Machine learning models for objective evaluation of visual acuity based on pattern-reversal visual evoked potentials

Abstract

Abstract Background To explore the development of machine learning models for objective evaluation of visual acuity based on pattern-reversal visual evoked potentials (PRVEPs) as a reference for improving medical selection and identification methods for aircrew visual function. Methods Twenty-four volunteers were recruited and forty-eight eyes were divided into four groups of 1.0, 0.8, 0.6, and 0.4 (decimal vision) according to visual acuity. The relationship between visual acuity, peak time, or amplitude of P100 that was recorded at 5.7°, 2.6°, 1°, 34′, 15′, and 7′ visual angles was analyzed using repeated-measures analysis of variance. Receiver operating characteristic (ROC) curves were used to compare the effectiveness of the P100 peak time and amplitude as objective indicators of visual acuity at different viewing angles. Based on meaningful P100 peak time and amplitude, four machine learning algorithms were used to construct objective assessment models for visual acuity, and the efficacy of different models was compared by repeated sampling comparisons and ten-fold cross-validation. Results The main effects of P100 peak time and amplitude between different visual acuity and various viewing angles were statistically significant (P<0.05). Visual acuity in general was negatively correlated with peak time and positively correlated with amplitude. The peak time initially shortened with increasing examination angle and gradually lengthened after the minimum value was reached at 1°. At the 1° visual angle, there were statistically significant differences when comparing the peak times between the vision groups with each other (P<0.05), and the amplitudes of the vision reduction groups were significantly lower than that of the 1.0 vision group (P<0.05). The correlations between peak time, amplitude, and visual acuity were all highest at 1° (r = 0.740, 0.438). For different binarization classifications of visual acuity, the maximum area under the ROC curves (AUC) of the peak time or amplitude at all visual angles was higher than 0.8 (P < 0.001). Among the objective assessment models of visual acuity constructed by different machine learning algorithms, multilayer perceptron (MLP) had the best performance in distinguishing 1.0 from 0.8, 0.6, and 0.4 visual acuities, with an accuracy of 93.75% for ten-fold cross-validation. Conclusions The P100 peak time and amplitude obtained at multiple views of PRVEPs are correlated with visual acuity; machine learning models based on this can be used to assist in the objective evaluation of visual acuity.