A Machine Learning Based Melanoma Skin Cancer using Hybrid Texture Features

There is a great rise in patients afflicted with the disease Melanoma, the most unpredictable and potentially fatal cancer. If treated in time, the fatality rate can be reduced significantly in most cases. Active Melanoma discovery in the incipient stage is difficult to recognize. As the availability of qualified dermatologists is scarce, a computerized automated method of finding Melanoma accurately is urgently needed. The aforementioned research aims to accomplish this by using supervised machine learning algorithms that are capable of accurately identifying and classifying melanoma to separate skin lesions from the images collected from the combined dataset of the ISIC archive and the PH2 dermoscopic images. For segmenting the tumor portion, the Otsu Thresholding algorithm is employed. The segmented images are applied to a preprocessing network to extract features of ABCD, GLCM, LBP and Histogram of Gradients (HG). Later, these features are trained with different supervised classifiers to generate the model. Later the test data is fed to the model to obtain the predictive labels. The results of the proposed skin melanoma detection prove that skin melanoma detection with XGBoost provides superior results compared to the existing works in the literature.