AI-assisted readout of automated skin prick testing in children

In this video, Sven Seys presents real-world paediatric data evaluating the performance of an AI-assisted readout method for the Skin Prick Automated Test device. The system standardises skin prick testing by applying 12 simultaneous pricks, followed by imaging of the skin reactions after 15 minutes. These images are analysed by an AI algorithm, which detects wheal reactions and measures the longest wheal diameter. Physicians can subsequently review, validate or adjust the AI-generated measurements.

The analysis included more than 27,000 pricks from over 2,300 children across multiple European hospitals and practices. Overall, AI measurements showed a strong correlation with physician-validated measurements. Physicians adjusted a minority of AI measurements, and only 2.7% of all pricks resulted in a change in clinical interpretation. Performance remained high across age groups, although misclassification was slightly more frequent in younger children (2-4 years and 5-11 years) compared with older children (12-17 years).

These findings confirm that AI-assisted readout of automated skin prick testing performs well in routine clinical practice, correctly classifying more than 97% of paediatric test results without requiring physician intervention. Further analyses are planned to explore factors that may explain the observed differences between younger and older children.