Date Published: February 6, 2019
Publisher: Public Library of Science
Author(s): Grzegorz Surówka, Maciej Ogorzalek, Seyedali Mirjalili.
This article refers to the Computer Aided Diagnosis of the melanoma skin cancer. We derive wavelet-based features of melanoma from the dermoscopic images of pigmental skin lesions and apply binary C-SVM classifiers to discriminate malignant melanoma from dysplastic nevus. The aim of this research is to select the most efficient model of the SVM classifier for various image resolutions and to search for the best resolution-invariant wavelet bases. We show AUC as a function of the wavelet number and SVM kernels optimized by the Bayesian search for two independent data sets. Our results are compatible with the previous experiments to discriminate melanoma in dermoscopy images with ensembling and feed-forward neural networks.
Melanoma, the neoplasm of the pigment cells of the skin, is still a challenge both for clinicians and CAD (Computer Aided Diagnosis) specialists. Observations with the naked eye and with instruments, especially with popular optical or even with digital dermoscopes require long-term experience which is hard to achieve not only for general practitioners but also for dermatologists. Diagnosis of mature melanoma moles, due to asymmetry, variety in colors or border irregularity, may not be difficult, which is quite the contrary to early melanoma lesions that lack those indications. Effective treatment of melanoma i.e. a high (>95%) 5 or 10-year survival rate consists in early detection and resection of the malignant skin lesion . When not excited at an early stage, melanoma penetrates deep from epidermis to the skin and finally transfers to the lymph nodes and other internal organs by metastasis. At this stage the mortality rate is extremely high and especially for at least a decade has become a medical problem. This problem refers to all countries but particularly these where melanoma morbidity rate is elevated. Statistics says that women got melanoma moles on the legs and men on the back.
Our setup was coded first with LIBSVM  and then recoded to Matlab . The experiments ran on a 4-core i7 workstation with 32GB RAM. We cross-validated (10-fold CV) our SVM classifiers on the standardized (N(0, 1)) predictor vectors.
Computer aided diagnostic systems are common. For the early detection of cutaneous melanoma they play a crucial role to support clinics and general practitioners. Our work contributes to this effort in the field of feature extraction. Appropriate features project the information from a (dermoscopy) image into a space where classes (malignant and benign lesions) are well separated. This step is critical for the performance of the classifier and the learning procedure.