Analysis and classification of normal and pathological skin tissue spectra using neural networks

An innovative spectroscopic diagnostic method has been developed for investigation of different regions of normal human skin tissue, as well as cancerous and precancerous conditions ill vivo, ex vivo and in vitro. This new method is a combination of fiber-optical evanescent wave Fourier Transform infrared (FEW-FTIR) spectroscopy and fiber optic techniques using low-loss, highly flexible and nontoxic fiber optical sensors. The FEW-FTIR technique is nondestructive and very sensitive to changes of vibrational spectra in the IR region without heating and staining and thus altering the skin tissue. A special software package was developed for the treatment of the spectra. This package includes a database, programs for data preparation and presentation, and neural networks for classification of disease states. An unsupervised neural competitive learning neural network(1) is implemented for skin cancer diagnosis. In this study, we have investigated and classified skin tissue in the range of 1400 to 1800 cm(-1) using these programs. The results of our surface analysis of skin tissue are discussed in terms of molecular structural similarities and differences as well as in terms of different skin states represented by eleven different skin spectra classes.