An expert system for cutaneous blood flow in melanocytic skin lesions

A basic tool in microcirculation research is laser Doppler fluxmetry (LDF). The chaotic behaviour of the measured LDF-time series acquires mathematical tools like, for instance, Wavelets. The notion of contrast is known as useful tool to measure differences between two LDF-time series [K. Brauer, Chaos, Attraktoren und Fraktale, Logos, Berlin, 2002]. The one time series arises from the blood flow in healthy skin and the other from a pigmented symmetric contra lateral skin lesion. Our approach is based on taking the contrast from all shorter non-overlapping time intervals of approximate length 5 or 10 seconds. This gives a sample or more precisely, a time series of contrast values. Our goal is an expert system to decide between malign and beligne lesions by estimating the probability for a maligne lesion. As a data base we again use the same data set as [H.-M. Hafner, K. Brauer, M. Eichner, A. Steins, M. Mohrle, A. Blum and M. Junger, Wavelet analysis of cutaneous blood flow in melanocytic skin lesions, J. Vasc. Res., submitted]. The statistical tool is logistic regression. We can show that 93% of data are correctly classified. If we check the expert system against the independent data base of the Greifswald dermatology department we get 78% correctly classified cases. Further work must be done to find a well distributed data base for an expert release system.