Feature selection for optimized skin tumor recognition using genetic algorithms

摘要: In this paper, a new approach to computer supported diagnosis of skin tumors in dermatology is presented. High resolution surface profiles are analyzed recognize malignant melanomas and nevocytic nevi (moles), automatically. the first step, several types features extracted by 2D image analysis methods characterizing structure profiles: texture based on cooccurrence matrices, Fourier fractal features. Then, feature selection algorithms applied determine suitable subsets for recognition process. Feature described as an optimization problem approaches including heuristic strategies, greedy genetic compared. As quality measure subsets, classification rate nearest neighbor classifier computed with leaving-one-out method used. Genetic show best results. Finally, neural networks error back-propagation learning paradigm trained using selected sets. Different network topologies, parameters pruning investigated optimize performance classifiers. With optimized system 97.7% achieved.