Evaluation of Pulsed Electromagnetic Field as a Treatment for Atopic Dermatitis in a Mouse Model Using Image Analysis Based on Machine Learning

초록

Background Atopic dermatitis (AD) is a chronic inflammatory skin disease generally treated using local or systemic corticosteroids. However, the long-term use of steroids decreases their performance, highlighting the need to develop safe and non-invasive treatment methods. Pulsed electromagnetic field (PEMF) is a non-invasive treatment clinically used to resolve diabetic skin ulcers. This study quantitatively analyzes the effect of PEMF using machine-learning techniques. Methods Six-week-old hairless mice were randomly assigned to a normal group and 2 PEMF groups. The mice in the PEMF groups were treated with 2,4-dinitrochlorobenzene solution to induce uniform lesions that mimic that of AD for 5 weeks; PEMF stimulation was performed in the 2 PEMF groups during the last week. Then, images of the inflammatory process were obtained using 2 different cameras. The t-distributed stochastic neighbor embedding (t-SNE) method and 2 convolutional neural network (CNN) models were compared to distinguish between the atopic and normal groups. Results The CNN models outperformed the t-SNE method and were thus used to evaluate the PEMF test set. The CNN models revealed that the post-treatment PEMF group data were more likely to be normal than the atopic group. Conclusion The study has shown potential for applying deep learning techniques to the diagnosis of various skin lesions. In future, this technique could be applied to humans in clinical trials to verify whether the classification performance reached clinically acceptable level.

키워드