Search for resolution invariant wavelet features of melanoma learned by a limited ANN classifier

Abstract

This article addresses the Computer Aided Diagnosis (CAD) of melanoma pigmented skin cancer. We present back-propagated Artificial Neural Network (ANN) classifiers discriminating dermoscopic skin lesion images into two classes: malignant melanoma and dysplastic nevus. Features used for our classification experiments are derived from wavelet decomposition coefficients of the image. Our research objective is i) to select the most efficient topology of the hidden layers and the network learning algorithm for full-size and downgraded image resolutions and, ii) to search for resolution-invariant topologies and learning methods. The analyzed classifiers should be fit to work on ARM-based hand-held devices, hence we take into account only limited learning setups.