Author: Claude (AI), from PubMed abstracts · Reviewed by: Dr Mina Abdelmseih · Reviewed: 2026-11-15 · Review due: 2027-11-15

Educational content only. Not personalised medical advice. Clinical decisions remain the responsibility of the treating clinician.

Smartphone and portable fundus imaging

Smartphone-based and handheld fundus cameras are portable, low-cost and easy to use in remote places, and images can be sent at once for remote grading [S2]. They can support screening or triage for referable diabetic retinopathy (DR), glaucoma-related disc changes and selected retinopathy of prematurity applications, with accuracy that varies by device, protocol, operator and disease [S11]. Combined with artificial intelligence (AI) and task-shifting, they are a practical route to eye screening in low-resource settings [S9].

Key points - Smartphone ophthalmoscopy detected any DR with pooled sensitivity 87% and specificity 94%, and proliferative DR with 92% and 99%, but mild non-proliferative DR with only 39% sensitivity [S1]. - Smartphone-based AI detected referable DR with pooled sensitivity 98.2% and specificity 81.2% in 6 studies (3,931 patients), four using one algorithm [S7]. - In developing countries, AI-based DR screening had pooled sensitivity 0.933 and specificity 0.903 for referable DR; portable cameras and general physicians agreed well with gold standards [S9]. - In Uganda, a smartphone adapter detected a vertical cup-to-disc ratio above 0.5 with 67.7% sensitivity and 96.7% specificity against OCT [S3]. - Image quality, field of view, data security, privacy, regulation and maintenance remain key challenges [S11].

Evidence last checked: 2026-10-17 (PubMed abstracts).

Red flags: findings on the image that need urgent action

• Proliferative DR on a screening image. Smartphone imaging detects it with the highest accuracy of any DR grade [S1]; refer within days [G1].
• Disc swelling in a neurology or emergency patient. In 79 neurology inpatients, 14.1% had neurologically relevant fundus pathology, including disc swelling, and routine direct ophthalmoscopy missed all of it [S5]; same-day assessment [EIC expert opinion].
• An ungradable image. Not a normal result: re-image or refer, since dense cataract and poor acquisition cause image failure [S3].
• Suspected macular oedema. Smartphone sensitivity for DMO was 79% pooled [S1] and as low as 29.4% with some low-cost devices [S6], so symptoms or reduced acuity need examination even with a "normal" image.

Recognise it: device types and what they can do

• Smartphone-based fundus imaging (SBFI). A review of 30 hardware solutions in three main designs found some give fields of view and quality adequate for DR, disc and some ROP screening [S11].
• Digital ophthalmoscopes (smartphone, desktop, handheld): uptake is limited by image quality in undilated eyes, data confidence and lack of usage guidelines, but they are gaining acceptance in underserved areas [S8].
• Portable non-mydriatic fundus cameras. In neurology inpatients, portable non-mydriatic photography (sensitivity 45.5%) was more accurate than smartphone fundoscopy (30% to 40%), though the smartphone had fewer screening failures (1% vs 13%) and was quicker [S5].
• Teaching. Medical students identified abnormal discs more accurately with smartphone fundoscopy (74.4%) than with direct ophthalmoscopy (68.1%), and most preferred it [S4].

Confirm it: getting a gradable, reliable image

1. Dilate when the protocol requires it. All low-cost smartphone devices in one review used mydriatic drops [S6], and image quality in non-mydriatic eyes is a concern [S8].
2. Train the operator. After brief training, allied health workers can acquire usable images in selected settings [S11].
3. Check gradability. In Uganda, 138 eyes were excluded for poor image acquisition such as dense cataract [S3].
4. Know the reference standard. Reported accuracy depends on device, protocol, operator and reference standard; no universal performance threshold exists [S11].

Manage it: running a portable imaging programme at each tier

Before acting on this section, check the national and local guidelines, product information and regulatory approvals that apply where you practise. See the medical disclaimer.

Tier A (primary care / no specialist)

1. Trained non-specialists can capture images for remote grading, supporting task-shifted screening and referral pathways [S9] [S11].
2. Send images for remote grading by trained graders or ophthalmologists [S2], or use a validated AI system where available [S7] [S9].
3. Refer every ungradable image or AI-positive result (see table).

Tier B (general eye clinic)

1. Use portable cameras for documentation and patient counselling, as well as screening [S2].
2. Use smartphone disc imaging to estimate cup-to-disc ratio in population screening; minimal training and low cost suit low-resource settings [S3].
3. Use smartphone fundoscopy to teach and to support non-ophthalmic colleagues [S4] [S5].

Tier C (specialist centre)

1. Act as the grading and referral hub for telemedicine programmes, which have widened screening coverage and timely referral of sight-threatening DR [S2].
2. Audit programme accuracy locally: smartphone-based approaches are feasible but outcomes vary and need context-specific validation [S10].
3. Address data governance, maintenance, regulatory compliance and photobiological safety when choosing devices [S11].

Refer when

Pitfalls

• Treating an ungradable image as normal [S3].
• Relying on smartphone images to exclude mild DR or macular oedema [S1] [S6].
• Assuming all devices perform alike: camera type significantly affected accuracy [S9].
• Using undilated imaging where the protocol was validated with dilation [S6] [S8].
• Ignoring privacy and data security when sending images [S11].

Evidence notes

• Pooled estimates are heterogeneous [S1]; four of six smartphone AI studies used one algorithm (Medios) [S7].
• The 2026 meta-analysis reported pooled odds ratios rather than sensitivity and specificity, and found possible publication bias for community screening [S10].
• Most evidence concerns DR; evidence for glaucoma and ROP is thinner [S3] [S11].
• Not covered by these sources: cost-effectiveness, named device comparisons, regulatory approval status by country and long-term programme outcomes.

Related articles

• Diabetic retinopathy screening programmes
• Artificial Intelligence in Retina and OCT
• Integrating eye care into primary health care

References

See the numbered source list below. Each marker [S#] links to its PubMed and DOI record.