A deep-learning model using ultra-widefield fundus autofluorescence (UW-FAF) images reveals that artificial intelligence may help accurately estimate the visual function of patients with retinitis pigmentosa.
Study design
This multicenter, retrospective study evaluated 1274 eyes of 695 consecutive patients with retinitis pigmentosa (RP). Ultra-widefield fundus autofluorescence and ultra-widefield pseudocolor (UWPC) images were utilized to train, validate, and test deep-learning models for the estimation of visual function. Estimations of mean deviation on Humphrey visual field analyzer, central retinal sensitivity, and BCVA were obtained, and the results were compared with actual values to determine accuracy.
Outcomes
Of the 3 image types evaluated (UW-FAF, UWPC, and UW-FAF/UWPC combination), the deep-learning model that used UW-FAF images alone provided the most accurate estimations for mean deviation, central sensitivity, and BCVA (standardized regression coefficient = 0.684, 0.697, 0.309, respectively).
Limitations
The data is limited by the retrospective nature of the report. Only eyes with classic findings of RP were included, so results may not be applicable to more atypical cases. Eyes with coexisting ocular conditions common in RP, such as macular edema or epiretinal membrane, were excluded. In addition, the study was conducted in Japan and thus is limited to patients of Japanese origin. As is the case with deep-learning models, what the model is actually using from the UW-FAF images to estimate visual function is unclear.
Clinical significance
UW-FAF images may be an easily accessible imaging modality to track visual function in patients with RP, a condition in which visual acuity may be poor and for which other testing, such as electroretinogram, may be more difficult to obtain. In addition, the study offers a use case for deep-learning models in ophthalmology, herein to determine which imaging modalities may be best suited for clinical monitoring.
Dr. M. Ali Khan discloses financial relationships with Allergan, Apellis Pharmaceuticals, Genentech (Consultant/Advisor); Regeneron Pharmaceuticals (Grant Support).