High-fidelity and high-resolution phase mapping of granites via confocal Raman imaging

In physical sciences such as chemistry and earth sciences, specifically for characterization of minerals in a rock, automated, objective mapping methods based on elemental analysis have replaced traditional optical petrography. However, mineral phase maps obtained from these newer approaches rely on conversion of elemental compositions to mineralogical compositions and thus cannot distinguish mineral polymorphs. Secondly, these techniques often require laborious sample preparations such as sectioning, polishing, and coating which are time-consuming. Here, we develop a new Raman imaging protocol that is capable of mapping unpolished samples with an auto-focusing Z-mapping feature that allows direct fingerprinting of different polymorphs. Specifically, we report a new methodology for generating high fidelity phase maps by exploiting characteristic peak intensity ratios which can be extended to any multi-phase, heterogenous system. Collectively, these enhancements allow us to rapidly map an unpolished granite specimen (~ 2 × 2 mm) with an exceptionally high accuracy (> 97%) and an extremely fine spatial resolution (< 0.3–2 µm).

www.nature.com/scientificreports/ Developing a Raman phase mapping protocol using a mineralogical approach. Phase maps have spatial information about the presence and absence of a particular mineral. In the past, studies have generated phase maps by plotting contrast images using peak intensities 15,29,30 , but well-segmented and definitive images are limited. Here, we have developed a four-step methodology to obtain definitive phase maps which has been illustrated for biotite-1 mineral in the same tile which was used in Fig. 1b 31 . In order to obtain phase maps, these peaks were assigned as characteristic peaks whose peak intensity ratio had a unique value of 2.7 for this mineral. This characteristic peak intensity ratio was used to generate a contrast image (Fig. 2a) whose gray value at every location is proportional to that ratio. Figure 2b shows the image filtered for the required range of ratios (0 to 2.7) which was then subjected to a Fourier transform to segment the required region as shown in Fig. 2c. The accompanying histogram of this image is expected to have peaks corresponding to areas where the mineral is present and absent. Finally, this image in Fig. 2c was then thresholded using the Isodata 32 algorithm in ImageJ resulting in a phase map which is a definitive binary image that shows the presence and absence of the mineral (Fig. 2d). This process was subsequently adopted for all minerals observed in the selected tile and the stepwise results are shown in Fig. 3. All of these minerals are common in granite 33,34 , and using this new methodology, an unambiguous identification is now possible.  www.nature.com/scientificreports/ Mineral phase maps using an elemental approach. EDS coupled with SEM is also another tool to obtain the spatial distribution of different minerals by mapping the elements [35][36][37] . In general, however, it is challenging to observe and compare data collected on the same area of the sample when there are variations in sample orientation, composition, and topography. In a previous approach 38 , compositional differences were observed between different phases using combined EDS elemental maps and back scattered images for quantitative analysis of minerals. Here, however, we have employed a different approach of using the gray value of a point to map it to a mineral based on characteristic Si weight % values. Elemental maps were subjected to ZAF (Atomic number, Absorbance and Fluorescence) corrections to obtain weight % maps resulting in pixels with gray values which are proportional to the weight of the elements. Then, we generated a grayscale Si relative weight % map by combining all the elemental weight % maps using Eq. 1 and have reported the outcome in Fig. 4.
where Wt i is the weight of ith element (i = 1-8: Si, Al, Fe, K, Na, O, Ca, and Mg). This map shows the Si relative weight concentration at every point in the scan area for which phase assignment was done using theoretical values based on known chemical formulae 39 with a tolerance of 0.015%. For minerals such as biotite and hornblende which occur as a series of minerals in nature with known end-members 40 , the corresponding Si weight % ranges were used. The mapped minerals were color coded to generate a composite EDS map which is explained fully in the next section.
Multi-modal correlative imaging. By assimilating both the EDS and Raman imaging maps developed in previous sections along with backscattered and secondary electron imaging, we now report a multi-modal correlative dataset for granite-1 specimen in Fig. 5. The raw Raman images obtained by stitching all the 12 individual images of the observed minerals obtained from characteristic peak intensity ratios are shown in Fig. 5a. The final image ("raw image") in Fig. 5a shows the preliminary Raman contrast map for the entire area capturing sum of intensities at 185.1 and 510.9 cm −1 to illustrate the location of different minerals. The definitive, unambiguous phase maps generated using our novel four-step methodology are shown in Fig. 5b. Two types of biotite were observed (biotite-1 and biotite-2), likely with the same chemical composition but different atomic structure and/or crystalline orientation due to which the characteristic Raman peaks (at 185.1, 675.7, 721.6, and 767.3 cm −1 ) have different relative intensities [41][42][43] . This difference was accurately recorded while obtaining contrast images and thus the minerals were mapped separately. The composite Raman phase map of the region shown in optical image (Fig. 1a) was created by color-coding the individual phase maps and combining them, resulting in the map shown in Fig. 5d. This map was obtained by overlapping all the Raman phase maps which resulted in the mineral region, a non-assigned region which corresponds to the area where none of the detected minerals were present, and an overlap region which partly includes the boundaries of minerals such as biotite and hornblende. In this region, hornblende marks the transition of the biotite to the amphibolite series 44 , which has been accurately observed due to the high resolution possible with Raman imaging. Similarly, other elemental maps show elemental presence only in the region where the minerals which have these elements are present. As Raman spectroscopy can detect differences in mineral structure 42,46 and crystalline orientation 41 , it was possible to distinguish biotite-1 and biotite-2 from one another, whereas with EDS such a distinction was not possible. Specifically, while the composite EDS and Raman maps show an overall good correlation in terms of the type and location of minerals, polymorphs of biotite can only be distinguished by Raman imaging. Figure 5e shows EDS and Raman composite maps of marked boxes in Fig. 5d. The area enclosed in the red and black boxes in Fig. 5d show a continuous map of quartz and a mineral boundary between albite and hornblende respectively. On one hand, a low-resolution map of EDS is good for a   24,47 . The area of the overlap region in Raman composite map exceeds that of non-assigned region in the EDS composite map for quartz and albite. This resulted in low actual mineral percentages in Raman composite map for these two minerals (41.8% and 3.6% for quartz and albite respectively). Biotite and hornblende had low overlaps only in the mineral boundaries due to which the percentages in the Raman composite map (24% for biotite and 16.2% for hornblende) did not reduce significantly.
On the other hand, there are notable disagreements when it comes to the total non-assigned and overlap regions. The EDS composite map has an overall higher amount of non-assigned region, primarily due to its lower spatial resolution (5.12 μm/pixel vs 1.96 μm/pixel). EDS map has zero overlap since every point that gets scanned gets assigned to only one type of mineral based on its Si weight % value. The scan time and resolution values (84 min at 1.96 μm/pixel for Raman and 120 min at 5.12 μm/pixel for EDS) show that Raman spectroscopy is considerably better in terms of time-resolution trade-off. www.nature.com/scientificreports/ We adopted the developed methodology to two other granites with very different compositions (granite-2 and granite-3 abundant in microcline and albite respectively) and obtained a high level of agreement (> 97%) between the mapped phases. The corresponding phase maps are reported in supplementary Fig. 1. and the quantitative phase percentages along with the level of agreement of mapped phases observed by both the techniques are reported in Table 1. The phase percentages quantified by both techniques are compared in supplementary Fig. 2 and we report a high coefficient of determination (0.99) corroborating the developed methodology.
Finally, the back-scattered image of granite-1 in Fig. 5d shows a clear contrast between regions that contain heavy and light elements 48 , in agreement with phases observed by EDS and Raman. The brightest area corresponds to the biotite and hornblende region as it is known that these minerals contain a significant amount of Fe in them 49,50 , followed by orthoclase, albite, and quartz which have elements with relatively lower atomic weights 27 . The secondary electron image and optical image in Fig. 5d reveal the presence of surface heterogeneities and some cracking in the hornblende and quartz region, underscoring the capability of Raman imaging to work well on unpolished specimens.

Conclusion
Our novel approach to generate quantitative mineral phase maps can, in principle, be extended to any multiphase heterogeneous system without the need to prepare polished specimens. Several studies in the past have had limitations, such as absence of high spatial resolution maps 15,29 as well as phase maps that are often not definitive enough [51][52][53] . Thus, an important implication of our results is that we can obtain high-fidelity and high-resolution phase maps of minerals without the use of thin sections or polished specimens which are often laborious to prepare. These definitive phase maps are able to quantify all present minerals, uniquely distinguish polymorphs of a mineral, and also clearly demarcate mineral transition regions. Furthermore, it was also possible to clearly differentiate minerals such as albite and orthoclase which have close principal characteristic Raman peaks by utilizing the unique peak intensity ratio method reported here. The composite phase maps of minerals generated from Raman spectroscopy and EDS had a high level of agreement proving that accurate phase mapping can be done on any rock specimen using the proposed methodology. Finally, Raman imaging has distinct advantages in quantitative mineral mapping and identification when compared to SEM-EDS such as high-spatial resolution and detection of mineral polymorphs.

Materials and methods
Optical imaging and Raman spectroscopy. Granite rock specimens of approximate dimension 50 × 10 × 2 mm obtained from Ward's Science were used to perform the analysis. For granite-1 specimen, Raman Spectroscopy was performed on a scan area of 2.21 × 2.36 mm to characterize the specimen. To obtain scans in high resolution, the scan area was divided into a grid of 3 columns and 4 rows, and images (785 µm × 677 µm) Table 1. Mineral distribution of the selected area as quantified from composite Raman and EDS phase maps. The fourth column shows level of agreement (LOA) for mapped phases which is defined as: 100 − σ, where σ is the standard deviation of quantitative phase percentages obtained from Raman and EDS. The average LOA (97.4%) is for the entire scan area for all 3 granites.

Sample
Mineral Raman (%) EDS (%) LOA (%) Note www.nature.com/scientificreports/ were captured using a 10X objective lens (working distance-17.3 mm, numerical aperture-0.3). The stitched optical image taken in high resolution and one of the tiles used to stitch the image (tile 8 in 3rd row and 2nd column) are shown in Fig. 1. The area in Fig. 5e was captured using a 50X objective (working distance-1 mm, numerical aperture-0.8). Polarized Raman spectra were obtained (Nanophoton Raman 11) using a 532 nm laser of spot size 0.41 µm for 50X objective (1.08 µm for 10X objective) and a 600 gr/mm grating with a slit width of 50 µm. The measurement dimension in every tile using the laser was set to 785.4 × 677.4 µm and an excitation power of 0.32 mW was used. Exposure time was set to 1 sec / line of pixels totaling to 7 min per tile. The spectral resolution was 1.8 cm −1 and the spectra were acquired in the wavenumber range of 200 cm −1 to 2000 cm −1 . The Raman spectra were subjected to baseline correction on RAMAN Viewer, proprietary software developed by Nanophoton Corporation. Raman contrast images were obtained by plotting characteristic peak intensity ratios for all the minerals at the following wavenumbers: quartz (SiO 2 ) − I 466 /I 207.2 cm −1 (ratio-4.5), orthoclase (KAlSiO 3 ) − I 515.2 /I 508.8 cm −1 (shoulders of peak at 515.2 cm −1 , ratio-2.5), albite (NaAlSiO 3 ) − I 508 / I 517 cm −1 (shoulders of peak at 510.9 cm −1 , ratio-2.5), biotite-1 (K(Mg,Fe) 3  Raman phase mapping. A four-step methodology was developed to obtain definitive phase maps for all minerals. Contrast images were generated after selecting characteristic peak intensity ratios for every mineral. For example, the peaks at 185.1 and 675.5 cm −1 were chosen for biotite. This characteristic peak intensity ratio was used to generate a contrast image whose gray value at every location is proportional to the ratio ranging from − 10 to + 10. This image was filtered for the required range of characteristic peak intensity ratios (0 to 2.7 for biotite). In this image, all the pixels that have a ratio of 0 or smaller and 2.7 or greater get assigned to a value of 0 and 255 respectively in the pixel scale. The range was so chosen such that the maximum observed ratio for a particular set of characteristic peaks was set as the upper limit. Using ImageJ, this image was then subjected to a Fourier transform, and subsequently a bandpass filter was applied to segment the mineral of interest from the rest. A histogram of gray values of this image was generated which had peaks corresponding to minerals present and absent. This image was then thresholded corresponding to the peaks of required minerals to generate phase maps which were essentially binary images showing presence and absence of minerals. The stepwise results for the selected tile are shown in Fig. 3. This process was adopted for minerals observed in all the tiles to obtain definitive phase maps.
Electron imaging. Electron imaging and EDS mapping was done on unpolished, flat specimens using an environmental scanning electron microscope coupled with an Energy Dispersive X-ray Spectrometer (EDS). Secondary electron and back scattered images were captured in low vacuum mode with an accelerating voltage of 10.00 kV, spot size of 5 nm, and a chamber pressure of 1 Torr. Images were obtained with a working distance of 10 mm and 40X magnification in such a way that the entire scan area observed in the Raman microscope was visible. The EDS maps were generated with a dwell time of 250 µs and a resolution of 1024 × 800 pixels using a EDAX light element EDS module. Si, Al, Fe, Mg, K, Na, and Ca were selected based on preliminary EDS scans performed on the scan area as well as knowledge of the mineral compositions. The biotite-1 mineral (Fig. 5d) observed from Raman imaging was used as a marker to locate the same area on the SEM, using the backscattered image. This image was used to align and orient the EDS images with the optical image obtained from the Raman spectrometer, which ensured that Raman and EDS showed the same area. The raw images obtained using both techniques were slightly larger (~ 2.5 × 2.5 mm) than the images that were adjusted for orientation, translation, and finally cropped to the same size (2.21 × 2.36 mm).