Fig. 3: Performance of the AI-assisted readout method.

A Treating physicians were asked to evaluate the measurement by the AI algorithm and adjust the measurement in case of need. The proportion of wheals that required adjustment, as well as the proportion of test results that shifted from positive to negative or vice versa, were analyzed in the test cohort (n = 95 patients and 1140 wheals). B The time to perform readout of the test by SPAT, with or without AI-assisted readout, is represented (n = 100 patients randomly selected from the validation cohort). Data are presented as Tukey box and whiskers, including median with interquartile range. Repeated measures ANOVA and Sidak’s multiple comparison test were performed to compare AI-assisted readout to manual readout.