Day: October 30, 2026

Title: Detection of Anaemia from Retinal Fundus Images via Deep Learning

Authors: Akinori Mitani, Abigail Huang, Subhashini Venugopalan, Greg S. Corrado, Lily Peng, Dale R. Webster, Naama Hammel, Yun Liu, and Avinash V. Varadarajan

Publication Source: Nature Biomedical Engineering (2019)

Anaemia affects over 1.6 billion people worldwide, yet conventional screening relies on invasive blood draws that require trained clinical staff and specialized laboratory equipment. To address this barrier, medical researchers are exploring non-invasive diagnostic tools that leverage artificial intelligence to identify systemic health conditions through routine eye examinations.

A landmark study published in Nature Biomedical Engineering investigated whether deep learning algorithms could accurately predict haemoglobin (Hb) concentration and screen for anaemia using standard color retinal fundus photographs.

What Was the Study Testing?

Researchers trained and evaluated an advanced deep neural network on a massive dataset from the UK Biobank, analyzing fundus photographs from over 57,000 individuals alongside basic demographic metadata.

The AI system was evaluated across several key parameters:

• Non-Invasive Prediction: Estimating continuous haemoglobin levels directly from retinal blood vessel morphology without drawing blood.
• Metadata Integration: Combining fundus images with basic non-invasive physical data (age, sex, and blood pressure) to enhance diagnostic precision.
• Diabetic Subgroup Screening: Assessing screening performance specifically in diabetic patients, who routinely undergo annual retinal fundus examinations.

Key Findings: Diagnostic Accuracy & Model Insights

The trial demonstrated that deep learning applied to retinal photographs offers a highly reliable method for non-invasive anaemia screening:

1. Robust Predictive Performance

• Haemoglobin Estimation: The combined AI model achieved a mean absolute error (MAE) of 0.63 g/dL (95% CI: 0.62–0.64) in predicting blood haemoglobin concentration.
• Anaemia Detection: The model achieved an Area Under the Curve (AUC) of 0.88 (95% CI: 0.86–0.89) for detecting general anaemia and an AUC of 0.95 (95% CI: 0.93–0.97) for moderate-to-severe anaemia.
• Diabetic Subgroup Consistency: Among diabetic participants (n = 539), the system maintained an impressive AUC of 0.89, proving its viability as an automated add-on during routine diabetic retinopathy screening.

2. Interpretability & Visual Saliency

• Optic Disc & Vascular Focus: Saliency map visualizations revealed that the neural network primarily focuses on fine anatomical structures around the optic disc and main retinal vessel branches.
• Biomarker Discovery: The model relies on subtle microvascular variations and vessel coloration differences that are imperceptible to the human eye.

What Does This Mean for Healthcare & Screening?

The ability to detect anaemia from fundus photographs transforms standard eye exams into powerful systemic health assessments. Because retinal photography is fast, painless, and already widely available in eye clinics and optical centers, incorporating AI screening could catch undiagnosed anaemia early without extra blood draws or patient discomfort.

This technology holds immense potential for opportunistic screening—enabling patients visiting an eye clinic for routine checkups to receive immediate insights regarding their blood health.

The Bottom Line

AI-driven retinal analysis offers a promising, painless alternative to conventional blood tests for anaemia screening. While further prospective validation in broader global populations is ongoing, deep learning model integration into clinical workflows represents a major leap forward in non-invasive preventive care.

Always consult a qualified healthcare provider or ophthalmologist to discuss comprehensive screening options and blood health evaluations.

Read the full study in Nature Biomedical Engineering.

Title: Application of Comprehensive Artificial Intelligence Retinal Expert (CARE) System: A National Real-World Evidence Study

Authors: Duoru Lin, Jianhao Xiong, Congxin Liu, Lanqin Zhao, Zhongwen Li, Shanshan Yu, Xiaohang Wu, Zongyuan Ge, et al., Daniel SW Ting, Tien Y Wong, Yuzhong Chen, and Haotian Lin

Publication Source: The Lancet Digital Health (2021)

Delivering specialized ophthalmic care in primary and community settings remains a major global challenge due to shortages of trained eye specialists. While AI algorithms have shown great success in controlled laboratory settings, proving their consistency across diverse real-world healthcare environments requires rigorous multi-center testing.

A nationwide prospective study published in The Lancet Digital Health evaluated the Comprehensive AI Retinal Expert (CARE) system—a unified deep-learning network designed to screen for 14 common retinal diseases simultaneously across real-world clinical workflows in China.

What Was the Study Testing?

Unlike traditional AI systems that use separate models for each disease, the CARE system utilizes a single multi-label convolutional neural network (Inception-ResNet-V2) trained on over 200,000 fundus photographs.

The system was evaluated across 35 real-world clinical environments:

• Diverse Clinical Environments: Deployed across tertiary hospitals, community clinics, and health physical examination centers.
• Broad Spectrum Coverage: Trained to detect 14 major retinal abnormalities, including referable diabetic retinopathy, glaucomatous optic neuropathy, pathological myopia, and age-related macular degeneration.
• Cross-Ethnicity & Hardware Generalization: Validated against external public datasets (such as EyePACS in the U.S.) to verify reliability across different ethnic groups and camera models.

Key Findings: Real-World Performance & Efficiency

The extensive real-world trial validated the robustness and operational efficiency of the CARE system:

1. Specialist-Level Diagnostic Accuracy

• Tertiary Hospitals: CARE achieved a mean AUC of 0.965 across 14 conditions, matching the diagnostic performance of senior ophthalmologists.
• Community Clinics & Exam Centers: Maintained outstanding accuracy in primary care settings, achieving mean AUCs of 0.983 in community hospitals and 0.953 in physical examination centers.
• Global Generalizability: When tested on the U.S. EyePACS dataset, CARE achieved an AUC of 0.960 for referable diabetic retinopathy, demonstrating strong performance across diverse populations and imaging hardware.

2. High Computational Efficiency

• Low Hardware Demand: By using a single neural network rather than 14 isolated models, CARE required only 3–6 GB of GPU RAM during operation.
• Field Deployability: The low computational footprint allows the software to run locally on standard desktop hardware in remote, low-resource primary clinics without requiring continuous cloud connectivity.

What Does This Mean for Healthcare & Screening?

The CARE study proves that a single AI model can reliably screen for multiple eye conditions simultaneously in real-world settings. By triaging patients at the primary care level, the system ensures that high-risk individuals are promptly referred to specialists while preventing secondary and tertiary hospitals from becoming overwhelmed with routine checkups.

This approach bridges the gap between urban medical hubs and underserved rural communities, bringing expert-level diagnostic support directly to primary healthcare providers.

The Bottom Line

The CARE system demonstrates that multi-disease AI screening is highly effective, efficient, and ready for deployment in everyday clinical settings. It provides a scalable solution for expanding access to specialized eye care worldwide.

Consult your eyecare professional to learn more about AI-assisted screenings and comprehensive eye health assessments.

Read the full open-access study in The Lancet Digital Health.

Title: Application of Artificial Intelligence System for Screening Multiple Fundus Diseases in Chinese Primary Healthcare Settings: A Real-World, Multicentre and Cross-Sectional Study of 4795 Cases

Authors: Chufeng Gu, Yujie Wang, Yan Jiang, Feiping Xu, Shasha Wang, Rui Liu, Wen Yuan, Nurbiyimu Abudureyimu, et al., Qinghua Qiu, Zhi Zheng, Deng Liu, and Jili Chen

Publication Source: British Journal of Ophthalmology (2024)

Disparities in healthcare access between urban centers and remote regions present a significant challenge for early eye disease detection. Deploying artificial intelligence in primary care clinics offers a scalable solution to standardizing diagnostic quality across geographically and economically diverse populations.

A multi-center, real-world study published in the British Journal of Ophthalmology evaluated the Airdoc Retinal AI System (ARAS) across six primary healthcare centers in Shanghai (a high-income urban metropolis) and Xinjiang (a lower-income western region).

What Was the Study Testing?

The study evaluated 4,795 participants to assess the diagnostic efficacy of ARAS in primary care settings and analyze regional differences in fundus disease prevalence.

The system combines a dual-module architecture:

• Precision Localization: Uses a YOLO-v3 object detection network to identify key anatomical landmarks (the macula and optic disc).
• Multi-Task Classification: Employs an EfficientNet-B3 network to classify 14 distinct fundus conditions simultaneously.
• Comparative Regional Analysis: Compares disease patterns between urban Shanghai and rural Xinjiang to optimize local healthcare resource allocation.

Key Findings: Diagnostic Accuracy & Regional Patterns

The study confirmed high screening accuracy while uncovering distinct regional health profiles:

1. Diagnostic Efficacy in Primary Care

• Referable Diabetic Retinopathy (RDR): Achieved an overall accuracy of 0.93, with a sensitivity of 0.87 and specificity of 0.93.
• Glaucomatous Optic Neuropathy (GON): Delivered an accuracy of 0.93, sensitivity of 0.95, and specificity of 0.92.
• Pathological Myopia (PM): Recorded an accuracy of 0.93, sensitivity of 0.85, and specificity of 0.94.

2. Regional Disease Spectrum Variations

• Shanghai (Urban): Displayed a significantly higher prevalence of pathological myopia and retinal drusen, reflecting lifestyle factors associated with dense urban environments.
• Xinjiang (Rural/Developing): Exhibited higher rates of referable diabetic retinopathy, macular edema, and retinal vein occlusion among middle-aged and older adults, pointing to a need for stronger chronic disease management in remote areas.

What Does This Mean for Healthcare & Screening?

The study highlights how AI screening can adapt to different regional healthcare demands. In urban settings, AI systems help manage high volumes of myopia and age-related macular conditions. In remote or resource-limited regions, they serve as a critical frontline defense for detecting unmanaged diabetic complications and glaucoma before irreversible vision loss occurs.

By automating early disease detection in primary care, health systems can streamline referral workflows, reduce patient travel burdens, and deploy medical resources where they are needed most.

The Bottom Line

AI-powered fundus screening bridges regional healthcare gaps, delivering consistent, high-accuracy diagnostic support in both urban hospitals and rural community clinics.

Schedule a visit with an eye care specialist to explore how modern imaging and AI tools can help safeguard your vision.

Read the full study in the British Journal of Ophthalmology.