A skin lesion classification method based on expanding the surrounding lesion-shaped border for an end-to-end Inception-ResNet-v2 classifier

Rym Dakhli,Walid Barhoumi
DOI: https://doi.org/10.1007/s11760-023-02577-0
2023-05-15
Abstract:The latest computer vision and machine learning technologies have introduced various computer-aided diagnosis (CAD) systems to automate the early diagnosis of skin lesions. Nevertheless, improvements made by CAD systems are not optimal because of the similarity in the appearance of skin lesions of different classes as well as the limitations of segmentation. In addition, according to dermatologists, the shape of the lesion and its infiltration into the surrounding skin are decisive information for the diagnosis. Inspired by this idea, the proposed method is based on gradually expanding the lesion border by including the lesion-shaped border area in order to associate the input image with the corresponding skin cancer type using an end-to-end Inception-ResNet-v2 classifier. The main contribution of this work lies in investigating the Inception-ResNet-v2 model exclusively on the expanded lesion-shaped border. In fact, the obtained results showed that the proposed method is effective in achieving precision rates of 95.6%documentclass[12pt]{minimal}usepackage{amsmath}usepackage{wasysym}usepackage{amsfonts}usepackage{amssymb}usepackage{amsbsy}usepackage{mathrsfs}usepackage{upgreek}setlength{oddsidemargin}{-69pt}egin{document}$$95.6\%$$end{document} and 97.26%documentclass[12pt]{minimal}usepackage{amsmath}usepackage{wasysym}usepackage{amsfonts}usepackage{amssymb}usepackage{amsbsy}usepackage{mathrsfs}usepackage{upgreek}setlength{oddsidemargin}{-69pt}egin{document}$$97.26\%$$end{document} on HAM10000 and PH2 datasets, respectively.
engineering, electrical & electronic,imaging science & photographic technology
What problem does this paper attempt to address?