A Hybrid PSO-DEFS Based Feature Selection for the Identification of Diabetic Retinopathy

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic Retinopathy (DR) is a significant cause of blindness, often linked to uncontrolled blood insulin levels.
• Diabetic Macular Edema (DME) is the primary cause of visual impairment in diabetic patients.
• Optical Coherence Tomography (OCT) is crucial for diagnosing and monitoring DME, but existing methods face accuracy and time complexity challenges.

Purpose of the Study:

• To develop a more accurate and efficient hybrid method for detecting Diabetic Retinopathy (DR) and Diabetic Macular Edema (DME).
• To overcome the limitations of existing DME detection techniques, specifically low accuracy and high computational cost.

Main Methods:

• Image preprocessing involved green channel extraction and median filtering.
• Feature extraction utilized Histogram of Oriented Gradient (HOG) and Complete Local Binary Pattern (CLBP) for gradient and texture analysis.
• A hybrid feature selection combining Particle Swarm Optimization (PSO) and Differential Evolution Feature Selection (DEFS) was employed to reduce time complexity.
• Classification was performed using a binary Support Vector Machine (SVM) followed by Multi-Layer Perceptron (MLP) for final DR patient categorization.

Main Results:

• The proposed hybrid approach demonstrated superior performance compared to existing methods.
• Key performance metrics including accuracy, sensitivity, and specificity were significantly improved.
• The combined PSO and DEFS feature selection effectively minimized processing time.

Conclusions:

• The developed hybrid DR detection method offers enhanced accuracy and efficiency for diagnosing DME.
• This approach provides a promising tool for clinical assessment and management of diabetic eye disease.
• The integration of advanced feature extraction and optimization techniques leads to improved diagnostic outcomes.