ULTRAWIDE-FIELD COLOR FUNDUS IMAGES AND ARTIFICIAL INTELLIGENCE FOR DIAGNOSIS OF DIABETIC RETINOPATHY: A Systematic Review and Meta-Analysis.
Kunihiko Hirosawa, Takenori Inomata, Ken Nagino et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review and meta-analysis assessed artificial intelligence (AI) for diabetic retinopathy (DR) evaluation using ultrawide-field fundus images. The review included 17 studies, with four contributing to the diagnostic-accuracy meta-analysis; all used Optos software. For AI-based DR screening, pooled sensitivity was 85.0%, specificity was 72.5%, and area under the receiver operating characteristic curve was 0.870 (0.838-0.897). The authors judged sensitivity acceptable but specificity limited, and advised caution because the evidence base was small.
Key findings
- Across four studies, pooled AI screening sensitivity was 85.0% and specificity was 72.5%.
- The summary area under the receiver operating characteristic curve was 0.870 (0.838-0.897).
- The systematic review included 17 studies, and all used Optos software.
Why it matters
Ultrawide-field imaging captures a larger retinal area without pupil dilation, and AI assessment could support DR screening. The pooled results suggest reasonable sensitivity but limited specificity, which is relevant when considering the balance between detecting disease and generating false-positive assessments. The small evidence base and use of a single platform constrain confidence in broader applicability.
What to keep in mind
Only four studies contributed to the meta-analysis, and all included studies used Optos software. The authors note that larger, diverse datasets and external validation across platforms are needed.