Backpropagation-Based Non Linear PCA for Biomedical Applications

Abstract

Machine learning methodologies such as artificial neural networks (ANN), fuzzy logic or genetic programming, as well as principal component analysis (PCA) and intelligent control have been recently introduced in medicine. ANNs imitate the structure and workings of the human brain by means of mathematical models able to adapt several parameters. ANNs learn the input/output behavior of a system through a supervised or an unsupervised learning algorithm. In this work, we present and demonstrate a new pre-processing algorithm able to improve the performance of an ANN in the processing of biomedical datasets. The algorithm was tested analyzing lung function and fractional exhaled nitric oxide differences in the breath in children with allergic bronchial asthma and in normal population. Classification obtained using non linear PCA based on the new algorithm shows a better precision in separating asthmatic and control subjects.