Artificial Intelligence (AI)-assisted readout method for the evaluation of skin prick automated test results

Background: The skin prick test (SPT) is the gold standard for diagnosing allergic sensitization to aeroallergies. A novel device, Skin
Prick Automated Test (SPAT), has previously demonstrated reduced variability and more consistent test results compared to
manual SPT. The current study aimed to develop and validate an artificial intelligence (AI) assisted readout method to support
physicians in interpreting skin reactions following SPAT. Methods: 963 patients with suspected aeroallergies underwent SPT using
SPAT for ten common allergens. To train and validate the AI algorithm, respectively 7812 (651 patients, 75%) and 2604 (217
patients, 25%) wheals were manually annotated by a person blinded to the outcome of the AI. The longest wheal diameter was
measured by the treating physician and compared to the AI measurement. The AI-assisted readout was validated on a separate
test cohort of 95 patients (1140 wheals). Results: The AI measurements of the longest wheal diameter exhibited a strong
correlation with the physician’s measurements. The AI algorithm showed a specificity of 98.4% and sensitivity of 85·0% in
determining positive or negative test results in the validation cohort. In the test cohort, physicians adjusted 5·8% of AI
measurements, leading to a change in the test interpretation for only 0·5% of cases. AI-assisted readout significantly reduced
inter- and intra-observer variability and readout time compared to manual physician measurements.Conclusion: The AI-assisted
readout software demonstrated high accuracy, with minimal misclassification of test results. Adding AI to SPAT further improved
standardization across the SPT process, significantly reducing observer variability and time to readout.