Abstract
Purpose :
Visual field testing using standard thresholding procedures is time-consuming, and prolonged test duration leads to patient fatigue and decreased test reliability. Different visual field testing algorithms have been developed to shorten testing time while maintaining accuracy. However, the performance of these algorithms depends heavily on prior knowledge and manually crafted rules that determine the intensity of each light stimulus as well as the termination criteria, which is suboptimal. We leveraged deep reinforcement learning (DRL) to find better decision strategies for visual field tests, thereby reducing test duration while retaining testing accuracy compared to conventional visual field testing algorithms.
Methods :
In our algorithm, the procedure of white-on-white static automated perimetry (SAP) is formulated as an extensive-form game. Multiple intelligent agents are employed to interact with the patient, with each agent controlling the test on one of the testing locations in the patient’s visual field. Through training, each agent learns an optimized policy that determines the intensities of light stimuli and the termination criteria at different steps, which minimizes the error in sensitivity estimation and test duration at the same time. In simulation experiments, we compare the performance of our algorithm against two conventional visual field testing algorithms, Zippy Estimation for Sequential Testing (ZEST) and Spatially Weighted Likelihoods in Zippy Estimation by Sequential Testing (SWeLZ), on our visual field dataset with 504 normal and 758 glaucomatous visual fields.
Results :
Experimental results show that, quantitatively, for the same error level, our algorithm is 34.0% and 25.2% faster than ZEST for normal and glaucomatous visual fields, and 11.5% and 10.3% faster than SWeLZ for normal and glaucomatous visual fields for a similar error level.
Conclusions :
The policies learned by our algorithm can achieve a better trade-off between estimation accuracy and testing time compared to conventional visual field testing algorithms. By retaining testing accuracy with reduced test duration, our algorithm increase test reliability, increase clinic efficiency with shorter testing times, improve patient satisfaction with decreased patient exhaustion and increased comfort, and translationally affect clinical outcomes.
This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.