The aim of MEDEA project is to develop a machine learning based diagnostic tool aiming at improving the accuracy of the automatic skin cancer detection and helping the clinical staff in developing and using semiautomatic techniques for extracting and combining different acquisition methods. The objectives of the project will be achieved by annotating and classifying image data provided by one of the most important italian dermoscopic institute IDI-IRCCS in the context of an ongoing joint collaboration. The annotated data will be used to create a large training set and a test set to validate both image segmentation and automatic classification. Since the boundary irregularity of skin lesions is of clinical significance for distinguishing between malignant melanomas and benign moles, an accurate contour extraction can provide useful features for dealing with the problem of early detection of malignant skin lesions. The set of extracted features will be used to provide the input for an automatic classifier, whose training will benefit from the large number of image samples that MEDEA can exploit.
To achieve the goals of MEDEA, the images coming from the dataset of the Istituto Dermopatico dell'Immacolata (IDI-IRCCS), will be used. In particular a large database consisting of more then 20000 skin lesion images has been provided in the context of a joint collaboration between DIAG and IDI-IRCCS. The images will be classified according to the clinical diagnosis and labeled in order to create a training dataset and a test dataset. In addition the images coming from the publicly available dataset PH2 will be used in order to compare the results of the proposed research with respect to the state of the arts methods.