Automated interpretation of optic nerve images: a data mining framework for glaucoma diagnostic support

Area of Science:

• Ophthalmology
• Medical Imaging
• Computational Biology

Background:

• Confocal Scanning Laser Tomography (CSLT) provides high-resolution optic disc images crucial for glaucoma diagnosis and monitoring.
• Interpreting CSLT images for glaucoma can be challenging, necessitating advanced analytical tools.
• Automated analysis can improve the accuracy and efficiency of glaucoma assessment.

Purpose of the Study:

• To develop and evaluate a hybrid framework integrating image processing and data mining for CSLT optic nerve image analysis.
• To support the automated diagnosis and monitoring of glaucoma by analyzing optic disc shape characteristics.
• To visualize sub-types of glaucomatous optic disc damage.

Main Methods:

• Utilized Zernike moments for extracting shape information from optic disc images.
• Employed machine learning classifiers including Multi Layer Perceptrons, Support Vector Machines, and Bayesian Networks for feature selection and classification.
• Applied Self-Organizing Maps for clustering optic disc images to identify glaucomatous damage subtypes.

Main Results:

• The framework successfully derived shape information from CSLT images.
• Machine learning models demonstrated capability in distinguishing between healthy and glaucomatous optic discs.
• Clustering revealed distinct patterns of glaucomatous optic disc damage.

Conclusions:

• The proposed hybrid framework offers an automated and objective approach to analyzing CSLT optic nerve images.
• This method has the potential to significantly aid in both the diagnosis and ongoing monitoring of glaucoma.
• Further development could enhance clinical decision-making in ophthalmology.