A Study on the Classification of Atopic Dermatitis by Spectral Features of Hyperspectral Imaging

Abstract

Atopic dermatitis (AD) is a chronic skin disease typically evaluated using the Invesrigator's Global Assessment (IGA), SCORing Aoptic Dermatitis (SCORAD), and Eczema Area and Severity Index (EASI) (SCORAD and EASI; S/E). However, the subjectivity of the diagnostic process remains limited. Hyperspectral cameras can detect optical signals to simultaneously generate morphological information and collect spatial information. The depth of skin penetration of caries can be determined according to the wavelength of light and its structural feature. Thus, after dividing the colors and information into six datasets, classified and analyzed their features and severities using representative diagnostic evaluation tools. First, the expansion of blood vessels and structural changes in the lesion area were confirmed between 550 and 650 nm, and information analysis of the lesion features was confirmed through the difference in the rate of change between the visible and near-infrared (NIR) areas. Subsequently, severity classification of the datasets for each wavelength of AD was performed. IGA-NIR-DenseNet121 showed the best classification performance, with 88% accuracy and S/E-G-ResNet18 showed 97% accuracy. Thus, IGA is sensitive to NIR, whereas SCORAD and EASI are sensitive to color. Although the overall accuracy was lower than existing RGB images, the t-distributed stochastic neighbor embedding results showed that the severity groups were well distinguished. The newly constructed dataset and severity classification study by wavelength suggested generalizations and methods for diagnosing and evaluating AD. Additionally, it shows the possibility of expanding research by identifying features that appear in the deep skin layers.