Multivariate Analysis Applied to Polymer Imaging Data Obtained by Near-Field Infrared Microscopy

Chemical imaging techniques such as mass spectrometry (MS) imaging and imaging spectroscopy have grown to be important in a variety of fields. Infrared spectrum information, for example is essential to evaluate organic and biological samples. Recently, near-field spectroscopy techniques have been developed that enable higher spatial resolution above the one usually obtainable due to wavelength limitations. In terms of chemical imaging for organic materials, time-of-flight secondary ion mass spectrometry (TOF-SIMS) is one of the powerful techniques because of extremely high sensitivity and high spatial resolution of approximately 100 nm. Since TOF-SIMS does not always provide complete information on complex samples, a complementary technique of similar spatial resolution is required. Near-field infrared microscope (NFIR) is the most promising candidate for a complementary analysis method along with TOF-SIMS. It is, however, often difficult to interpret NFIR data because of the low signal intensity in near-field infrared. Multivariate analysis techniques such as principal component analysis (PCA), which have successfully been applied to TOF-SIMS imaging data, would also likely be helpful for NFIR data interpretation. In this study, a multicomponent model polymer sample was measured with NFIR and then the image data along with the complex NFIR spectra were analysed by PCA. As a result, the components in the model sample can be separately displayed based on groups of peaks specific to every component indicated by PCA. [DOI: 10.1380/ejssnt.2017.19]


INTRODUCTION
Chemical imaging has become important in a variety of fields such as life science, material science and medical applications.There are several basic techniques that have been applied to chemical imaging.For example, mass imaging, including secondary ion mass spectrometry [1][2][3][4][5], is one of the most powerful imaging techniques because it provides detailed images with high spatial resolution of approximately 100 nm using time-of-flight secondary ion mass spectrometry (TOF-SIMS) [1].Spectroscopy techniques such as infrared (IR) spectroscopy [6,7] are also useful for obtaining imaging and chemical structure information.In terms of spatial resolution, techniques using light of longer wavelength typically inferior.However, near-field infrared (NFIR) microscopy [7][8][9][10] was developed to overcome the spectral resolution limitation of the optical diffraction limit and enabled a spatial resolution of several hundred nanometers [7].Chemical images of high spatial resolution similar to that of TOF-SIMS imaging can be obtained by using this technique [11].Although TOF-SIMS generally provides richer chemical information than IR spectroscopy, it must be used with other techniques to complement the quantitative analysis due to matrix effects [12] that often occur when taking TOF-SIMS measurements.On the other hand IR spectroscopic information is sometimes insufficient for the discrimination of materials having the same groups, such as pro- * Corresponding author: aoyagi@st.seikei.ac.jp teins.Therefore, complimentary analysis using NFIR microscope and TOF-SIMS imaging would be more powerful than using either one of the techniques by itself.
As the first stage of the evaluation of complimentary analysis using NFIR microscope and TOF-SIMS, a simple model sample comprising four layers made up of three polymers, polyethylene glycol (PEG), polystyrene (PS) and polycarbonate (PCB), was employed.The model sample has been used in a previous evaluation of the application of multivariate analysis to TOF-SIMS data [5,13].Figure 1 shows TOF-SIMS images of the polymer sample.Each polymer can be displayed based on one of the secondary ion peaks specific to each polymer.Since signals by NFIR are generally very low, it is often difficult to find out one peak or just a few peaks specific to a target material.Even if the specific peaks are found, the image obtained of one of the peaks is usually unclear to evaluate the distribution of interest.Therefore, principal component analysis has been applied to interpret NFIR data and to extract the important peaks in the groups related to each material in the sample.

A. Sample Preparation
Four layers made up of three polymers, polyethylene terephthalate (PET), polystyrene (PS), polycarbonate (PCB) and PS, were compressed to connect with one another using a universal film maker (S.T.Japan Inc., Tokyo, Japan).Then a cross-section of the layers was sliced using a utility knife and then the sample was pressed again using a handy type press maker STJ-0130 (S.T.Japan Inc., Tokyo, Japan).Finally, layered polymers were immobilized on an aluminium-coated glass substrate.A schematic image of the sample is shown in Fig. 2.

B. NFIR measurement
The polymer sample was measured using a near-field infrared microscope (NFIR 200, JASCO Co., Japan) with a probe of 1 µm diameter (JASCO Co.) over a 20 µm × 10 µm (for x and y orientations, respectively) region of the sample.A spectrum ranging from 800 to 4000 cm −1 was measured every 1 µm.The resolution of the wavenumber was 8 cm −1 .All of the spectra were normalized by a background spectrum of an Al mirror sample.Since the cross-section of the polymer layers is thin, the measured spectra were thought to be reflection spectra transformed by Kramers-Kronig transformation [14,15].

C. TOF-SIMS measurement
The polymer sample was measured using TOF.SIMS 5 (ION-TOF GmbH, Germany) with 30 kV Bi ++ 3 while maintaining the total primary ion dose at less than 10 12 ions/cm 2 .The secondary ion images were obtained over a 300 µm × 300 µm region of the sample.A low-energy electron flood gun was used for charge neutralization.

D. Data analysis using PCATOF-SIMS measurement
Principal component analysis (PCA) is described in detail elsewhere [4,5,13], but the main concept is briefly explained.PCA involves a mathematical procedure transforming a number of variables into a smaller number of uncorrelated variables called principal components (PCs).In terms of chemical mapping data containing chemical spectra having a number of peaks, the peaks are generally categorized into several important components, PCs.The important several PCs often indicate the existence of particular materials or important features of the sample data.The first principal component accounts for as much of the variability in the data as possible, and each succeeding component accounts for as much of the remaining variability as possible.Moreover, PCA is one of the unsupervised pattern recognition techniques, and therefore provides results unbiased by human input.
Prior to PCA, the intensity of every wave number was collected for every mapping data point.For data analysis, the first half of the mapping data, an area of 20 µm × 10 µm, was used.It was found there are 831 wavenumber peaks at 231 (21 × 11) pixels.The data matrix of 831 × 231 was thus used for PCA. Figure 3 shows a schematic of the data analysis and the mapping of the data structure.Each pixel has a spectrum containing a number of peaks.It is recommended to use absorbance because it is difficult to interpret the PCA results of transmittance.The data sets were treated with mean-centring or auto scaling using the PLS Toolbox (Eigenvector Research Inc., WA, USA) and Matlab (Mathworks, MA, USA) before PCA.

E. Results and Discussion
Several points in the model sample that consisted of four layers of three polymers were measured using a NFIR microscope (NFIR-200).The maximum scanning area of NFIR-200, 20 µm by 20 µm, is not large enough to cover four-polymer-layer regions.Each NFIR mapping data point thus contains at most two polymers.peaks is low, it is difficult to display a clear distribution mapping for each polymer based on just one specific peak.Therefore, all of the peaks at every pixel were analysed by PCA in order to extract a group of peaks specific to each polymer.Figure 5 shows the main PCA result of the NFIR data measured over an area between PS and PCB.PC1 score mapping displayed homogeneous distribution and was not useful for separating the two polymers.PC1 would correspond with common factors of the polymers or contaminants, which have been suggested by TOF-SIMS measurement in a previous study [5].In contrast, the PC2 positive and negative high score areas reflect PS and PCB, respectively, because PCB and PS are distributed on the right and left sides, respectively.Thus, the PCB and PS areas were clearly separated out by PCA.From the references [16,17] and FT-IR measurement spectra of PCB, the wavenumber peaks specific to PCB are 1227, 1175, 1289 cm −1 , 1266, 1291 and 1792 cm −1 .In the same manner, the peaks specific to PS are 840, 1152, 1180 cm −1 and 1182 cm −1 .According to the loadings of PCs 1 and 2 (Table I), the peaks related to PCB and PS were sug-gested correspond to wavenumbers related to the PCB and PS structures.
Figure 6 shows the score image for PC1 (82.10%) and for the other NFIR data measured in an area between PS and PET, and Table II shows the wavenumber peaks having higher loadings for PC1.The PET-related wavenumber peaks, 1578, 1615, 1454 and 1234 cm −1 [18], are suggested by the loadings of PC1.They have higher loadings and therefore they correspond with the PC1 score image.In the measurement area, PET is distributed on the right side and PS distributes at the left side edge.Since there is a large trench between PET and PS, the PS area was not effectively measured.Therefore, the contribution of PS-related wavenumber peaks is relatively low, although some of the PS-related wavenumber peaks, such as 3091, 3054, 2850 cm −1 , are detected in the relatively lower PC1 loadings.The data displayed in Figure 7 shows both the PS and PET areas, although the resolution of the wavenumber is lower at 16 cm −1 than for the other measurements.According to the loadings of PC1 from the data in Fig. 7, wavenumber peaks related to PS, which are 3091, 3054, 2850 cm −1 and 906 cm −1 , and those re-   lated to PET, which are 1578, 1615, 1454 and 1234 cm −1 , are indicated.Using one of the indicated peaks at 1578 cm −1 , the distribution of PET can be mapped from the original NFIR data, as shown in Figure 8. Regarding PS, it was difficult to display a clear PS distribution pattern based on one of the specific peaks, probably because the influence of the trench between PS and PET was so great.Thus, it is indicated that PCA is a powerful means for interpreting complicated and low signal NFIR mapping data.Even though mapping based on one of the specific peaks is not useful for extracting the distribution information of one of the materials in a given sample, the PCA score images are useful for providing clear distribution patterns because they are based on some of the specific peaks.Moreover, NFIR and TOF-SIMS are supportive of each other and for both multivariate analysis techniques can be applied in the same procedure.TOF-SIMS generally provides more detailed chemical information.For example, contamination was clearly indicated by the TOF-SIMS results, although it was difficult to extract from the NFIR results.However, in terms of quantitative analysis, IR spectroscopy is often more reliable than TOF-SIMS because TOF-SIMS data is often highly influenced by matrix effects.

III. CONCLUSIONS
It is indicated that near-field infrared microscope (NFIR) mapping data with complex near-field infrared spectra can be interpreted by principal component analysis (PCA).From the original NFIR data, it is often difficult to obtain the distribution of a target material because signals are generally very low.Nevertheless, even when signals are too low to distinguish from noise, the target material distribution can be obtained based on a group of peaks specific to the target materials that is suggested by PCA.Since NFIR provides a spatial resolution of less than 1 µm, it should be used together with other chemhttp://www.sssj.org/ejssnt(  ical imaging techniques of high spatial resolution, such as time-of-flight secondary ion mass spectrometry (TOF-SIMS).

FIG. 3 .
FIG. 3. The structure of the NFIR spectrum and mapping data.Spectra show transmittance.

TABLE I .
Wavenumber peaks having high values for PCs 1 and 2 loadings.

TABLE II .
J-Stage: http://www.jstage.jst.go.jp/browse/ejssnt/) e-Journal of Surface Science and Nanotechnology Wavenumber peaks having the highest and lowest values of PC 1 loadings