TORONTO: A trial-oriented multidimensional psychometric testing algorithm.

Area of Science:

• Ophthalmology
• Computational Neuroscience
• Psychophysics

Background:

• Traditional Bayesian adaptive methods for sensory threshold determination focus on single thresholds.
• Existing visual field testing methods do not leverage spatial patterns for improved efficiency.
• Exploiting spatial patterns is crucial for enhancing visual field test efficiency.

Purpose of the Study:

• To introduce TORONTO, a novel Bayesian adaptive method for simultaneous multi-threshold determination.
• To evaluate TORONTO's performance against existing algorithms in terms of speed and accuracy.
• To generalize Bayesian adaptive methods to exploit spatial patterns in visual field testing.

Main Methods:

• TORONTO generalizes the QUEST/ZEST algorithm to estimate multiple thresholds concurrently.
• It employs a trial-oriented approach, updating all tested locations after each trial using reference data patterns.
• Techniques are developed to address limitations in reference data availability.

Main Results:

• TORONTO demonstrated superior speed and accuracy in computer-simulated visual field tests across various reliability conditions (FP=FN=3%, 15%, 30%).
• In reliable conditions (3%), TORONTO achieved median termination in 153 trials with 2.0 dB RMSE, twice as fast as ZEST with equal accuracy.
• Under higher false positive/negative rates (15% and 30%), TORONTO consistently outperformed ZEST in speed (2.2x faster at 15%) and accuracy (better RMSE).

Conclusions:

• TORONTO is a highly efficient and accurate algorithm for determining multiple sensory thresholds, particularly in visual field testing.
• The method effectively utilizes spatial patterns to accelerate threshold determination.
• TORONTO offers significant advantages over existing methods, especially under varying subject reliability conditions.