Automatic Segmentation and classification of Skin Lesion Images

The ultimate aim of this work is to provide an automatic detection of melanoma skin lesion images using a computer-aided diagnosis (CAD) system. This work presents the different steps of such a process. We detail in this paper the segmentation step by describing the different used methods in the literature and propose a hybrid approach that can be integrated in our system. We use three different methods of automatic segmentation based on thresholding, morphology functions and active contours (snakes). The malignancy signs are quantified in a set of parameters that summarize the geometric and photometric characteristics of the lesion. Parameters the more robust and most discriminative have been kept for the classification. These parameters constitute the entry of the stage of classification by neural network. We evaluate our work on different cases of color skin lesion images from digitized slides data base selected by expert dermatologists from the hospital "CHU de Rouen" France