Skin Disease Classification Using A New Unsupervised Competitive Learning Neural Network

摘要

An unsupervised neural competitive learning neural network is implemented for skin cancer diagnosis. The input neuron data set is the Fourier Transform infrared (FT-IR) spectrum obtained by a new Fiberoptic Evanescent Wave Fourier Transform Infrared (FTIR-FEW) spectroscopy method in the range of 1480 to 1850cm~(-1). In this paper, we modified the competitive learning (CL) network. The total square distance TSD is introduced to define the class center vector and the learning rate η is optimized such that the center adjustment has reached its highest speed. With a proper tolerance value, 62 FTIR-FEW spectral data files taken from normal skin tissues, benign tumors and melanomas can be successfully classified. In each spectral data file, there are 401 absorbance values in the interested range and the whole spectrum is directly inputted to the CL network without any feature extraction. The modified CL network has a high efficiency and good stability.