Cancer diagnosis based on discrete wavelet transformation of mass spectrometry proteomic data
P. Maass (Center for Industrial Mathematics, University of Bremen, Bremen, Germany), T. Alexandrov (Center for Industrial Mathematics, University of Bremen, Bremen, Germany) and H. Thiele (Bruker Daltonics GmbH, Bremen, Germany)
We present a procedure of automatic cancer detection using high-resolution mass spectrometry proteomic data. This procedure is based on discrete wavelet transformation (DWT) and consists of: (1) DWT of the spectra, (2) statistical selection of discriminative wavelet coefficients, (3) construction of a support vector machine classifier with double cross-validation. The procedure is developed in collaboration with Bruker Daltonics company, Bremen, Germany, and evaluated on a data set collected at the Leiden University Medical Center, Leiden, the Netherlands. For these data our procedure provided almost perfect total recognition rate (97.3%), sensitivity (98.4%), and specificity (95.8%).