Detection of Skin Cancer Using Bi-Directional Emperical Mode Decomposition and GLCM

Abstract

The most frequent type of cancer in humans is the skin cancer and it can be lethal. It affects in copious forms such as basal, melanoma, and squamous cell carcinoma. Among these, melanoma case is severe, most dangerous and unpredictable. When it is diagnosed in the early stages, it can be controlled and cured considerably. Thus, a novel computational approach using texture feature fusion and machine learning techniques is proposed to diagnose and classify the skin lesions as benign or malignant. The workflow of this approach is preprocessing for noise and hair strands removal, segmentation of the cancer affected region, validation of the segmentation methods, statistical feature extraction, principle feature selection, classification as benign or malignant and performance estimation of the classifier algorithm. The Otsu thresholding, enhanced Otsu thresholding and watershed segmentation methods are implemented and the segmented images are validated using the Jaccard index and Dice index. Further, several features derived from texture, colour, and shape of the segmented images are fused and fed to the variants of the Support Vector Machine (SVM) classifier after the significant features selection process and the performance of the classifiers are evaluated. The results show that cubic SVM classifier (98%, 100%, and 99%) and Fine Gaussian SVM classifier (100%, 100% and 100%) performs well in terms of sensitivity, specificity and accuracy for the considered image dataset. Hence, the proposed method can be used for early detection classification of melanoma.