Hybrid multi-kernel SVM algorithm for detection of microaneurysm in color fundus images

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Science

Background:

• Diabetic retinopathy (DR) is a leading cause of vision loss in diabetic patients.
• Early detection of DR, particularly microaneurysms, is crucial for timely intervention.
• Current screening methods may benefit from more efficient automated systems.

Purpose of the Study:

• To develop an automatic screening system for early diabetic retinopathy detection.
• To utilize a combination of shape and texture features for reduced computational burden.
• To evaluate a novel hybrid multi-kernel support vector machine classifier.

Main Methods:

• An automatic screening system was developed using combined shape and texture features from retinal images.
• A hybrid multi-kernel support vector machine classifier was constructed by combining base kernels.
• The system was validated on public datasets (Retinopathy Online Challenge, DIARETdB1, MESSIDOR) and a newly developed AGAR300 dataset.

Main Results:

• The proposed system achieved high performance, with FROC scores of 0.503 (ROC), 0.481 (DIARETdB1), and 0.464 (MESSIDOR).
• On the AGAR300 dataset, the system outperformed existing microaneurysm detection algorithms in FROC, AUC, F1 score, precision, sensitivity, and specificity.
• The hybrid multi-kernel SVM approach demonstrated superior performance compared to recent deep learning techniques.

Conclusions:

• The developed automatic screening system is effective for early diabetic retinopathy detection.
• The hybrid multi-kernel SVM classifier offers a computationally efficient and robust solution.
• The system's validated performance across multiple datasets ensures its reliability for clinical application.