Related Experiment Video
Updated: Nov 21, 2025
05:51
Smartphone Fundus Photography
Published on: July 6, 2017
39.8K
Diabetic Retinopathy Screening Using Artificial Intelligence and Handheld Smartphone-Based Retinal Camera
Fernando Korn Malerbi1,2, Rafael Ernane Andrade1,3, Paulo Henrique Morales1,2
1Department of Ophthalmology and Visual Sciences, Federal University of São Paulo, São Paulo, Brazil.
Journal of Diabetes Science and Technology
|January 13, 2021
Summary
A deep learning algorithm achieved high accuracy in detecting diabetic retinopathy (DR) using portable retinal cameras. This technology shows promise for expanding DR screening programs in high-burden areas.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Diabetic retinopathy (DR) screening programs increasingly adopt portable retinal cameras and deep learning (DL) algorithms.
- Evaluating these novel tools is crucial for real-world application in high-burden settings.
Purpose of the Study:
- To assess the diagnostic accuracy of a DL algorithm for DR detection.
- To evaluate the performance of portable handheld retinal cameras in a large, diverse type 2 diabetes population.
Main Methods:
- Fundus photographs were taken using a portable retinal camera (Phelcom Eyer).
- Diabetic retinopathy classification was performed by human graders and a DL algorithm (PhelcomNet).
- Diagnostic accuracy metrics (AUC, sensitivity, specificity) for detecting more than mild DR were calculated.
Main Results:
- The DL algorithm demonstrated high sensitivity (97.8%) and specificity (61.4%) with an AUC of 0.89.
- Portable camera image quality was adequate in over 80% of participants.
- All false negatives were moderate non-proliferative diabetic retinopathy (NPDR) cases upon human review.
Conclusions:
- The DL algorithm exhibits good diagnostic accuracy for detecting more than mild DR in real-world, high-burden conditions.
- Portable retinal cameras and AI tools have the potential to enhance the reach of DR screening programs.
