Component spectra extraction from terahertz measurements of unknown mixtures

Abstract

The aim of this work is to extract component spectra from unknown mixtures in the terahertz region. To that end, a method, hard modeling factor analysis (HMFA), was applied to resolve terahertz spectral matrices collected from the unknown mixtures. This method does not require any expertise of the user and allows the consideration of nonlinear effects such as peak variations or peak shifts. It describes the spectra using a peak-based nonlinear mathematic model and builds the component spectra automatically by recombination of the resolved peaks through correlation analysis. Meanwhile, modifications on the method were made to take the features of terahertz spectra into account and to deal with the artificial baseline problem that troubles the extraction process of some terahertz spectra. In order to validate the proposed method, simulated wideband terahertz spectra of binary and ternary systems and experimental terahertz absorption spectra of amino acids mixtures were tested. In each test, not only the number of pure components could be correctly predicted but also the identified pure spectra had a good similarity with the true spectra. Moreover, the proposed method associated the molecular motions with the component extraction, making the identification process more physically meaningful and interpretable compared to other methods. The results indicate that the HMFA method with the modifications can be a practical tool for identifying component terahertz spectra in completely unknown mixtures. This work reports the solution to this kind of problem in the terahertz region for the first time, to the best of the authors' knowledge, and represents a significant advance toward exploring physical or chemical mechanisms of unknown complex systems by terahertz spectroscopy.

Terahertz spectral unmixing based method for identifying gastric cancer

“…As human tissue progresses from normal to cancerous, the tissue structure and internal components change significantly, which results in weight changes in the decomposed peaks. This allows the proposed method to not only correctly predict the number of pure components but also obtain the identified pure spectra that exhibited good similarity with the true spectra (Li et al 2015). The successful application of the HMFA method in infrared and Raman spectra shows that it can be used to effectively extract the composition spectra of mixtures (Kriesten et al 2008b).…”

“…Previous studies of terahertz spectral unmixing techniques have shown good results in the detection of simple mixtures (Li et al 2015). This paper proposes a terahertz spectral unmixing method for biological tissue without prior knowledge, and provides an explanation of the terahertz data of cancerous tissue from the perspective of the pathologic mechanisms.…”

HYPERION: Hyperspectral Penetrating-type Ellipsoidal Reconstruction for Terahertz Blind Source Separation

“…NMF is the unmixing method for blind separation problems designed according to the non-convex optimization approaches. HMFA is the unmixing method based on peak fitting and has demonstrated its efficacy in mid-infrared and THz bands [33,44]. The comparison among HYPERION, nICA, NMF, and HMFA under different noise SD is shown in Figure 5.…”

“…This is a blind spectral unmixing problem waiting to be answered, especially in THz spectroscopy, THz hyperspectral imaging, and THz biophotonics fields. Recently, research groups have implemented THz spectral unmixing methods for material blind separation -hard modeling factor analysis (HMFA) [33], nonnegative matrix factorization (NMF) [34], and independent component analysis (ICA) [35]. However, all the conventional THz spectral unmixing methods still rely on properties of material information, the size of a measured dataset, and the signal-to-noise ratio of the dataset, which severely limits their scope of practical use.…”

A qualitative analysis method for multi-component gas mixtures via terahertz rotational spectroscopy

“…Hard modeling factor analysis (HMFA) 12 based on the indirect hard modeling method 13 was applied to resolve the terahertz spectra of unknown mixtures. 14 A terahertz spectral unmixing method by introducing modified HMFA was also used to identify gastric cancer from normal tissue. 15 These methods are suitable for solid and liquid mixtures with severe spectral overlapping whose absorption peaks are relatively sparse.…”