A Novel Hybrid System for Skin Lesion Detection

In this paper, a new intelligent method of classifying benign and malignant melanoma lesions is implemented. The system consists of four stages; image pre-processing, image segmentation, feature extraction, and image classification. As the first step of the image analysis, pre-processing techniques are implemented to remove noise and undesired structures from the images using techniques such as median filtering and contrast enhancement. In the second step, a simple thresholding method is used to segment and localise the lesion, a boundary tracing algorithm is also implemented to validate the segmentation. Then, a wavelet approach is used to extract the features, more specifically wavelet packet transform (WPT). Finally, the dimensionality of the selected features is reduced with principal component analysis (PCA) and later supplied to an artificial neural network and support vector machine classifiers for classification. The ability to correctly discriminate between benign and malignant lesions was about 95% for the Artificial Neural Network and 85% for the Support Vector Machine classifier.