BIBS, a New Descriptor for Melanoma/Non-Melanoma Discrimination
Saeid Amouzad Mahdiraji, Yasser Baleghi, Sayed Mahmoud Sakhaei · 2018
Malignant melanoma is known as the deadliest type of skin cancer. Nowadays, several automatic diagnosis systems have been developed in order to improve the detection of melanoma lesions. Boundary fluctuation of skin lesions is an important factor to discriminate different types of lesions. This paper introduces a new border descriptor that properly quantifies contents of the concave contours. The Fourier and wavelet descriptors of the proposed boundary intersection-based signature (BIBS) are obtained. Support vector machine (SVM) is used for classification process. It is shown that the BIBS has better performance in comparison with conventional shape signature. Also, the combination of the BIBS Fourier descriptor and a shape and textural feature set achieves the best accuracy of 90.34%, sensitivity of 92.02%, specificity of 88.08%, and precision of 91.72% with less feature dimension amongst previous works on Dermatology Information System and DermQuest datasets.