A back propagation neural network for the classification of visual field data

Publisher Website
Google Scholar
Abstract
The computer assisted moving eye campimeter (CAMEC) is a visual field analyser that uses a moving target to hold the patient's fixation while the stimuli are presented. A neural network has been used to classify visual field data in the CAMEC format. This has the potential to provide self-testing diagnostic visual field examinations without the presence of skilled clinical personnel. A three-layer back propagation network was designed, with 110 units in the input layer, each unit corresponding to a test point on the CAMEC grid, a hidden layer of 40 units and an output layer of 22 units, each one corresponding to a particular type of visual field defect. Four hundred and ninety simulated field plots were generated in the CAMEC format by experienced ophthalmologists. These data were split into a training sets of 440 plots and a test set of 50 plots. Training of the neural network was accomplished by cycling the training set up to 50,000 times through the network. The effect of changing the number of hidden units was investigated using the test set of field plots. Optimizing the network configuration results in a classification accuracy of 98.2% for the training set of data and 96% for the test set.

This publication has 2 references indexed in Scilit: