The database of the Service of Dermatology at the Pedro Hispano Hospital (HPH) has about 4000 clinical cases using dermoscopic images. This represents a valuable resource, but its current format is not adequate for a reliable and accurate automatic analysis. Thus, the database will be converted into a common format and enriched with new information. A number of representative cases from the main pathologies will be selected for evaluation. The trained volunteers from the HPH will evaluate these images following the various early warning methods and register the detailed scoring on the database.

As the images obtained through dermoscopy have a great diversity, the algorithm will have to be able to remove noise and other undesired features in the image, and to correctly segment the lesion. Several image segmentation methods will be applied to this set of images. The performance of the various methods will be compared with the strong and weak points identified.

Develop algorithms to extract meaningful features from the segmented dermoscopic images, and to develop a lesion analysis system able to identify dangerous lesions and to provide a tentative classification. Two approaches will be followed concerning the extraction of meaningful features from the images. The first approach will be inspired in variables evaluated by the specialists on early warning screening methods. We will also follow a second approach which does not attempt to mimic human evaluation but selects mathematical features instead.

Evaluation of the automatic diagnosis system developed on the ADDI project in real life clinical trials. As soon as a beta version of the software is available, it will be installed at the HPH (Dermatology Service). The system will then be used in clinical practice, both in dermatoscopy and in tele-dermatoscopy, where the problems and limitations of the system should be detected.

Developing a software prototype to automatically classify the dermoscopic images and perform the tele-dermoscopy. The software should be able to perform the full range of tasks with minimum operator intervention. It should nevertheless provide details of intermediate stages, such as the segmentation of the lesion, the features values extracted, the values estimated for the parameters of the existing early warning screening methods, and the tentative lesion classification. The aim of this task will be also to provide the medical community with a simulation tool for dermoscopy diagnosis of skin lesions providing an excellent environment for education and training purposes.

A complementary objective of the project is to use the enriched database and the software prototype created for simulation and educational purposes. An educational tool will be developed in the later stages of the project, which will be suitable for clinicians and health-care staff, as well as researchers in the image processing area, in order to become familiar with dermatology, dermoscopy and tele-dermoscopy.