more than one thousand people are screened for DR in a single morning.
Such protocol may also be suitable for a staged mydriasis strategy: due to pupillary reflex secondary to the camera flash, the second image is harder to obtain without pharmacological mydriasis. In that sense, the ungradable rate is expected to be higher with two photos. Our objective was to evaluate the performance of a DR screening protocol that employed a single retinal photo per eye, obtained with a handheld retinal camera and evaluated by an embedded AI system.
This retrospective study enrolled a convenience sample of individuals aged over 18 years old with a previous type 1 or type 2 diabetes mellitus DM diagnosis who were summoned to attend the Blumenau Diabetes Campaign, a DR screening strategy that occurred from February to November at the city of Blumenau, Southern Brazil.
The study protocol was approved by the ethics Committee of Fundação Universidade Regional de Blumenau After signing informed consent, participants answered a questionnaire for demographic and self-reported clinical characteristics: age, gender, income, profession, educational level, type of diabetes, and diabetes duration.
After answering the questionnaire, patients underwent ocular imaging. Imaging acquisition protocol and expert reading are detailed elsewhere [ 8 ].
Image acquisition was performed by a team of previously trained medical students, at public primary care health units. Human image reading was performed in a store-and-forward fashion at EyerCloud platform Phelcom Technologies LLC, Boston, MA by a single retinal specialist FMP after anonymization and quality evaluation.
This ground truth analysis by a human grader was performed using two images per eye. Classification of DR was given per individual, considering the most affected eye, according to the International Council of Ophthalmology Diabetic Retinopathy ICDR classification.
Patients with ungradable fundus images had their anterior segment evaluated for cataracts or other media opacities. No information other than ocular images was available for the reader, and the human grader was masked to the automated evaluation described below.
Images corresponding to one macula-centered image of each eye, were graded by an AI system trained with the Kaggle Diabetic Retinopathy dataset EyePACS and transfer learning with a dataset of approximately 16, fundus images captured using Phelcom Eyer.
The system was previously validated for the detection of more than mild DR mtmDR , details of which have already been described by our group elsewhere [ 8 ]. Only individuals who had images with enough quality were included in the analysis. Softmax normalized the respective neuron input values, creating a probabilistic distribution in which the sum will be 1; the prediction corresponding to the interval between 0 and 1, indicating the likelihood of DR.
To visualize the location of the most important regions obtained by CNN, to discriminate between classes, the Gradient-Based Class Activation Map GradCam was used; it generates a heat map EyerMaps, Phelcom Technologies LLC, Boston, MA which highlights the detected changes Fig.
Example of heatmap visualization. A, C and E Color fundus photograph depicting clinical signs of diabetic retinopathy such as hard exudates and hemorrhages. B, D and F Overlay with the heatmap visualization can help identify lesions, flagged in a color scale, from blue low importance to red high importance.
Data were collected in MS Excel files Microsoft Corporation, Redmond, WA, USA. Statistical analyses were performed using SPSS The 0. Diagnostic accuracy is reported according to the Standards for Reporting of Diagnostic Accuracy Studies STARD [ 9 ]. The remaining individuals average age Diabetes duration was Rates of insulin use, daily glycemic monitoring and treatment for systemic hypertension were Even though Individuals who were illiterate or who had not completed elementary school were DR classification according to the ground truth was as follows Table 1 : absent Figure 2 depicts the Standards for Reporting of Diagnostic Accuracy Studies STARD diagram for the algorithm mtmDR output.
PPV and NPV for mtmDR were Area under the receiver operating characteristic ROC curve was 0. Standards for Reporting of Diagnostic Accuracy Studies STARD diagram for the algorithm mtmDR output. We herein report the results of automatic analysis for the detection of mtmDR with a single retinal image, obtained with a portable smartphone-based retinal camera.
The high sensitivity of the embedded AI system in our real-world sample compares well with previous reports of other automated systems [ 5 , 10 , 11 ]. The first two AI systems approved by the FDA for DR screening rely on protocols of two retinal images per eye and use traditional, tabletop retinal cameras: Idx DR [ 5 ] and EyeArt [ 6 ], and a recent study that validated seven AI systems for DR screening in the real world based on protocols of two retinal images per eye found sensitivities ranging from Portable handheld, low-cost retinal cameras have the potential to broaden the reach of DR screening programs, widening geographic areas and reaching populations that otherwise would not be screened by traditional methods [ 12 ]; such aspects potentially increase program´s efficiency due to higher coverage and increased adherence.
A handheld device with integrated AI analysis has been reported in a screening performance with four fundus images per eye [ 13 ] with sensitivity of Another recent study with a handheld device and the same AI system, but a protocol of five fundus images per eye, reported a sensitivity of We have studied the performance of AI on a protocol based on a single fundus image per eye.
It has been established that, regarding expert human reading, a single image protocol loses diagnostic accuracy in comparison to a two-images protocol [ 15 ]. However, with automatic reading, performance was considered satisfactory for screening, with the obvious advantages of obtaining one single image per eye; efforts to facilitate the process and make it less time-consuming are warranted to increase efficiency.
Interestingly, macula-centered images have been considered to correspond to the most important region for deep learning systems in the evaluation of DR [ 16 ]. We have attained comparable diagnostic accuracy in comparison with the results reported by Nunez do Rio and colleagues [ 17 ]: their performance of a Deep Learning algorithm for the detection of referable DR analyzing only one retinal image per eye was as follows: sensitivity of The performance of our strategy also compares well with results from trained human readers when analyzing one image per eye [ 18 ].
Comparing to other algorithms, it has a relatively low specificity This approach helps to minimize costs while improving the specificity of the method through evaluation by specialists for those patients who truly require it. In this same strategy, Xie demonstrated that assistive and non-autonomous systems exhibit greater cost-effectiveness when compared to purely autonomous systems [ 19 ].
Improving the algorithm technology may increase this specificity without losing its main characteristic of a high sensitivity method for mass screening programs.
Another important aspect to discuss is the pupil status for retinal imaging, this might affect the number of ungradable images and AI performance.
Piyaseana and cols reported that the proportion of ungradable images in non-mydriatic settings was The present study was conducted in a mass screening program and pupil dilation was performed to ensure a faster imaging acquisition.
One strategy could be to use the staged mydriasis and dilate just those patients where image quality was not sufficient without pupil dilation. The PPV reported in the present study was A recent study that validated seven AI systems for DR screening in the real world based on protocols of four retinal images per eye found PPVs ranging from Regarding the population evaluated in the present study, even though the screening was performed on a State that presents the 3rd highest human development index HDI of the country [ 21 ], over half of participants had their first fundus evaluation during this initiative, despite having a diabetes duration of Brazil is considered to host the sixth biggest population of individuals with diabetes worldwide [ 22 ]; being a country with continental dimensions and heterogeneous realities, Brazil also has many differences regarding social and economic aspects.
As an example, a comparison between data collected in Blumenau Southern Brazil and Itabuna, situated in Bahia state Northeastern Brazil , ranked 22nd for HDI, shows significant differences on the health profile of patients who underwent DR screening: the present sample from Blumenau, consisting of individuals aged In contrast, a sample of individuals with diabetes from Itabuna aged Despite the southern region of Brazil being one of the most developed in the country, with the municipality of Blumenau boasting one of the highest Human Development Index HDI levels nationwide, access to early detection of diabetic retinopathy remains highly limited.
Thus, implementing mass screening programs and potentially incorporating regular and continuous assessment utilizing portable cameras in primary healthcare facilities could help decrease waiting times and improve access.
This approach would serve as an effective strategy to mitigate diabetes-related blindness cases. A sentence has been included in the discussion to address this aspect. We believe the main strength of this study is to present an automatic system with a potential to yield a high sensitivity for DR screening after evaluation of a single retinal image per eye; of note, the sensitivity attained was higher than the pre-specified endpoint for FDA approval of an automatic DR screening system [ 5 ].
Further steps for a DR screening program that would deploy the present tool could include acquisition of a second fundus image per eye only for detected cases, thereby rendering the screening process simpler for most patients, who would only need one image; further studies are needed to investigate this hypothesis.
Our study has several limitations, the most notable of which is that human grading was performed by only one specialist, a potential source of bias. Additionally, automatic evaluation was performed only on images with sufficient quality, limiting partially our conclusions regarding the real world, when a considerable rate of patients has ungradable images, mainly due to cataracts.
Furthermore, diabetic maculopathy was not evaluated with gold standard methods; instead, its presence was inferred in non-stereoscopic images. Finally, the lack of comprehensive clinical and laboratory data is also a limitation of the current study.
This study presents a new concept of a single-image approach for diabetic retinopathy screening. However, due to its methodological limitations, particularly the fact that it had only one evaluator, its results need to be interpreted with caution.
A high sensitivity prototocol was obtained for DR screening with a portable retinal camera and automatic analysis of only one image per eye.
Further studies are needed to clarify whether a simpler strategy as compared to the traditional, two images per eye protocol, could contribute to superior patient outcomes, including increased adherence rates and increased overall efficacy of DR screening programs.
Scanlon PH. The contribution of the English NHS Diabetic Eye Screening Programme to reductions in diabetes-related blindness, comparisons within Europe, and future challenges. Acta Diabetol. Epub Apr 8. PMID: ; PMCID: PMC Malerbi FK, Melo GB.
Feasibility of screening for diabetic retinopathy using artificial intelligence, Brazil. Bull World Health Organ. Epub Aug Song A, Lusk JB, Roh KM et al. Practice Patterns of Fundoscopic Examination for Diabetic Retinopathy Screening in Primary Care. JAMA Netw Open. Ruamviboonsuk P, Tiwari R, Sayres R et al.
Real-time diabetic retinopathy screening by deep learning in a multisite national screening programme: a prospective interventional cohort study.
Lancet Digit Health Apr;4 4 :e— We kindly request any interested parts to contact the authors directly for obtaining access to the database when applicable.
Klein R, Klein BEK. Epidemiology of eye disease in diabetes. In: Flynn HW Jr, Smiddy WE, editors. Diabetes and ocular Disease: past, present, and future therapies. Cham: The foundation of the American Academy of Ophthalmology; Google Scholar.
World Health Organization. Tool for the assessment of diabetic retinopathy and diabetes management systems. Genebra: WHO; Shi L, Wu H, Dong J, Jiang K, Lu X, Shi J.
Telemedicine for detecting diabetic retinopathy: a systematic review and meta-analysis. Br J Ophthalmol. Article PubMed Google Scholar. Early Treatment Diabetic Retinopathy Study Research Group. Grading diabetic retinopathy from stereoscopic color fundus photographs—an extension of the modified Airlie House classification.
ETDRS report number Article Google Scholar. Sociedade Brasileira de Diabetes. Diretrizes da Sociedade Brasileira de Diabetes — São Paulo, SP: A. Farmacêutica, Fong S, Aiello LP, Gardner TW, King GL, et al. Diabetic retinopathy. Diabetes Care. Hilgert GR, Trevizan E, de Souza JM.
Uso de retinógrafo portátil como ferramenta no rastreamento de retinopatia diabética. Rev Bras Oftalmol. Russo A, Morescalchi F, Costagliola C, Delcassi L, Semeraro F. Comparison of smartphone ophthalmoscopy with slit-lamp biomicroscopy for grading diabetic retinopathy. Am J Ophthalmol. e1 Epub Nov 7 PMID: Toy BC, Myung DJ, He L, et al.
Smartphone-based dilated fundus photography and near visual acuity testing as inexpensive screening tools to detect referral warranted diabetic eye disease. Bolster NM, Giardini ME, Bastawrous A. The diabetic retinopathy screening workflow: potential for smartphone imaging.
J Diabetes Sci Technol. Williams GA, Scott IU, Haller JA, Maguire AM, Marcus D, McDonald HR. Single-field fundus photography for diabetic retinopathy screening. Ryan ME, Rajalakshmi R, Prathiba V, Anjana RM, Ranjani H, Narayan KMV, et al.
Comparison among methods of retinopathy assessment CAMRA study. Abràmoff MD, Lavin PT, Birch M, et al. Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices.
NPJ Digit Med. Vedula SS, Tsou BC, Sikder S. Artificial intelligence in clinical practice is here—now what? JAMA Ophthalmol. Download references. We thank Dr. Igor F. Teodoro and Dr. Carlos Augusto S.
Borges for their contributions in the development of this study. Daniel Ferraz for the reviews. The project received financial support from FAEPA Foundation for the Support of Teaching, Research and Service of the University Hospital - FMRP-USP.
Division of Ophthalmology, Ribeirão Preto Medical School, University of São Paulo, , Bandeirantes Ave, Ribeirão Preto, SP, , Brazil. Jéssica Deponti Gobbi, João Pedro Romero Braga, Moises M. Lucena, Victor C. Department of Applied Mathematics and Statistics, University of São Paulo, São Carlos, Brazil.
Department of Ophthalmology, National University Hospital, Singapore, Singapore. You can also search for this author in PubMed Google Scholar. RJ was the primary contributor to research design. JG, VB, JB and MM were responsible for research execution and data acquisition. RJ, DF, MF, and VK were the primary contributors to data analysis and interpretation.
Manuscript was prepared by RJ, JB, VB, MM, JG, with critical revisions provided by RJ, DF and VK. Correspondence to Rodrigo Jorge. Every volunteer received clear explanations about the involved procedures and filled in a declaration of informed consent prior to their participation.
Sponsors 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.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution 4. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material.
If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.
Reprints and permissions. Gobbi, J. et al. Efficacy of smartphone-based retinal photography by undergraduate students in screening and early diagnosing diabetic retinopathy.
Int J Retin Vitr 8 , 35 Download citation. Received : 04 March Accepted : 23 May Published : 07 June Anyone you share the following link with will be able to read this content:. Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative. Skip to main content. Search all BMC articles Search. Download PDF. Original article Open access Published: 07 June Efficacy of smartphone-based retinal photography by undergraduate students in screening and early diagnosing diabetic retinopathy Jéssica Deponti Gobbi 1 , João Pedro Romero Braga 1 , Moises M.
Lucena 1 , Victor C. Bellanda 1 , Miguel V. Abstract Background To evaluate the efficacy of retinal photography obtained by undergraduate students using a smartphone-based device in screening and early diagnosing diabetic retinopathy DR.
Methods We carried out an open prospective study with ninety-nine diabetic patients eyes , who were submitted to an ophthalmological examination in which undergraduate students registered images of the fundus using a smartphone-based device.
Results Concerning the presence or absence of DR, we found an agreement rate of Conclusion The smartphone-based device showed promising accuracy in the detection of DR Background: Diabetic retinopathy DR is one of the most important complications of Diabetes Mellitus DM and its incidence is intrinsically related to the duration of the disease and level of glycemic control.
Materials and methods Patients and ethics We conducted a prospective, open study, collecting data from diabetic patients eyes at the diabetic retinopathy screening clinic of Hospital das Clínicas de Ribeirão Preto HC-FMRP-USP , a high complexity general hospital in Brazil. Ophthalmological evaluation During their appointment for diabetic retinopathy evaluation, patients in the study underwent two types of assessments: one being standard seven field color stereoscopic photography of the fundus captured by an experienced nurse through a tabletop fundus camera Canon CR-2 Digital Non-Mydriatic Retinal Camera—demonstrated on Fig.
Full size image. Side view of the smartphone-based device used in the study. Results Demographics Participants had a mean age of Table 1 Demographic data concerning all 99 patients included in the study Full size table.
Table 2 Frequency of diagnoses comparing the degree of retinopathy as determined by the smartphone-based device and the gold standard Full size table. Table 3 Values of interobserver and intraobserver agreement when the presence or absence of DR Full size table.
Table 4 Interobserver and intraobserver agreement values for the presence of proliferative or non-proliferative DR Full size table. Table 5 Sensitivity and specificity of smartphone-based device ocular fundus images according to diabetic retinopathy severity scale Full size table.
Discussion Our study was able to verify that retinal images obtained by undergraduate students using a smartphone-based device showed satisfactory performance when compared to the reference standard for the diagnosis of DR. Conclusion High cost and low availability of eye examination, especially when requiring in-site experts, represent an important limitation for DR screening.
Availability of data and materials All data generated in this study, including the images obtained through both the analysed method and the gold standard, were saved on private cloud storage Google Drive ® for patient safety and privacy.
Abbreviations DR: Diabetic retinopathy DM: Diabetes mellitus CI: Confidence Interval NPDR: Non proliferative diabetic retinopathy PDR: Proliferative diabetic retinopathy. References Klein R, Klein BEK. Google Scholar World Health Organization. Google Scholar Shi L, Wu H, Dong J, Jiang K, Lu X, Shi J.
Article PubMed Google Scholar Early Treatment Diabetic Retinopathy Study Research Group. Article Google Scholar Sociedade Brasileira de Diabetes.
Article Google Scholar Hilgert GR, Trevizan E, de Souza JM. Google Scholar Russo A, Morescalchi F, Costagliola C, Delcassi L, Semeraro F. Article PubMed Google Scholar Toy BC, Myung DJ, He L, et al. Article PubMed Google Scholar Bolster NM, Giardini ME, Bastawrous A.
Article Google Scholar Williams GA, Scott IU, Haller JA, Maguire AM, Marcus D, McDonald HR. Article PubMed Google Scholar Ryan ME, Rajalakshmi R, Prathiba V, Anjana RM, Ranjani H, Narayan KMV, et al.
Thank you for retinopahy nature. You are using a imaginh version with limited support for CSS. To obtain the best experience, we recommend imagin use rstinal more Fueling before a game to Diabettic browser or turn off compatibility mode in Retijopathy Explorer. In the retinoopathy, to Energy metabolism and cardiovascular health continued support, we imagibg displaying the site without styles and JavaScript. Diabetes is a globally prevalent disease that can cause visible microvascular complications such as diabetic retinopathy and macular edema in the human eye retina, the images of which are today used for manual disease screening and diagnosis. This labor-intensive task could greatly benefit from automatic detection using deep learning technique. We also provide novel results for five different screening and clinical grading systems for diabetic retinopathy and macular edema classification, including state-of-the-art results for accurately classifying images according to clinical five-grade diabetic retinopathy and for the first time for the four-grade diabetic macular edema scales. Megh Ketur Fueling before a gameMuscle definition secrets HenryBernard SzirthIamging Bhagat; Utility Liver detoxification program Remote Point-Of-Care Diabetlc Imaging for Screening and Diagnosis of Diaabetic Retinopathy: A Pilot Study. Purpose : In the Diabetic retinopathy retinal imaging era, Size diversity technologies are rising to Diabeitc forefront Fueling before a game contactless ophthalmic care. Point-of-care Optical Coherence Tomography OCT and fundus photography remotely analyzed by an off-site retina specialist tele-R must be validated for screening retinal disorders. This study assesses the feasibility of tele-R as a screening tool for diabetic retinopathy DR in an outpatient clinical setting. Methods : Rtinopathy retrospective study was conducted on 16 patients 32 eyes, 28 with DR, and 4 controls presenting to the retina clinic RC of an urban academic medical center. Diabettic OCT-B and 45 Diabegic fundus photographs of the posterior pole were taken using a Topcon Maestro 3D OCT-1 unit, and 3D topographical maps of the macula were generated.
0 thoughts on “Diabetic retinopathy retinal imaging”