Automated Detection of Retinal Nerve Fiber Layer by Texture-Based Analysis for Glaucoma Evaluation

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma, a leading cause of blindness, involves damage to the retinal nerve fiber layer (RNFL).
• Early detection of RNFL changes is crucial for effective glaucoma management.
• Current detection methods may lack the precision needed for early-stage diagnosis.

Purpose of the Study:

• To propose and evaluate an automated method for detecting RNFL loss using texture features.
• To leverage co-occurrence matrices and a backpropagation neural network for RNFL detection.
• To improve the accuracy and efficiency of identifying glaucomatous optic neuropathy.

Main Methods:

• Extraction of two texture features: correlation and autocorrelation from co-occurrence matrices.
• Selection of relevant features using a correlation feature selection method.
• Classification of RNFL status (normal vs. loss) using a backpropagation neural network.

Main Results:

• The proposed method was evaluated on 40 retinal fundus images (160 sub-images for training).
• An overall accuracy of 94.52% was achieved in detecting RNFL abnormalities.
• The results indicate robust performance in distinguishing normal RNFL from RNFL loss.

Conclusions:

• The developed automated method demonstrates high accuracy in RNFL detection.
• This approach shows significant potential for early glaucoma diagnosis and monitoring.
• Texture analysis combined with neural networks offers a promising avenue for ophthalmic image analysis.