Spectral-spatial feature fusion of hyperspectral imaging for skin lesion classification

Atopic dermatitis (AD) is a chronic inflammatory skin disease characterized by poorly defined borders and a widespread distribution. Clinical evaluations of AD commonly rely on surface imaging, limiting insight into subsurface skin changes. This study investigates the utility of hyperspectral imaging for visualizing subsurface skin changes and identifying optimal data combinations to assess AD severity. A feature-level fusion model was constructed to extract both spectral and spatial features across different skin layers. The classification performance of each configuration was analyzed to assess its effectiveness in determining the severity of AD. The combination of Range C (750–900 nm) and spatial features achieved the best severity classification performance among the fusion models, yielding an accuracy of 0.95 ± 0.07. Combinations including Range C consistently outperformed the others, suggesting that this spectral band significantly contributed to accurate severity assessment. This study advances AD evaluation by shifting from surface-focused methods to a multilayered approach that incorporates deep structural information.