Validation of AI-based software for objectification of conjunctival provocation test

BACKGROUND: Provocation tests are widely used in allergology to objectively reveal patients’ sensitivity to specific allergens. The objective quantification of an allergic reaction is a crucial characteristic of these tests. Because of the absence of objective quantitative measurements, the conjunctival provocation test (CPT) is a less frequently used method despite its sensitivity and simplicity. OBJECTIVE: We developed a new artificial intelligence (AI)-based method, called AllergoEye, for quantitative evaluation of conjunctival allergic reactions and validated it in a clinical study. METHODS: AllergoEye was implemented as a 2-component system. The first component is based on an Android smartphone camera for screening and imaging the patient’s eye, and the second is personal computer–based for image analysis and quantification. For the validation of AllergoEye, an open-label, prospective, monocentric study was carried out on 41 patients. Standardized CPT was performed with sequential titration of grass allergens in 4 dilutions, with the reaction evaluated by subjective/qualitative symptom scores and by quantitative AllergoEye scores. RESULTS: AllergoEye demonstrated high sensitivity (98%) and specificity (90%) as compared with human estimation of allergic reaction. Tuning cutoff thresholds allowed us to increase the specificity of AllergoEye to 97%, at which point the correlation between detected sensitivity to allergen and specific IgE carrier–polymer system class becomes obvious. Strikingly, such correlation was not found with sensitivity to allergen detected on the basis of subjective and qualitative symptom scores. CONCLUSION: The clinical validation demonstrated that AllergoEye is a sensitive and efficient instrument for objective measurement of allergic reactions in CPT for clinical studies as well as for routine therapy control.