AN IMAGE MODEL FOR QUANTAL RESPONSE ANALYSIS IN PERIMETRY

Measurement of the patient's ''seeing thresholds'' at different locati ons in the visual field is an important diagnostic tool for glaucoma a nd other eye diseases. A Markov random field model is developed and us ed for efficient estimation of the thresholds and for classification o f points as ''(normal'' or ''defective''. The model allows for non-hom ogeneous spatial dependence and non-symmetric marginal distributions a nd has physically interpretable parameters. ICM threshold estimation r esulted in 10-30% (depending on the patient population) reduction of m ean square error as compared to currently used procedures and in a fai r agreement between true and estimated defect status. Marginal posteri or mean estimates had the same efficiency, but required more computati on time. Non-standard features of the problem are: (i) a non-homogeneo us directional dependence, (ii) thresholds are only measured indirectl y, by binary responses to questions, where the probability of response depends on the threshold and the stimulus level.