Machine Learning-Based Evaluation of Skin Prick Tests using Photoplethysmography Imaging
Year: 2026
Type: conference-paper
Abstract: Skin prick testing (SPT) is widely used to diagnose immediate allergic hypersensitivity, but conventional visual assessment may be affected by interobserver variability and neighboring reactions. This study investigated photoplethysmography imaging (PPGI) combined with machine learning for the objective classification of SPT reactions. Skin perfusion was monitored in 18 subjects before allergen application and at 1, 5, 10, 15, and 20 minutes afterwards. Following motion compensation and signal filtering, perfusion parameters were calculated for 324 samples, including 134 positive and 190 negative reactions. Individual parameters were evaluated using receiver operating characteristic analysis, while their combination was assessed using a support vector machine (SVM) with subject-wise cross-validation. Discriminatory performance improved over time, with individual parameters achieving area under the curve (AUC) values above 0.90 at later measurement points. Excluding spatially influenced samples increased the maximum AUC to approximately 0.96. Using the complete dataset, the SVM achieved strong discrimination from 5 minutes onwards. Its best performance occurred at 10 minutes, with an AUC of 0.937, accuracy of 0.867, sensitivity of 0.812, and specificity of 0.909. PPGI combined with machine learning may therefore provide a non-contact, quantitative complement to conventional SPT assessment.
Authors: Sona Sikynova, Vladimir Blazek, Stefan Borik
Institutions: University of Žilina, RWTH Aachen University
Language: English
Subfield: Biomedical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7225411940