Feed Forward Neural Network for Autofluorescence Imaging Classification

The key elements in cancer diagnostics are the early identification and estimation of the tumor growth and its spread in order to determine the area to be operated on. The aim of our study was to develop new methods of analyzing autofluorescence images which will allow us an objective and accurate assessment of the location of a tumor and will also be helpful in determining the advancement of the disease. The proposed classification methods are based on neural network algorithms. An Olympus company endoscopic system was used for an autofluorescence intestine imaging study. The autofluorescence imaging analysis process can be divided into several main stages. The first step is preparation of a training data set. The second one involves selection of feature space, namely the selection of those features which enable distinguishing the pathologically altered areas from the healthy ones. Final stages of the analysis include pathologically changed tissue classification and diagnosis.