Contrast sensitivity function enhances the prediction of treatment outcome and recurrence in amblyopia via machine learning

Abstract

Purpose : Amblyopia is the most common cause of unilateral visual impairment in children. Although effective treatments are available, treatment outcomes are variable, and the condition recurs in up to 25% of the patients. Here we developed a machine learning analysis using artificial intelligence techniques to explore associations in visual function parameters for patients with amblyopia. We hypothesized that machine learning could predict amblyopia treatment outcomes at 3- and 6 months post-diagnosis and identify patients at higher risk of recurrence within one year of successful treatment.

Methods : A total of 643 patients were eligible for modelling: 434 for treatment response prediction and 209 for recurrence prediction. For the treatment response model, we used retrospective baseline data to predict treatment response (visual acuity improved by 1 line, 1.5 lines and 2 lines, respectively) at 3 and 6 months. For the recurrence model, we used retrospective endpoint data to predict recurrence within one year of successful treatment. Functional vision measures from vision acuity (VA) and contrast sensitivity function (CSF) assessments were used as predictive features in the models. The predictive accuracy of the predictive features individually and in combination was evaluated. The performance of the predictive models was evaluated using the area under the receiver operating characteristic curve (AUC).

Results : In the treatment response prediction model, combining features from VA and CSF assessments gave the highest accuracy, with AUCs of 0.863 and 0.815, 0.769 and 0.743, 0.734 and 0.689 for outcome predictions after 3 and 6 months of treatment under the improved 1-,1.5- and 2-line criteria, respectively. In the recurrence prediction model, features from the CSF assessment gave rise to an AUC of 0.909 compared to 0.539 for VA. Combining VA and CSF features did not improve model performance. While worse initial visual functions led to better treatment outcomes, unbalanced binocular contrast sensitivity after successful treatment led to a higher amblyopia recurrence.

Conclusions : CSF data enhanced the accuracy of treatment response prediction models and were sufficient for the prediction of amblyopia recurrence in their own right. The ability to accurately predict amblyopia treatment response and recurrence has the potential to improve amblyopia management.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.