Retinal image analysis for disease screening through local tetra patterns.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Age-related Macular Degeneration (AMD) and Diabetic Retinopathy (DR) are leading causes of global vision impairment.
• Accurate and early detection of these retinal diseases is crucial for effective treatment and vision preservation.

Purpose of the Study:

• To investigate the potential of texture analysis in color fundus images for discriminating between healthy and diseased retinas (AMD, DR).
• To evaluate the efficacy of Local Tetra Patterns (LTrP) as a texture descriptor for retinal disease classification.
• To develop an automated screening approach for retinal diseases based on texture characteristics.

Main Methods:

• A novel retinal background characterization approach using Local Tetra Patterns (LTrP) was developed.
• Texture classification was performed on AMD, DR, and normal retinal images across five experimental setups.
• Multiple classifiers (AdaBoost, c4.5, logistic regression, naive Bayes, neural network, random forest, support vector machine) were tested on public datasets (ARIA, STARE, E-Optha).
• LTrP performance was compared against other texture descriptors like local phase quantization, local binary pattern, and local derivative pattern.

Main Results:

• The LTrP method achieved Area Under the ROC Curve (AUC) and f-score values consistently above 0.78 and 0.746, respectively.
• Exceptional performance was observed with AUC and f-score values exceeding 0.995 for DR and AMD detection when using a random forest classifier.
• The proposed texture analysis technique demonstrated high discrimination potential for retinal diseases.

Conclusions:

• Local Tetra Patterns (LTrP) effectively utilize retinal image texture for discriminating between healthy and diseased states (AMD, DR).
• The developed method shows significant promise as a component of automated screening systems for early detection of retinal diseases.
• Texture-based analysis offers a viable alternative to lesion segmentation for retinal disease identification.