ADProNet: a few-shot prototype network for erythema classification of childhood atopic dermatitis

The assessment of the severity of erythema in pediatric atopic dermatitis (AD) highly relies on the experience of doctors, with significant subjective differences and a lack of high-quality annotated images. To this end, this study proposes a classification model based on a few-shot prototype network, ADProNet, aiming to achieve objective and accurate severity grading using a small number of samples. This study employs a few-shot learning framework (EasyFSL) to construct a classification model for pediatric AD erythema, using the pre-trained Resnet18 as the feature extraction backbone to achieve accurate classification of pediatric AD erythema under few-shot conditions. The image data was collected from the Capital Center For Children's Health，Capital Medical University in Beijing, China, between July and September 2025. Based on a dataset of 140 images of children’s AD erythema with three severity levels, this method achieved better performance than traditional models in both 1-shot and 5-shot tasks, with an accuracy of 88.16% and an F1 score of 88.17%. This method can achieve high classification accuracy with a small amount of data, demonstrating certain practical value. It provides a new technical approach for assessing the severity of pediatric AD, which helps improve the diagnostic efficiency of pediatric AD.