Segmentation and classification of melanoma and benign skin lesions

Abstract The incidence ofmalignant melanoma has been increasing worldwide. An efficient non-invasive computer-aided diagnosis (CAD) is seen as a solution to make identification process faster, and accessible to a large population. Such automated system relies on three things: reliable lesion segmentation, pertinent features’ extraction and good lesion classifier. In this paper, we propose an automated system that uses an Ant colony based segmentation algorithm, takes into consideration three types of features to describe malignant lesion:geometrical properties, textureand relative colors from which pertinent ones are selected, and uses two classifiers K-Nearest Neighbor (KNN) and Artificial Neural Network (ANN). The objective of this paper is to test the efficiency of the proposed segmentation algorithm, extract most pertinent features that describe melanomas and compare the two classifiers. Our automated system is tested on 172 dermoscopic images where 88 are malignant melanomas and 84 benign lesions. The results of the proposed segmentation algorithm are encouraging as they gave promising results. 12 features seem to be sufficient to detect malignant melanoma. Moreover, ANN gives better results than KNN.