AUTOMATIC DETECTION OF GLAUCOMATOUS VISUAL-FIELD PROGRESSION WITH NEURAL NETWORKS

Objective: To evaluate computerized neural networks to determine visua l field progression in patients with glaucoma. Methods: Two hundred th irty-three series of Octopus G1 visual fields of 181 patients with gla ucoma were collected. Each series was composed of 4 or more reliable v isual fields from patients who had previously undergone automated peri metry. The visual fields were independently evaluated in a masked fash ion by 3 experienced observers (K.N.-M, M.W., and J.C.) and were judge d to show progression based on the agreement of 2 observers. The stabl e and progressed series were matched for mean defect at baseline. The threshold data were submitted to a back propagation neural network tha t was trained to classify each series as stable or progressed. Two thi rds of the data were used for the training and the remaining one third to test the performance of the network. This was repeated 3 times to classify all of the series (changing the training and test series). Re sults: Fifty-nine series of visual fields showed progression and 151 w ere judged stable. Neural network sensitivity was 73% and specificity was 88% (threshold for progression = 0.5). The concordance of the neur al network with the observers was good (0.50 less than or equal to kap pa greater than or equal to 0.64). Conclusions: A neural network can b e trained to recognize visual field progression in good concordance wi th experienced observers. Neural networks may be used to aid the physi cian in the evaluation of glaucomatous visual field progression.