Implementation of Deep Learning Models for Skin Cancer Classification

Abstract

Melanoma skin cancers are most threatening disease. Manual detection of melanomas using dermoscopic images is very time-consuming method which also demands a high level of competence. An accurate and prompt diagnosis needs the development of an intelligent classification system for the detection of skin cancer. This paper implements deep learning models for skin cancer classification and integrates features obtained from several feature extraction methods. Pre-processing, feature extraction, classification, and performance evaluation are phases of proposed approach. Any superfluous noise in the edges is removed during the pre-processing stage. The Gaussian filter method is used to improve image clarity and remove unwanted pixels. The detection of melanoma cells is based on features such as lesion segmentation and colour of images. The contour approach, contrast, Grey scale approaches, lesion segmentation using U-NET are employed for feature extraction. Deep learning-based classifiers such as ResNet50 and CNN architecture are used to classify based on extracted features. Classification techniques use these qualities to identify malignant and affected skin areas. Sensitivity, specificity, accuracy, and F-score are some of the performance measurement criteria used to evaluate the suggested approach. The classifiers are used on the HAM10000 dataset. On the HAM10000 dataset, the suggested framework outperformed existing melanoma detection systems.