Retinal blood vessel segmentation with neural network by using gray-level co-occurrence matrix-based features.

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Accurate extraction of retinal vessels is vital for diagnosing various ocular diseases.
• Existing methods for retinal vessel segmentation face challenges in achieving high sensitivity and specificity.

Purpose of the Study:

• To propose a novel supervised approach for retinal vessel extraction.
• To leverage the gray-level co-occurrence matrix (GLCM) for enhanced feature extraction from retinal images.
• To improve the accuracy and efficiency of retinal vessel segmentation.

Main Methods:

• Utilizing the green channel of retinal images for optimal vessel-non-vessel contrast.
• Applying the gray-level co-occurrence matrix (GLCM) to capture spatial structural information.
• Employing a neural network (multilayer perceptron) trained with GLCM-derived features for classification.

Main Results:

• The proposed method demonstrated superior performance compared to existing techniques on the DRIVE and STARE datasets.
• Key performance metrics including sensitivity, specificity, area under the ROC curve, and accuracy were significantly improved.
• Experimental results confirm the high efficiency of the proposed algorithm.

Conclusions:

• The integration of GLCM feature extraction with neural network classification offers a highly effective method for retinal vessel segmentation.
• This approach provides a robust and accurate tool for automated analysis of retinal images.
• The findings suggest a significant advancement in automated retinal image analysis for clinical applications.