Autoregressive modelling of chromatographic signals from urine samples for prostate cancer diagnosis

Abstract

This article evaluates autoregressive modeling as a feature extraction method in a database of chromatographic signals from urine samples for non-invasive diagnostic support of prostate cancer in response to the research question: Can chromatographic signals from urine be characterized and used as a non-invasive method for cancer diagnosis? For this purpose, a database of 18 patients, 9 diagnosed with prostate cancer and 9 control patients, is consolidated, statistical methods are implemented to generate autoregressive coefficients from the data signals, and finally, the principal component analysis technique is applied for crossclass classification. As a result, a correct classification was obtained in the total number of samples validating the autoregressive modelling as a feature extraction method in contrast to the conventional methodology usually followed in chromatographic signal processing.