Tensor decomposition-based feature extraction for noninvasive diagnosis of melanoma from the clinical color image

We propose a feature extraction method for noninvasive diagnosis of melanoma based on tensor decomposition of the clinical color image of skin lesion. Extracted features are elements of the core tensor in the corresponding Tucker3 model, and represent spatial-spectral profile of the lesion. In contrast to majority of methods that exploit either texture or spectral diversity of the tumor only, this method simultaneously captures spatial and spectral characteristics. The proposed procedure is demonstrated on a problem of noninvasive diagnosis of melanoma from cost-effective auto-fluorescence color images of skin lesions, with overall sensitivity 82.1% and specificity 86.9%.