Optimized auxiliary classifier Wasserstein generative adversarial network fostered skin cancer classification from...

Skin cancer is a disease that affects people of all ages. Automated diagnosis of skin cancer reduces the rate of death by detecting the disease at primary phase. Visual inspecting at the clinical inspection of skin lesion is one of the hard procedure because the similarity between the lesions exists. In this manuscript, Optimized Auxiliary Classifier Wasserstein Generative Adversarial Network fostered Skin Cancer Classification from Dermoscopic Images (OAC-WGAN-SCC-DI) is proposed. Initially, the input Skin dermoscopic images are engaged from the dataset of Skin Lesion Images for Melanoma Classification. The Dynamic Context-Sensitive Filter was used in removing noise and increasing the quality of Skin dermoscopic image. Next, these pre-processed images are given to Classic Semantic Segmentation Algorithm for segmenting ROI region.The segmented ROI region is given into Dual-Domain Feature Extraction for extracting Radiomic features such as Grayscale statistic features and Haralick Texture features. The extracted features are given into the Auxiliary Classifier Wasserstein Generative Adversarial Network (ACWGAN) which classifies the skin cancers, like Melanocytic nevus, Basal cell carcinoma, Actinic Keratosis, Benign keratosis, Dermatofibroma, Vascular lesion including Squamous cell carcinoma. In general, ACWGAN does not show any optimization adaption methods to determine the optimum parameterto offer accurate skin cancer classification. Artificial Humming Bird Optimization Algorithm is proposed in this manuscript to optimize ACWGAN classifier that classifies skin cancer precisely. The proposed OAC-WGAN-SCC-DI is implemented using MATLAB. To classify Skin cancer, performance metrics like precision, accuracy, F1-score, Recall (Sensitivity), Matthew's correlation coefficient, specificity, Jaccard co-efficient, Error rate, ROC, computational time are considered. Performance of the OAC-WGAN-SCC-DI approach attains 13.11%, 27.12% and 18.73% high specificity, 29.13%, 23.04% and 19.51% lower computation Time, 22.29%, 5.365%, 1.551% and 3.915% higher ROC and 28.65%, 3.98%, and 17.15% higher MCC compared with existing methods such as Skin cancer classification of Convolutional Neural Network with optimized squeeze Net by Bald Eagle Search optimization (DCNN-SCC-DI) and Skin cancer detection of Convolutional Neural Network using Gray Wolf Optimization (CNN-TL-SCC-DI), Hybrid convolutional neural networkswith SVM classifier for categorization of skin cancer (SVM-SCC-DI) respectively.