Eye disease affects 2.2 billion people globally, which in turn adversely affects schooling, economic productivity, and participation in social life. The primary conditions contributing to visual impairment and blindness include cataracts, age-related macular degeneration (AMD), glaucoma, diabetic retinopathy (DR), refractive error, and presbyopia. Early detection of eye disease can provide substantial benefits in prompting treatment to reduce progression and mitigate disability. Compared with other regions, South Asia has the most cases of visual impairment due to cataracts and uncorrected refractive error. The combination of poverty, poor living and working environments, and limited health care access have long endangered eye health in Bangladesh. Coastal Bangladesh is particularly impacted by eye disease due to economic deprivation and limited healthcare access. The coastal population mostly works in fishing and agriculture, have prolonged sunlight exposure, and inadequate occupational eye protection. This low-lying region, with 35 million people, is especially vulnerable to climate disasters and global warming. High rates of chronic disease, especially diabetes mellitus Type 2 and hypertension, coupled with limited screening and treatment, shape the area's health profile, with the increasing prevalence of eye diseases such as DR, glaucoma, and visual impairment. To address the issues of poor health, accessibility, and affordability of eye care, Artificial Intelligence (AI) applications, such as Artificial Intelligence (AI)-assisted fundus imaging, can be applied in eye screening. Medical AI applications have the potential to improve the quality and efficiency of healthcare, reduce healthcare costs, optimize treatment plans, and bolster the development of primary healthcare. They can identify presumptive DR, hypertensive retinopathy (HR), AMD, and glaucoma by analyzing the retina and optic disc of fundus images with moderate accuracy and high efficiency, thus helping address the lack of local eye care professionals. Data Yakka developed a human-AI collaboration that delivers affordable and transformative community-based eye screening to underserved communities in the coastal Bangladesh region of Char Fasson. The "Amar Chokh Amar Alo" (My Eyes, My Light) initiative creates and implements comprehensive eye screening that combines AI-assisted eye screening and grassroots partnerships with trusted non-health non-governmental organizations (NGOs). It has three objectives: 1) Enhancing accessibility and affordability of eye screening; 2) Supporting high quality and efficient treatment of those problems detected via screening, 3) Collecting fundus images to refine or train AI algorithms in the future. This project was designed to evaluate the feasibility, performance, equity, and cost of this model of eye screening and its implications for global eye disease. The implementation of participant recruitment, data collection, screening, and follow-up was separated into twelve steps. This standardized framework ensured the integration of screening with data collection and follow-up eye care services. Based on risk stratification by diabetes, hypertension, age 50+ years, and/or optometrist recommendation, fundus imaging was offered selectively to higher-risk patients.
Setting Location and Participants Selection of a target region within Bangladesh considered four factors that would facilitate successful screening: 1) Relatively high population density and adequate transportation network, 2) the presence of a Cyclone Preparedness Program (CPP) volunteer network, 3) An eye care hospital willing to serve referred clients, and 4) Support and assistance from local and regional government. Based on these criteria, the investigators selected and implemented the screening project in Char Fasson of coastal Bangladesh. Char Fasson is a sub-district (Upazila) of the Bhola District and has a population of 518,792. While the investigators initially considered another coastal sub-district of comparable size, Char Fasson was chosen because of data collection challenges in the original choice. Char Fasson showed greater levels of poverty and more limited healthcare access while governmental coordination was exemplary. Within Char Fasson, the investigators made special efforts to serve outlying, smaller islands within the sub-district. Partnership Team Successful delivery of the program relied on effective collaboration across a range of diverse partners, including major operational organizations, clinical affiliates, commercial suppliers, and technical advisors. The major partners responsible for day-to-day work included the organizer, Data Yakka (Palo Alto, California, USA) which is a healthcare technology company specializing in AI-assisted medical screening platforms and data management systems, the regional government (Char Fasson Upazila), the Bangladesh Disaster Preparedness Centre (BDPC, Dhaka), and the national government's Cyclone Preparedness Program (CPP, Dhaka). Among them, BDPC provided personnel for program execution and supervision while CPP mobilized its volunteers for paid work in clinical site staffing and program outreach to remote areas. The total number of CPP community volunteers in coastal Bangladesh is around 76,000, with 3,300 in Char Fasson. The main clinical partner was the Dr. K. Zaman Bangladesh National Society for the Blind Eye Hospital (BNSB, Mymensingh, Bangladesh), which provided on-site ophthalmology and optometry staffing as well as remote fundus image review and clinical consultation as needed. The Ad-din Medical College Hospital (Dhaka) provided additional resources. Clients requiring cataract surgery were referred to regional facilities vetted by BNSB or to services provided by the Ad-Din Hospital. Commercial supplier partners included VisionSpring (New York, New York, USA), a not-for-profit organization providing low-cost reading glasses that the program distributed free to those in need, Topcon Corporation (Tokyo, Japan) for its NW-500 robotic fundus imaging camera, and Aurolab (Madurai, India) for its HAWK I T2 slit lamp examination device. Thirona Retina (Nijmegen, Netherlands) supplied its RetCAD AI software algorithms for the detection of suspected DR, AMD, and glaucoma. The software evaluates image quality, generates heatmaps with deep learning, and provides disease likelihood scores. Previous research demonstrated that RetCAD showed high performance: 96% sensitivity and 94% specificity on the Messidor-2 dataset for DR, 95% sensitivity and 97% specificity on the private1 dataset for AMD, and 95% sensitivity and 86% specificity on the REFUGE dataset for glaucoma. Technical advisors, including Stanford University (Palo Alto, California, USA), Royal Victorian Eye and Ear Hospital (Melbourne, Australia), and Glaucoma Australia (Sydney, Australia), shared their expertise, guided technical program components, and provided external clinical oversight and case-by-case review for complex or unclear diagnosis. Data security The security and integrity of patient data was ensured through multiple layers of protection. The secure, encrypted patient data portal was developed by Data Yakka using Java (Oracle, Austin, Texas, USA) with Angular for data entry and Spring Boot for backend micro services management, including a program dashboard for real-time monitoring. Patient information was securely stored in PostgreSQL RDS (Amazon Web Services, Seattle, Washington, USA), while fundus images were stored in Amazon S3 with server-side encryption using managed keys. This protected all sensitive data from unauthorized access. Personal identifiers were segregated from clinical details to enhance patient privacy. Keycloak was implemented for secure authentication and authorization. A comprehensive audit trail was maintained, logging all user interactions (access, edits, and deletions) to ensure compliance with the European Union's General Data Protection Regulation (GDPR) and the U.S.'s Health Insurance Portability and Accountability Act of 1996 (HIPAA). This also provided immediate detection of potential data security threats. Finally, to enhance system reliability, the investigators deployed Multi-AZ, which enabled automatic failover, high availability in case of hardware disruptions, and disaster recovery. Ethical Considerations Ethical approval was obtained from the Institutional Ethical Committee of BNSB (reference date/number 118/2025). To comply with Bangladesh's data regulatory policies, consent was waived for most components of the screening process and was obtained for the fundus images (retina and optic disc) because they constitute personal health information, are stored electronically, and their use in the future is anticipated. Before taking fundus images, screening staff described the reason, purpose, acquisition, use, confidential storage, deletion, and withdrawal mechanism for the fundus images and the required personal information. Informed consent was then obtained from the participants, including their signatures to verify their voluntary participation. The study was conducted in strict compliance with Declaration of Helsinki and ethical international and country-specific requirements. 2.2 Screening \& Care Process Timeline The program duration was planned for 10 months. This paper covers the program's initial stage and is current as of early April 2025. From October to December 2024, the investigators carried out a three-month-long creation and refinement of the screening model and electronic platform. Next, in December 2024, the model was tested and further enhanced during a pre-opening testing period for troubleshooting and initial data collection. Between January 2025 and July 2025, full-time data collection is planned with participant recruitment and screening then follow-up. Process The implementation of participant recruitment, data collection, screening, and follow-up was separated into twelve steps. This standardized framework ensured the integration of screening with data collection and follow-up eye care services. Fundus imaging was offered selectively to higher-risk patients with specific criteria including: random blood glucose \>7.9 millimoles per liter (mmol/L), blood pressure ≥140/90 millimeters of Mercury (mmHg), age ≥50 years, or clinical suspicion of retinal pathology based on patient report or slit lamp examination. Data Analysis Statistical analyses will be performed using Microsoft Excel. After data collection and cleaning, the investigators calculated descriptive statistics, including percentages, means, and standard deviations (SD), to analyze proportions, central tendency, and dispersion of demographic and diagnostic data. For statistical comparisons, the calculations for statistical significance will use 2-tailed t-tests with a threshold of p\<0.05 considered statistically significant. Eye condition diagnosis was based on the presence of a condition in either one or both eyes. Estimates of cost will initially calculate costs associated with the first 8000 screened individuals. The investigators also projected trends in both screening numbers and costs through a hypothetical 6-month time point by extrapolating screening numbers and variable costs of the program to 6 months. Fixed costs are non-recurring, volume-independent expenses, including program computers, blood pressure and blood glucose measurement devices, and fundus imaging and slit lamp equipment. Variable costs include consumables and operation fees based on unit usage, calculated by multiplying unit costs by the total screening, such as consumable testing costs, eyeglasses distribution, and office supplies. Per-person cost breakdown followed a dual-tier approach: fixed costs were evenly allocated among all screened individuals, while marginal cost reflected the average variable expenditure per screening.
Study Type
OBSERVATIONAL
Enrollment
20,000
The eye disease screening process involved twelve distinct steps were that organized through a corresponding software platform (electronic health record). These steps included: (1) community awareness campaign, (2) participant registration, (3) blood pressure and finger-stick blood glucose measurement, (4) basic vision test, (5) on-site optometrist evaluation, (6) obtaining informed consent for imaging, (7) fundus imaging, (8) automated AI-based disease detection, (9) on-site ophthalmologist examination, (10) remote eye specialist review, (11) on-site counselor discussion, (12) referral for local surgery. This standardized protocol promoted the alignment of eye screening with data gathering and ongoing follow-up eye care interventions. All participants were offered steps 1 to 5. Those participants eligible for fundus imaging (Steps 6-8) included those with diabetes, hypertension, age 50+, or optometrist recommendation based on presenting symptoms.
Individuals diagnosed with high grade cataracts associated with moderate to sever visual impairment were offered cataract surgery, either through local surgery services or regional specialized vision services.
Shashibhushan Clinic
Char Fasson, Bhola, Bangladesh
RECRUITINGNumber of individuals screened
The feasibility goal was to demonstrate that high-quality free services could be provided to the population through the eye screening model's work-flow, computer platform, and use of AI-assisted interpretation of fundus images.
Time frame: 12 months
Eye Disease Prevalence
Prevalence of Various Eye Diseases among those screened, including fundus diagnoses (Diabetic Retinopathy, Hypertensive Retinopathy, Age-related Macular Degeneration, and Glaucoma), cataracts, refractive errors, and presbyopia.
Time frame: 12 months
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.