Morphometric analysis and classification of glaucomatous optic neuropathy using radial polynomials.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computational Biology

Background:

• Glaucomatous optic neuropathy diagnosis relies on accurate assessment of optic nerve head (ONH) morphology.
• Traditional methods for ONH analysis may have limitations in sensitivity and specificity for early glaucoma detection.
• Novel quantitative approaches are needed to improve the classification of glaucomatous optic neuropathy.

Purpose of the Study:

• To quantify optic nerve head (ONH) morphology using radial polynomials.
• To develop an automated decision tree algorithm for classifying glaucomatous optic neuropathy based on morphometric models.
• To compare the performance of this novel classification method against established procedures.

Main Methods:

• Patients with ocular hypertension or early glaucoma (n=179) and normal subjects (n=96) were assessed.
• Optic nerve head morphology was modeled using pseudo-Zernike radial polynomials from confocal scanning laser tomography data.
• A decision tree induction algorithm classified glaucomatous optic neuropathy; results were compared with expert assessment and Moorfields Regression Analysis (MRA).

Main Results:

• Morphometric models showed decreasing error with increased polynomial coefficients.
• Optimal classification using 64 features achieved 80% accuracy, 69% sensitivity, 88% specificity, and an 88% AUROC.
• Moorfields Regression Analysis (MRA) achieved 78% accuracy, 55% sensitivity, 95% specificity, and an 83% AUROC.

Conclusions:

• Pseudo-Zernike radial polynomials offer a compact and accurate representation of ONH morphology.
• The proposed morphometric classification method demonstrates superior sensitivity and comparable overall performance (AUROC) to MRA.
• This automated approach holds promise for improved early detection and classification of glaucomatous optic neuropathy.