Chapter 11 - Empirical wavelet transform-based fast deep convolutional neural network for detection and classification of melanoma

Abstract

Melanoma is a type of skin cancer which develops from melanocytes, responsible to provide skin color. The severity of melanoma is defined on the basis of different stages which depends upon the depth of penetration, and the early detection of melanoma at its prodromal stage is very crucial to stop its advancement. In this work, a novel variant of the deep convolutional neural network (DCNN) is developed and called as the fast deep convolutional neural network (fast-DCNN) to perform a binary classification of normal nevus and melanoma by using the dermoscopic images of the PH2 dataset. Furthermore, the discrete wavelet transform (DWT) and the empirical wavelet transform (EWT) are employed to decompose the images for obtaining the horizontal details and first sub-band images, respectively, for the purpose of multi-resolution analysis. The horizontal details and the first sub-band images are used as input to the fast-DCNN to categorize the images on the basis of probability density function by applying the most popular softmax classifier. Finally, the classification accuracy of fast-DCNN, DWT-based fast-DCNN, and the EWT-based fast-DCNN is compared, and the EWT-based fast-DCNN emerged as the best method to categorize the normal nevus and melanoma.