Referral system for hard exudates in eye fundus
Syed Ali Gohar Naqvi1, Muhammad Faisal Zafar1, Ihsan ul Haq1
1International Islamic University, Islamabad, Pakistan.
Computers in Biology and Medicine
|August 2, 2015
Summary
This study presents an automated system to detect hard exudates, a common cause of blindness in diabetic retinopathy. The system combines image analysis techniques for faster, more accurate diagnosis, aiding overloaded ophthalmologists.
Area of Science:
- Ophthalmology
- Medical Imaging
- Computer Vision
Background:
- Hard exudates are a prevalent sign of diabetic retinopathy, often leading to vision loss.
- Manual diagnosis by ophthalmologists is time-consuming and contributes to workload.
- Automated systems are needed to assist in the early detection and referral of diabetic retinopathy.
Purpose of the Study:
- To develop and evaluate an automated referral system for detecting hard exudates in retinal fundus images.
- To improve diagnostic efficiency and reduce the burden on ophthalmologists.
Main Methods:
- The system integrates Scale Invariant Feature Transform (SIFT), K-means Clustering, Visual Dictionaries, and Support Vector Machine (SVM).
- Performance was assessed using multiple fundus image databases, including a public one for comparison.
- The system was also tested with a Back Propagation Neural Network classifier.
Main Results:
- Achieved a maximum Area Under the Curve (AUC) of 97.02% and 95.02% accuracy on a single database.
- On mixed databases, the system recorded an AUC of 93.49% and 87.23% accuracy.
- Performance was found to be comparable or superior to existing systems.
Conclusions:
- The proposed automated system effectively detects hard exudates in fundus images.
- The system demonstrates robust performance across different image conditions and sources.
- This technology offers a valuable tool for early detection and management of diabetic retinopathy.
