CLASSIFYING VISUAL-FIELD DATA

We develop a prediction model for classifying an eye according to glau coma status based on its visual field. We develop measures of both dif fuse and localized defects in the visual field as potential predictors of glaucoma. To identify predictors of abnormal fields, we must descr ibe the variability in the fields of normal eyes, hence we first model the mean, variance and correlation structures of normal fields with u se of generalized estimating equations. The best measures of diffuse l oss include a field's mean level, contrasts of the upper and lower hal ves, and contrasts of the nasal and temporal halves, Local loss is mea sured by the depth, area, volume and location of the field's largest d efect. We develop logistic regression models to classify eyes as havin g glaucoma or not. We present ROC curves of the results that are highl y competitive with current clinical methods of classification.