A novel unsupervised initial pattern recognition algorithm based on wavelet transform and window selection

This paper presents an unsupervised initial pattern recognition approach to determine spectral regions contributing most to intrinsic clustering patterns in nuclear magnetic resonance (NMR) spectra associated with a given pathological stimulus using a small number of realizations based on wavelet transform coupled with a window selection algorithm. Results of different choices of window size are given. The pattern recognition algorithm proposed in this paper can be used as the first step in metabolomic data analysis followed by future supervised algorithms to refine potential biomarkers characterizing tissue-specific lesions further within these regions.