Unique corneal tomography features of allergic eye disease identified by OCT imaging and artificial intelligence.

The purpose of this study was to assess unique corneal tomographic parameters of allergic eye disease (AED) using optical coherence tomography (OCT) and artificial intelligence (AI). A total of 57 eyes diagnosed with AED were included. The curvature and aberrations of the air-epithelium (A-E) and epithelium-Bowman's layer (E-B) interfaces were calculated. Random forest AI models were built combing this data with the parameters of healthy, forme fruste keratoconus (FFKC) and KC eyes. The AI models were cross-validated with 3-fold random sampling. Each model was limited to 10 trees. The AI model incorporating both A-E and E-B parameters provided the best classification of AED eyes (area under the curve = 0.958, sensitivity = 80.7%, specificity = 98.5%, precision = 88.2%). Further, the E-B interface parameters provided the highest information gain in the AI model. A few AED eyes (n = 9) had tomography parameters similar to FFKC and KC eyes and may be at risk of progression to KC.