Evaluating the Efficiency and Precision of Bayesian Active Learning qReading in Low Vision.

Purpose: A novel Bayesian active learning method, the qReading method, was developed to measure reading speed versus print size function. Using a Bayesian adaptive procedure that selects stimuli to optimize information gain, the method has demonstrated high efficiency and precision in assessing individuals with normal vision.1,2 This study examines its efficiency and precision in patients with visual impairments. Methods: Reading functions of 20 visually impaired observers were measured. Each observer read 60 sentences at varying speeds and print sizes, determined by the qReading method. The area under the reading curve (AUC) quantified overall reading performance. The 68.2% half width of the credible interval (HWCI) of the posterior distribution of reading speed gauged precision. Results: AUC exhibited high variation initially and leveled off within 10 trials, with the final AUC value ranging between 1.75 and 2.51 across subjects. Estimation error decreased to 5% within 14 trials. Root mean square error dropped below 0.05 log10 units after 25 trials. HWCI of reading speed decreased to 0.05 log10 units within 18 trials. Whereas 60 trials took 11 minutes, fewer (e.g. 20) trials may suffice for precise estimates. Conclusions: The qReading method demonstrated outstanding precision and efficiency in visually impaired individuals, and may serve as a valuable tool in research and clinical assessment for low vision. Translational Relevance: The qReading method provides efficient and precise evaluation of reading performance in individuals with visual impairments, highlighting its potential clinical application for assessing reading function and monitoring interventions in low vision populations.