Vessel Density Features of Optical Coherence Tomography Angiography for Classification of Glaucoma Using Machine...

Area of Science:

• Ophthalmology
• Medical Imaging
• Machine Learning

Background:

• Glaucoma detection relies on identifying characteristic changes in the optic nerve and retinal nerve fiber layer.
• Optical coherence tomography angiography (OCTA) visualizes retinal vasculature, offering potential biomarkers for glaucoma.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) algorithms in classifying glaucoma using peripapillary vessel density features derived from OCTA.
• To compare the performance of different ML classifiers (SVM, Random Forest, Gaussian Naive Bayes) for glaucoma detection.

Main Methods:

• Four peripapillary vessel density features were extracted from OCTA images using threshold-based segmentation.
• These features were used to train and test three ML models: Support Vector Machine (SVM), Random Forest, and Gaussian Naive Bayes.
• Performance was assessed using the area under the receiver operating characteristic curve (AUC) and accuracy, with data split into 70% training and 30% testing sets.

Main Results:

• Glaucomatous eyes exhibited reduced vessel densities across various thresholds compared to healthy eyes.
• The SVM classifier achieved an AUC of 1 and accuracy of 1 at 70% and 100% thresholds for differentiating glaucoma.
• The Random Forest classifier demonstrated strong performance with an AUC of 0.993 and accuracy of 0.994 at 100% threshold.

Conclusions:

• ML models integrating peripapillary vessel density features from OCTA show excellent performance for glaucoma detection.
• The combination of total vessel and capillary density features in both whole and ring images is effective for diagnosing glaucoma.