An Automated Detection System for Microaneurysms That Is Effective across Different Racial Groups
George Michael Saleh1, James Wawrzynski2, Silvestro Caputo3
1Moorfields Eye Hospital NHS Foundation Trust, London, UK; Department of Computing, Faculty of Engineering, University of Surrey, Guildford, UK; National Institute for Health Research Biomedical Research Centre, Moorfields Eye Hospital NHS Foundation Trust and UCL Institute of Ophthalmology, London, UK.
Journal of Ophthalmology
|January 12, 2017
Summary
An automated system reliably identifies the absence of diabetic retinopathy (DR) in digital fundus images across diverse populations. This technology shows promise for scalable DR screening, efficiently identifying patients without the condition.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Diabetic retinopathy (DR) screening is crucial for preventing vision loss.
- Accurate identification of patients without DR is essential for efficient screening resource allocation.
- Automated analysis of digital fundus images (DFIs) is a key area in DR screening research.
Purpose of the Study:
- To evaluate a novel automated algorithm for detecting the absence of diabetic retinopathy (DR) in DFIs.
- To assess the algorithm's performance across diverse geographical locations and racial groups.
Main Methods:
- A retrospective, masked, and controlled study involving 17,850 DFIs from six countries.
- Comparison of automated system performance against human graders for DR detection.
- Analysis of system sensitivity and specificity across different countries and racial demographics.
Main Results:
- The automated system demonstrated high sensitivity (90.1%-93.5%) and specificity (79%-83.2%) for DR detection across all studied countries.
- Minimal variability in performance was observed between different countries and racial groups.
- The system effectively identified patients without DR.
Conclusions:
- The automated DR detection algorithm shows consistent and reliable performance across diverse populations.
- The findings suggest the scalability of this automated platform for widespread DR screening.
- This technology can facilitate the rapid identification of individuals without DR, optimizing screening services.
