Diagnostic support for glaucoma using retinal images: a hybrid image analysis and data mining approach

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Science

Background:

• Glaucoma diagnosis relies on objective methods for early detection.
• Modern imaging like Confocal Scanning Laser Tomography (CSLT) provides high-resolution optic nerve data.
• Automated analysis of CSLT images can improve glaucoma detection efficiency.

Purpose of the Study:

• To develop an automated system for glaucoma diagnosis using CSLT images.
• To investigate feature selection methods for optimizing classification accuracy.
• To evaluate the efficacy of moment methods and machine learning in distinguishing healthy from glaucomatous optic disks.

Main Methods:

• Utilized Confocal Scanning Laser Tomography (CSLT) for optic nerve imaging.
• Applied moment methods to extract image-defining features from CSLT data.
• Employed feature subset selection techniques to reduce the input feature space.
• Trained classifiers, including neural networks and support vector machines, for automated classification.

Main Results:

• Identified a subset of moment features that yield high classification accuracy.
• Demonstrated the ability to discriminate between healthy and glaucomatous optic disks.
• Successfully automated the analysis of optic disk topography and reflectance images.

Conclusions:

• The hybrid approach effectively uses CSLT image features for glaucoma detection.
• Feature selection is crucial for optimizing automated diagnostic systems.
• This method shows promise for objective and early glaucoma diagnosis.