Thermography based skin allergic reaction recognition by convolutional neural networks

Area of Science:

• Medical Diagnostics
• Artificial Intelligence in Medicine
• Image Analysis

Background:

• Allergic reaction classification is crucial in medicine but currently relies on subjective human assessment.
• Automated methods are needed to improve the accuracy and efficiency of allergy diagnosis.

Purpose of the Study:

• To develop and validate an automated approach for classifying prick allergic reactions.
• To utilize correlated visible-spectrum and thermal imaging for enhanced diagnostic capabilities.
• To segment individual allergen injection sites for detailed analysis.

Main Methods:

• Implementation of a neural network model for automated allergic reaction classification.
• Utilizing a dataset of 100 patients with 1584 allergen injections.
• Employing correlated visible-spectrum and thermal images of patient forearms.
• Development of image segmentation techniques for identifying injection areas.

Main Results:

• The automated system achieved high performance metrics: 0.98 ROC AUC, 0.97 AP, and 93.6% accuracy.
• Successful segmentation of multiple allergen injection areas from forearm images.
• Demonstrated the potential for reduced examination times and increased data consideration.

Conclusions:

• The proposed automated allergy recognition system demonstrates significant potential for accurate and efficient medical diagnosis.
• Neural network-based analysis of combined visible and thermal imaging offers a robust alternative to human-based classification.
• This technology could streamline allergy testing processes and improve patient care.