Classification of Skin Lesions by Fluorescence Diagnosis and Independent Component Analysis

H.G. Stockmeier, W. Bäumler, R.-M. Szeimies, F.J. Theis, E.W. Lang (Germany), and C.G. Puntonet (Spain)


skin lesions, neural networks, independent componentanalysis, fluorescence diagnosis


We develop an approach to automatically classify images of malignant and benign skin lesions obtained by Photodynamic Diagnosis. Our method does not use plain fluorescence intensity levels but relies on discriminative intensity patterns. Therefore a set of characteristic filter vectors is created for each class by independent component analysis. The results of the filtering process are fed into a supervised neural classifier. The classification results are promising. But further experiments with an increased data set taken under standardized image recording conditions seem necessary to yield still better classification rates.

Important Links:

Go Back