Learning from examples to automatically cluster pigmented skin lesions

Our goal was to model the ability of dermatologists to build consistent clusters of pigmented skin lesions in patients. A consensus clustering allows modeling the diversity of skin lesions in each patient as a result of the partitions proposed by nine dermatologists. To learn the dermatologists´ consensus clustering, we used two supervised clustering methods, namely the structural approach and the pairwise approach. These methods learnt similarity measures between individuals´ skin lesions to cluster future individuals´ sets of skin lesions in the same fashion as the dermatologists do. The agreement between partitions obtained from the sequential fusion of both methods and the consensus clustering matches dermatologists agreement.