Predicting glaucomatous progression in glaucoma suspect eyes using relevance vector machine classifiers for combined...

Area of Science:

• Ophthalmology
• Medical Imaging
• Machine Learning in Healthcare

Background:

• Glaucoma is a leading cause of irreversible blindness.
• Early detection of glaucomatous progression is crucial for timely intervention.
• Confocal scanning laser ophthalmoscopy (CSLO) and standard automated perimetry (SAP) are key diagnostic tools.

Purpose of the Study:

• To evaluate the predictive capability of baseline CSLO and SAP measurements for future glaucomatous progression.
• To assess the performance of relevance vector machine (RVM) classifiers in analyzing these datasets.
• To compare the accuracy of RVM analysis with established CSLO and SAP global indices.

Main Methods:

• A cohort of 264 eyes from 193 participants with normal baseline SAP results was analyzed.
• Eyes were classified as progressed or stable based on SAP Guided Progression Analysis or stereophotograph assessment.
• Baseline CSLO topographic parameters and SAP total deviation values were used to train and test RVMs with ten-fold cross-validation.

Main Results:

• RVM analysis of combined CSLO and SAP features achieved an area under the ROC curve (AUROC) of 0.805.
• RVMs trained on CSLO and SAP parameters alone yielded AUROCs of 0.640 and 0.762, respectively.
• Traditional CSLO (Glaucoma Probability Score) and SAP (mean deviation, pattern standard deviation) global indices showed limited discriminatory ability (AUROCs ≤ 0.620).

Conclusions:

• RVM analysis of baseline CSLO and SAP measurements demonstrates superior accuracy in predicting future glaucomatous progression compared to global indices.
• This machine learning approach holds promise for improving early detection and management of glaucoma in suspect eyes.
• Further validation in larger, diverse cohorts is warranted.