Glaucoma Detection Using Support Vector Machine Method Based on Spectralis OCT.

Area of Science:

• Ophthalmology
• Medical Imaging
• Machine Learning

Background:

• Spectralis optical coherence tomography (OCT) generates extensive data (114 features) for glaucoma assessment.
• Interpreting this large volume of information for accurate glaucoma diagnosis can be challenging.
• Machine learning (ML) offers a powerful approach to process complex datasets for disease detection.

Purpose of the Study:

• To evaluate the diagnostic performance of Spectralis OCT parameters in glaucoma detection using Support Vector Machine (SVM) classification.
• To assess the capability of ML models in discriminating between normal and glaucomatous eyes across different severity levels.
• To identify key OCT parameters for improved glaucoma diagnosis in the Taiwanese population.

Main Methods:

• Utilized Support Vector Machine (SVM) classification to analyze 114 Spectralis OCT features.
• Applied the SVM model to differentiate healthy eyes from glaucomatous eyes.
• Evaluated the model's performance using Area Under the Curve (AUC) metrics for overall detection and severity discrimination.

Main Results:

• The SVM model using all OCT features achieved an AUC of 0.82 for detecting glaucomatous eyes.
• The model effectively discriminated between normal and early, moderate, or severe glaucomatous eyes with AUCs of 0.78, 0.89, and 0.93, respectively.
• A combination of specific peripapillary and macular parameters (including MRW and GCL) yielded an AUC of 0.84.

Conclusions:

• Spectralis OCT parameters, when analyzed with SVM, demonstrate significant potential for glaucoma detection and management.
• Specific OCT features, such as Minimum Rim Width (MRW) and Ganglion Cell Layer (GCL) measurements, are valuable for early glaucoma diagnosis.
• The developed ML model shows promise for glaucoma assessment in the Taiwanese population, warranting further validation.