In this paper, a new approach to computer supported diagnosis of skin tumor
s in dermatology is presented. High resolution skin surface profiles are an
alyzed to recognize malignant melanomas and nevocytic nevi (moles), automat
ically. In the first step, several types of features are extracted by 2D im
age analysis methods characterizing the structure of skin surface profiles:
texture features based on cooccurrence matrices, Fourier features and frac
tal features. Then, feature selection algorithms are applied to determine s
uitable feature subsets for the recognition process. Feature selection is d
escribed as an optimization problem and several approaches including heuris
tic strategies, greedy and genetic algorithms are compared. As quality meas
ure for feature subsets, the classification rate of the nearest neighbor cl
assifier computed with the leaving-one-out method is used. Genetic algorith
ms show the best results. Finally, neural networks with error back-propagat
ion as learning paradigm are trained using the selected feature sets. Diffe
rent network topologies, learning parameters and pruning algorithms are inv
estigated to optimize the classification performance of the neural classifi
ers. With the optimized recognition system a classification performance of
97.7% is achieved. (C) 1999 Elsevier Science B.V. All rights reserved.