Feed-forward network for cancer detection

Abstract

Samples of patients with or without disease can be diagnosed by serum proteomic pattern. Protein mass spectra are created by applying Surface-Enhanced Laser Desorption and Ionization (SELDI). A clinic diagnostic test to improve cancer pathology may be accomplished by this technology. In this paper, aim at FDA-NCI Clinical Proteomics Program Databank, first preprocess carefully data, sort the key features according to class separability criteria and extract the key features according to principal component analysis(PCA), set the size of the hidden layer neurons based on experience. Percentage correct classification is 100%. The results of experiment are analyzed according to confusion matrix and the receiver operating characteristic plot.