Relevance Vector Machine for Combining OCT and Standard Automated Perimetry Results for Discriminating Between Healthy and Glaucoma Eyes

Abstract

 

Purpose:

 

To determine the diagnostic accuracy for classifying healthy and glaucomatous eyes using relevance vector machine (RVM) learning classifiers trained on retinal nerve fiber layer (RNFL) thickness measurements obtained by optical coherence tomography (OCT) and standard automated perimetry (SAP), independently and combined.

 

Methods:

 

222 subjects from the UCSD Diagnostic Innovations in Glaucoma Study (DIGS) were included. 66 eyes of 66 healthy controls and 156 eyes of 156 glaucoma patients with glaucomatous appearing optic discs and/or repeatable abnormal SAP results were imaged using Stratus OCT and tested using SAP (Humphrey Field Analyzer II with SITA) within three months of each other. RVMs were trained and tested on OCT– determined RNFL thickness measurements from 32 equal sectors obtained in the circumpapillary area under the instrument defined measurement ellipse and SAP threshold values from 52 points from the 24–2 grid plus age, independently and in combination. Ten–fold cross–validation was used to train and test RVM classifiers on unique subsets of the full 222–eye data set, and areas under the receiver operating characteristic curve (AUROC) for the classification of eyes in the test set were generated for full dimension and optimized input data sets. AUROC results from RVM trained on OCT and SAP alone and those from RVM trained on OCT and SAP in combination were compared. In addition these results were compared to currently available OCT measurements and SAP indices.

 

Results:

 

Results are presented in the Table below.

 

Conclusions:

 

RVM did not improve the classification performance of OCT or SAP compared to standard parameters. OCT results support previous results showing that combining OCT RNFL parameters using statistical methods does not improve performance. However, the probabilistic output generated by RVM may be clinically useful. When OCT and SAP data were combined and optimized using RVM, classification was improved slightly.  

 

Keywords: imaging/image analysis: clinical • visual fields • nerve fiber layer