Predicting Glaucomatous Progression from Baseline HRT and SAP Measurements Using Relevance Vector Machine (RVM) Classifiers

Abstract

 

Purpose:

 

To determine if future glaucomatous progression can be predicted from baseline HRT and SAP measurements analyzed using RVM classifiers.

 

Methods:

 

271 eyes of 196 participants in the Diagnostic Innovations in Glaucoma Study (DIGS) or the African Descent and Glaucoma Evaluation Study (ADAGES) were included. Baseline HRT, SAP and optic disc stereophotographs were required within a six-month window. All eyes had normal SAP results at baseline and had a minimum of 5 SAP tests over time. Eyes were labeled as progressed (n=48) or non-progressed (n=223) during follow-up based on SAP Guided Progression Analysis "likely progression" or masked assessment of serial stereophotographs. Non-progressed eyes were stable by both assessment methods for a minimum of 4 years. Baseline HRT-measured topographic parameters (n=117) and baseline total deviation (TD) values from the 24-2 SAP test-grid (n=52) were selected from each eye. Ten-fold cross-validation was used to train and test RVM classifiers using the HRT and SAP features from unique subsets of the 271 study eyes. Areas under the receiver operating characteristic curve (AUROC), representing the average prediction accuracies, were calculated using full and optimized (using backward elimination) feature sets. AUROC results from HRT, SAP and HRT and SAP combined were compared to currently available HRT (GPS) and SAP (MD, PSD) global indices.

 

Results:

 

Results are presented in the table below.

 

Conclusions:

 

RVM analyses of baseline HRT and SAP measurements can predict eyes that will show glaucomatous progression with a higher accuracy than the HRT and SAP global indices investigated. In this study, HRT GPS and SAP MD did not predict progression any better than chance.  

 

Clinical Trial:

 

http://www.clinicaltrials.gov NCT00221897

 

Keywords: imaging/image analysis: clinical • visual fields