Título: Browning index: An image processing technique to aid in the segmentation of skin lesions on dermatoscopic images
Autores: Edmilson Q. S. Filho, Evandro O. T. Salles, Jacques Facon, Patrick M. Ciarelli
Resumo: Although most people believe that melanoma cancer is restricted only to skin cancer, the great danger is the spread of metastases from the skin and subcutaneous tissues to any part of the body. This justifies that early detection of cutaneous melanoma cancer remains the primary factor in reducing the mortality of this type of cancer. Currently, non-invasive segmentation-based approaches represent one of the most efficient computational frameworks when it comes to melanoma recognition. Instead of using color images directly in RGB space, as in many works in the literature, we propose the use of browning indices to highlight the differences between the skin and the area of interest to increase the recognition rates of cutaneous melanoma. Three browning indices (Aimonino, Fetuga, and Lunadei) and three segmentation techniques were evaluated in the task of cancer segmentation on a public dataset of dermatoscopic images. In addition to these indices, the U-Net network was used for the purpose of comparison and it was also evaluated combined with browning indices. The experiment rates highlighted the potential of the browning indices to better perform melanoma segmentation. U-Net obtained the Jaccard Index and F1 Score of 0.594 and 0.805, respectively, against 0.719 and 0.862 achieved by the combination of U-Net and browning indices
Palavras-chave: Skin Cancer, Image Processing, Browning Index, Image Segmentation, U-Net
Páginas: 6
Código DOI: 10.21528/CBIC2023-097
Artigo em pdf: CBIC_2023_paper097.pdf
Arquivo BibTeX: CBIC_2023_097.bib