Visual field interpretation with a personal computer based neural network

Abstract

The Computer Assisted Touch Screen (CATS) and Computer Assisted Moving Eye Campimeter (CAMEC) are personal computer (PC)-based video-campimeters which employ multiple and single static stimuli on a cathode ray tube respectively. Clinical studies show that CATS and CAMEC provide comparable results to more expensive conventional visual field test devices. A neural network has been designed to classify visual field data from PC-based video-campimeters to facilitate diagnostic interpretation of visual field test results by non-experts. A three-layer back propagation network was designed, with 110 units in the input layer (each unit corresponding to a test point on the visual field test grid), a hidden layer of 40 processing units, and an output layer of 27 units (each one corresponding to a particular type of visual field pattern). The network was trained by a training set of 540 simulated visual field test result patterns, including normal, glaucomatous and neuro-ophthalmic defects, for up to 20,000 cycles. The classification accuracy of the network was initially measured with a previously unseen test set of 135 simulated fields and further tested with a genuine test result set of 100 neurological and 200 glaucomatous fields. A classification accuracy of 91-97% with simulated field results and 65-100% with genuine field results were achieved. This suggests that neural networks incorporated into PC-based video-campimeters may enable correct interpretation of results in non-specialist clinics or in the community.