Research on Accurate Diagnosis of Cutaneous Squamous Cell Carcinoma Based on Spatio-Spectral Fusion Features

Cutaneous squamous cell carcinoma (SCC), a prevalent non-melanoma skin malignancy, poses significant diagnostic challenges due to the limitations of conventional clinical methods. This study introduces an advanced diagnostic framework leveraging hyperspectral imaging (HSI) to enhance SCC detection accuracy. The proposed methodology integrates Gray-Level Co-occurrence Matrix, Gabor filters, and Local Binary Patterns for spatial feature extraction, combined with Gramian Angular Field, Markov Transition Field, and Recurrence Plot for spatio-spectral feature transformation. A novel multi-scale hybrid transformer (MSHT) model is developed to classify skin lesions using microscopic HSI data, capturing both local texture details and global spectral-spatial dependencies through hybrid convolutional and self-attention mechanisms. Comparative experiments demonstrate the MSHT model's superior performance, achieving sensitivities of 0.88, 0.84, and 0.87 for actinic keratosis (AK), seborrheic keratosis (SK), and SCC, respectively. This research establishes a robust diagnostic paradigm for SCC and advances the clinical application of HSI technology through rigorous validation.