3d-wavelet transform based skin cancer classification of VGG-16 network model

Abstract

The RGB dermoscopic skin images utilize the Deep Convolutional Neural Network (DCNN) model with the wavelet analysis of skin image transformations. This proposed Computer-Aided Diagnosis (CAD) based early diagnosis of malignant melanoma skin cancer image has three phases. In this, the acquired image utilizes the median filter to remove noise and skin hairs in the initial phase of pre-processing stage. The second phase employs the 3-DimensionalDiscrete Wavelet Transformation (3D-DWT) feature extraction method with Transfer Learning (TL) technique. Here the 2D-DWT uses the grayscale image and 3D-DWT uses the RGB color image as input for feature extraction, and transformation performancesare evaluated and compared. The result obtained shows that the 3D-DWT-based skin image transformation with 97% accuracy, 95.42% sensitivity, and 97.78% specificityprovides a high-resolution transformation of images. Hence, 3D-DWT transformations provide efficient feature datasets and CNN-based VGG-16 network classifiers for melanoma skin classification.