Related Experiment Video
Updated: Dec 22, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.2K
Deep Learning for Automated Sorting of Retinal Photographs
Tyler Hyungtaek Rim1, Zhi Da Soh2, Yih-Chung Tham1
1Singapore Eye Research Institute, Singapore National Eye Centre, Singapore; Ophthalmology & Visual Sciences Academic Clinical Program (Eye ACP), Duke-NUS Medical School, Singapore.
Ophthalmology. Retina
|May 5, 2020
Summary
A new automated system, RetiSort, accurately sorts retinal photographs by type and laterality. This artificial intelligence tool enhances data preparation for deep learning research in ophthalmology.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Big data and AI in healthcare require improved data quality and preparation methods.
- Automated systems are needed to streamline the processing of medical images.
Purpose of the Study:
- To develop an automated sorting system (RetiSort) for accurate labeling of retinal photograph types and laterality.
- To address the lack of systemic methods for improving data quality in healthcare AI.
Main Methods:
- RetiSort was developed using a 3-step process involving 2 deep-learning (DL) algorithms and 1 rule-based classifier.
- The system identifies the optic disc, sorts images into macular-centered, optic disc-centered, or other fields, and determines laterality (left/right eye).
- Accuracy was evaluated on 5000 images from the SEED study and 3 public databases.
Main Results:
- RetiSort achieved 99.0% accuracy on SEED study images (48/5000 mislabeled).
- External validation showed high accuracy: 99.2% on DIARETDB0, 100% on HEI-MED, and 98.0% on Drishti-GS.
- Saliency maps indicated key features used by the DL algorithm for laterality differentiation.
Conclusions:
- RetiSort demonstrates high accuracy as an automated retinal image sorting system.
- The system can significantly aid in data preparation for deep learning research involving retinal photographs.
- RetiSort has practical applications for improving efficiency and accuracy in ophthalmic AI research.
