Elsevier

Chemosphere

Volume 286, Part 2, January 2022, 131736
Chemosphere

Identification and visualisation of microplastics via PCA to decode Raman spectrum matrix towards imaging

https://doi.org/10.1016/j.chemosphere.2021.131736Get rights and content

Highlights

  • Raman imaging enables the direct visualisation and identification of microplastics.

  • Logic-based and PCA-based algorithm are compared to map image.

  • Logic-based algorithm can merge several images mapped at different characteristic peaks into one to increase the signal-noise ratio.

  • PCA-based algorithm can decode the Raman spectrum matrix in the absence of the standard Raman spectrum.

Abstract

To visualise microplastics and nanoplastics via Raman imaging, we need to scan the sample surface over a pixel array to collect Raman spectra as a matrix. The challenge is how to decode this spectrum matrix to map accurate and meaningful Raman images. This study compares two decoding approaches. The first approach is used when the sample contains several known types of microplastics whose standard spectra are available. We can map the Raman intensity at selected characteristic peaks as images. In order to increase the image certainty, we employ a logic-based algorithm to merge several images that are simultaneously mapped at several characteristic peaks to one image. However, the rest of the signals other than the selected peaks are ignored, meaning a low signal-noise ratio. The second approach for decoding is used when samples are complicated and standard spectra are not available. We employ principal component analysis (PCA) to decode the spectrum matrix. By selecting principal components (PC) and generating PC score curves to mimic the Raman spectrum, we can justify and assign the suspected items to microplastics and other materials. By mapping the PC loadings as images, microplastics and other materials can be simultaneously visualised. We analyse a sample containing two known microplastics to validate the effectiveness of the PCA-based algorithm. We then apply this method to analyse “unknown” microplastics printed on paper to extract Raman spectra from the complicated background and individually assign the images to paper fabric/additive, black carbon and microplastics, etc. Overall, the PCA-based algorithm shows some advantages and suggests a further step to decode Raman spectrum matrices towards machine learning.

Introduction

Environmental pollution resulting from microplastics has sparked global concerns due to their potentially adverse effects (Granek et al., 2020; Karbalaei et al., 2018; Sighicelli et al., 2018). Consequently, as emerging pollutants, microplastics have aroused considerable research. To address microplastic pollution issues, it is essential to establish accurate techniques to identify and quantify microplastics in environmental samples (Silva et al., 2018). The frequently used analytical methods include optical microscopy (Rodríguez-Seijo and Pereira, 2017; Sierra et al., 2020), scanning electron microscopy (SEM) and energy dispersive X-ray spectroscopy (EDS) (Blair et al., 2019; Gniadek and Dąbrowska, 2019; Tiwari et al., 2019; Wang et al., 2017), vibrational micro-spectroscopy (Raman spectroscopy and Fourier transform infrared (FTIR)) (Crichton et al., 2017; Käppler et al., 2016; Lenz et al., 2015; Primpke et al., 2017; Schymanski et al., 2018), thermal decomposition followed by mass spectroscopy (Dümichen et al., 2015; Fischer and Scholz-Böttcher, 2017; Hendrickson et al., 2018; Hermabessiere et al., 2018), and the combination of these methods (Shim et al., 2017). Among these methods, Raman is attracting particular attention, mainly due to its capability to enable non-destructive analysis of microplastics and even nanoplastics (down to 100 nm in size) (Araujo et al., 2018; Sobhani et al., 2020a). Another important advantage is that when confocal Raman is employed to scan a sample surface, the direct visualisation of the microplastics (5 mm- 1 μm) and nanoplastics (<1 μm) can be achieved via mapping a Raman image after collecting Raman spectra at the pixels of a pre-defined area (Sobhani et al., 2020a; Xu et al., 2020).

However, interpreting or decoding Raman data can be an arduous task (von der Esch et al., 2020). For example, Raman mapping usually scans 10-10,000 pixels for a sample to collect 10-10,000 sets of spectra. Consequently, a large data matrix comprising a multitude of signal intensities at different wavenumbers is created. Decoding such matrices to image and visualise targets can be conducted manually, but it is often time-consuming and commonly results in signal loss (Fang et al., 2020; Sobhani et al., 2020a). To illustrate, scanning a sample area of 10 μm × 10 μm with a pixel size of 0.1 μm × 0.1 μm creates 100 × 100 spectra as a matrix. To map a Raman image, the signal intensity at a selected wavenumber is extracted. In order to avoid interference, the wavenumber is selected corresponding to the peak that is characteristic and unique to a specific plastic, if the standard Raman spectrum is available. These intensity data are then used to generate an image relating to the scanned area. By doing so, the signal-noise ratio of a Raman image (or image certainty) from a matrix of 100 × 100 spectra can be significantly different from the signal-noise ratio from an individual spectrum. Although this process could be realised by software, the produced image only provides limited information at the selected wavenumber. In other words, the majority of the spectrum is not selected and analysed, meaning a low signal-noise ratio. In order to increase the signal-noise ratio, we can generate several Raman images by mapping multiple peaks. We then apply a logic-based algorithm to merge these images into one. Nevertheless, this method still has some limitations: i) loss of signal, because only several peaks from the spectrum are analysed, and the rest are ignored, ii) time-consuming computation given that the ASCII code of the matrix is often tens or even hundreds of megabytes, and iii) failure to cover meaningful signals from the weak spectrum/peak, either due to the background interference, or being shielded/overlapped by neighbouring signals, etc. iv) the logic-based algorithm itself can also lead to the signal loss, due to the threshold-based analysis to merge images.

Key information encompassed within Raman spectrum matrix is usually widely distributed throughout the dataset (Shinzawa et al., 2009). It is therefore important to interpret the Raman data in such a way that information loss and background interference are minimised (Hasegawa et al., 2000). To effectively analyse a large data matrix, several techniques are available (Araujo et al., 2018; Ilie et al., 2017; Levermore et al., 2020; von der Esch et al., 2020; Zhang et al., 2005). Some studies attempted to use automated algorithm-based particle selection prior to Raman analysis to reduce the number of pixels to be scanned, thereby shortening operation time (Araujo et al., 2018; von der Esch et al., 2020). Standard normal variance (SNV) was also used to achieve baseline correction and to smooth the Raman data of air-borne microplastic samples (Levermore et al., 2020). Principal component analysis (PCA) is another effective statistical method for handling a complex Raman data matrix, as it is able to reduce the dimensionality of a dataset while preserving the most critical features of the data (Halstead et al., 2018; Jolliffe and Cadima, 2016).

Over the past few years, PCA has been increasingly used to decode a Raman spectrum matrix to identify and differentiate particles, materials or cells in various fields of research, such as chemistry and environmental science (Ai et al., 2021; Ali et al., 2020; An et al., 2021; Ilie et al., 2017; Jin et al., 2021; Laptenok et al., 2020; Maslova et al., 2017; Pořízka et al., 2018; Samyn et al., 2020; Silva et al., 2020), biology (Falamas et al., 2021; Halstead et al., 2018; He et al., 2020; Meksiarun et al., 2017), microbiology (Bashir et al., 2021; Hanson et al., 2017; Yan et al., 2021), and forensic science (Buzzini et al., 2021; Osmani et al., 2020; Thomas et al., 2021). Therefore, PCA is deemed suitable for analysing microplastics via decoding their Raman spectrum matrix. In fact, PCA has already been used to decompose the spectrum matrix from FTIR, a similar and complementary technique to Raman spectroscopy (Gál et al., 2020; Li et al., 2019; Mecozzi and Nisini, 2019; Wander et al., 2020; Xu et al., 2018). Nevertheless, further research is still needed to verify the effectiveness of PCA in the context of Raman analysis of microplastic, to automatically decode the spectrum matrix for mapping and imaging. For example, when the Raman signal is mixed and shielded by a complex background, and when the standard spectrum of the plastic target is not available, we don't know whether the PCA-based algorithm is still a helpful decoding approach or not, and how effective it is when compared with the logic-based algorithm.

In this study we assume that the Raman spectrum matrix can be directly decomposed, via PCA, to two new matrices, including a PC score matrix and a PC loading matrix. The former is supposed to be related to the spectrum profile (“PCA spectrum”), and the latter to the Raman intensity (or vice versa). In this case, the mapping via the decomposed Raman intensity matrix will ideally generate an image that covers the whole PCA spectrum profile and thus contains all the key features of the whole Raman spectrum, meaning a higher signal-noise ratio than that from an individual Raman peak. Furthermore, the standard Raman spectrum of the target microplastic is not directly needed for the decoding and mapping, unless it is required to justify the decoding accuracy and to identify the plastic via comparison of the decoded spectrum (PCA spectrum) with the standard one. To confirm this assumption, in this study, a PCA-based algorithm is used to analyse and decode several Raman spectrum matrices, first containing two “known” microplastics to get validated, then containing several “unknown” items with complicated background. The decoded results are compared with the logic-based algorithm that we have used previously, to assess the advantages and limitations of PCA. It is expected that this study will be useful in directing future investigations into improving the interpretability of Raman spectroscopy in microplastic research, particularly when a spectrum matrix analysis is needed. The findings will potentially contribute to the knowledge with regards to the automation of microplastic Raman mapping via techniques such as machine learning.

Section snippets

Chemicals and microplastics

All microplastics and chemicals including sulphuric acid (H2SO4) and hydrogen peroxide (H2O2), were purchased from Sigma-Aldrich (Australia) and used as received unless further indicated. A microplastic mixture including polyethene (PE) and polyvinyl chloride (PVC) was selected as a model to validate the algorithm analysis (Sobhani et al., 2019).

Sample holder

A glass slide was cleaned by dipping into Piranha solution (2:1 H2SO4: H2O2, v/v) (be careful, this solution reacts vigorously with organic compounds!)

Colour off-setting

When the Raman intensity is mapped, we must select the characteristic peaks of a target plastic to avoid any overlap and interference. In this case, the standard Raman spectrum should be available, as shown in Fig. 1(a). For the mixture of PVC and PE, we selected the marked peak at 1059 cm−1 for PE and 695 cm−1 for PVC (Sobhani et al., 2019). At those two peaks, their Raman signals do not overlap with each other, which means there is almost no interference from each other for imaging.

The

Conclusion

We compared different methods of decoding the Raman spectrum matrix, including colour off-setting, logic-based algorithm and PCA-based algorithm. The logic-based algorithm can handle several characteristic peaks for intensity mapping when the standard Raman spectrum is available. However, the remaining parts of the Raman spectrum have been ignored, which means a low signal-noise ratio. On the other hand, PCA can be effectively used to automatically extract and decode the key information from

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

The authors appreciate the funding support from CRC CARE and the University of Newcastle, Australia. For the Raman measurements, we also acknowledge the use and support of the South Australian node of Microscopy Australia (formerly known as AMMRF) at Flinders University, South Australia, and the Key Lab of Urban Environment and Health, Institute of Urban Environment, Chinese Academy of Sciences, Xiamen 361021, P. R. China.

References (67)

  • A.H. Kuptsov

    Fourier transform Raman spectroscopic investigation of paper

    Vib. Spectrosc.

    (1994)
  • S.P. Laptenok et al.

    Stimulated Raman microspectroscopy as a new method to classify microfibers from environmental samples

    Environ. Pollut.

    (2020)
  • S. Lee et al.

    Characterization of VOCs, ozone, and PM10 emissions from office equipment in an environmental chamber

    Build. Environ.

    (2001)
  • R. Lenz et al.

    A critical assessment of visual identification of marine microplastic using Raman spectroscopy for analysis improvement

    Mar. Pollut. Bull.

    (2015)
  • X. Li et al.

    Enhancement in adsorption potential of microplastics in sewage sludge for metal pollutants after the wastewater treatment process

    Water Res.

    (2019)
  • O.A. Maslova et al.

    Raman imaging and principal component analysis-based data processing on uranium oxide ceramics

    Mater. Char.

    (2017)
  • M. Mecozzi et al.

    The differentiation of biodegradable and non-biodegradable polyethylene terephthalate (PET) samples by FTIR spectroscopy: a potential support for the structural differentiation of PET in environmental analysis

    Infrared Phys. Technol.

    (2019)
  • P. Pořízka et al.

    On the utilization of principal component analysis in laser-induced breakdown spectroscopy data analysis, a review

    Spectrochim. Acta B Atom Spectrosc.

    (2018)
  • A. Rodríguez-Seijo et al.

    Chapter 3-morphological and physical characterization of microplastics

    Compr. Anal. Chem.

    (2017)
  • J. Ruan et al.

    Controlling measures of micro-plastic and nano pollutants: a short review of disposing waste toners

    Environ. Int.

    (2018)
  • D. Schymanski et al.

    Analysis of microplastics in water by micro-Raman spectroscopy: release of plastic particles from different packaging into mineral water

    Water Res.

    (2018)
  • M. Sighicelli et al.

    Microplastic pollution in the surface waters of Italian Subalpine Lakes

    Environ. Pollut.

    (2018)
  • A.B. Silva et al.

    Microplastics in the environment: challenges in analytical chemistry - a review

    Anal. Chim. Acta

    (2018)
  • D.L. Silva et al.

    Raman spectroscopy analysis of number of layers in mass-produced graphene flakes

    Carbon

    (2020)
  • Z. Sobhani et al.

    Identification and visualisation of microplastics by Raman mapping

    Anal. Chim. Acta

    (2019)
  • Z. Sobhani et al.

    Identification and visualisation of microplastics/nanoplastics by Raman imaging (i): down to 100 nm

    Water Res.

    (2020)
  • Z. Sobhani et al.

    Identification and visualisation of microplastics/nanoplastics by Raman imaging (i): down to 100 nm

    Water Res.

    (2020)
  • M. Tiwari et al.

    Distribution and characterization of microplastics in beach sand from three different Indian coastal environments

    Mar. Pollut. Bull.

    (2019)
  • Z.-M. Wang et al.

    SEM/EDS and optical microscopy analyses of microplastics in ocean trawl and fish guts

    Sci. Total Environ.

    (2017)
  • P. Xu et al.

    Microplastic risk assessment in surface waters: a case study in the Changjiang Estuary, China

    Mar. Pollut. Bull.

    (2018)
  • L. Zhang et al.

    Multivariate data analysis for Raman imaging of a model pharmaceutical tablet

    Anal. Chim. Acta

    (2005)
  • J. Ai et al.

    Biochar catalyzed dechlorination – which biochar properties matter

    J. Hazard Mater.

    (2021)
  • W. An et al.

    Occurrence, spatiotemporal distribution, seasonal and annual variation, and source apportionment of poly– and perfluoroalkyl substances (PFASs) in the northwest of Tai Lake Basin, China

    J. Hazard Mater.

    (2021)
  • Cited by (55)

    View all citing articles on Scopus
    View full text