In collaboration with Isao Noda, I have been using Raman spectroscopy to study the properties of the bioplastic polyhydroxybutyrate hexanoate (PHBHx), which depend on the percentage of hexanoate (which produces propyl side branches to the polymer chain). The percentages of hexanoate determine the maximum crystallinity that the polymer can experience, and that determines its physical and chemical properties, which include its optical clarity, dyability, flexibility, and thermal properties (melting and glass transition temperature). It is useful to have an easy way to determine the composition; in this column, we describe how Raman spectroscopy was used to show that this is feasible at an accuracy greater than expected.

The structure of polyhydroxybutyrate hexanoate (PHBHx) is shown in Figure 1. Although the PHB homopolymer is highly crystalline, the insertion of the propyl group highlighted in red disrupts the ability of PHB to crystallize, resulting in a lower maximum crystallinity. The presence of the propyl group adds two methylene groups and can change the interaction of the methyl end group with the chain because it is separated from the chain by the two methylene groups. Thus, if one wants to quantitate the concentration of propyl units, it would be of interest to look in the region of the CH-stretching bands between 2700 and 3000 cm

. However, this cannot be easily done in the solid phase because of the random, consistent errors that would be introduced by the spectral changes, which would stem from crystallinity and orientation. Thus, we can either measure the spectra in the melt or in the solution. Because the melting temperatures may vary, thus producing other ambiguities in the spectra, we opted to look at the solution phase. The obvious problem with the solution phase was the presence of the CH bands of the solvent. But, even then, the obvious solution to this problem would be to dissolve the samples in a deuterated solvent, which is what we did. PHBHx is soluble in chloroform (CHCl

), so we used deuterated chloroform (CDCl

Figure 2 shows the spectra between 1900 and 3150 cm

 collected on the XploRA using the 1800 groves/mm grating, a 532-nm laser, the 45° mirror enabling access to the sample vials with the 4-cm lens, and the wavelength fixed so that the CD stretch at 2254 cm

 is observed simultaneously with the CH stretches of the polymer without scanning the monochromator, but still with good spectral resolution (~1.4 cm

 per pixel, with the sharpest band being the CD stretch at less than 7 cm

). Figure 2 is displayed with the CD stretch scaled the same in all spectra. Figure 3 shows the CH region extracted and scaled, so that the band at 2940 cm

 is the same in all spectra. It was not clear a priori that the band at 2940 cm

 would be the best band to use for normalization. Had the entire spectrum been acquired, we could have used the >C=O band whose concentration is independent of the substitution (refer to Figure 1). But, as you will see, the normalization we used produced good results.

The next step that was performed before trying to produce a model for prediction was to subtract the spectrum of the 0% hexanoate sample (that is, PHB homopolymer). The result of this operation is shown on Figure 4. In support of this method, we see that the subtracted signals more or less scale with the percentage substitution (I color code the spectra in the sense of a rainbow to easily visualize the behavior).

Using LabSpec to Produce a Model for Prediction 

Now that I have a data set from a sample set with known comonomer percentages and I can see that the spectra are varying systematically with the percentages, it is time to try using partial least squares (PLS) to determine if a quantitative predictive model can be derived from the spectra.

In LabSpec, this is easily done by importing a multifile with all of the relevant spectra. The user then inputs the values for the variable; in this case, the only variable was the percentage of Hx. The program automatically creates a model. Figure 5 shows the result of the model creation. The two blue tables were captured directly from LabSpec. The graph was created by simply copying the table on the left and pasting it into Excel. I have inserted the black line to indicate what a perfect fit would be (input value = calculated value). The plot in the middle was created in Excel because the 

-axis in the LabSpec plot was in the spectrum number rather than the percentage. Curiously, Excel sometimes reported that the plot was correlated. I would hope so, since we would like the predicted value to be a good approximation of the input value!

Polymers have become extremely useful materials for a myriad of commercial applications. In particular, their properties can be engineered for the degree of crystallinity, which, in the case of PHBHx, depends on the concentration of the hexanoate. The sample set that was studied here has seven samples with Hx concentrations between 0 and 13%. In the manufacturing process, it is useful to confirm a material’s concentration before initiating the manufacturing of the final product with a large amount of material. The results shown here indicate that a simple Raman measurement can estimate the concentration of an unknown in the low percentage range to ±0.5%. The immediate implication for the product quality control (QC) during the commercial production of many thousands of pounds of the material is obvious.

Articles

Column: Molecular Spectroscopy Workbench

Raman spectroscopy has the capability of accurately determining the physical and chemical properties of bioplastics (polyhydroxybutyrate hexanoate [PHBHx]), which helps determine the maximum crystallinity a polymer can experience.

Calibrating the Composition of a Copolymer

In Part I (February 2023) of this two-part series on artificial intelligence (AI), and its subfield machine learning (ML), we presented the variety of chemometric algorithms used to compare AI, ML, and chemometrics. These algorithms included those used for classification, regression, clustering, ensemble learning, signal processing, and component analysis. Now, in Part II, we discuss the applications of AI to electronic and vibrational spectroscopy. We also touch on some applications of deep learning (DL), which is a subfield of machine learning where more complex artificial neural networks (ANNs) with more hidden layers are used. This column article includes a number of selected references that discuss the application of AI in analytical chemistry and in molecular spectroscopy. We give a few early and late examples of AI and ML as applied to different vibrational spectroscopy methods, such as Raman, infrared (FT-IR), near-infrared (NIR), and ultraviolet–visible (UV-vis) spectroscopic techniques. This article is intended only as a sampling of the numerous research manuscripts addressing this subject.

Since 2010, there has been a rapid growth in the application of artificial intelligence (AI) approaches to provide data analysis and modeling solutions for a variety of analytical chemistry applications. AI, when properly applied and evaluated using domain knowledge, is a useful tool for routine processing of large quantities of data and for assisting humans in performing routine or repetitive tasks with high repeatability and accuracy. AI also has significant potential for automating well-understood complex tasks, and for advancing discovery of new molecules and materials by integrating and combining known physical science principles with new innovative concepts and approaches. We bring to the reader’s attention that, strictly speaking, AI does not always indicate the use of neural networks. Neural networks are one of the many techniques used in AI, but they are not the only ones. AI encompasses a wide range of techniques and approaches, including rule-based systems, decision trees, genetic algorithms, and others.

General Trends for AI in Analytical Chemistry

AI use is increasing rapidly for analytical chemistry applications. Review articles are a useful starting place to explore this growing field. A tutorial review of AI applications in analytical chemistry was published in 2021 (1). This review discusses the application advances in computer processing power and algorithms of AI in chemistry for creating new compounds and novel approaches in organic synthesis, drug discovery, and design of materials. The authors report that the use of AI to support analytical purposes has been mostly limited to data analysis methods, such as image recognition, vibrational spectroscopy, and mass spectrometry. The review describes basic concepts related to implementing AI, and states that there are many new opportunities for allying AI technology to spectroscopy and separation science as well as sensor output and design.

Another valuable review paper describes the trends in AI and the relationship of machine learning to chemometrics (2). In fact, this paper refers to chemometrics as an AI-based method, along with machine learning and deep learning (DL, a subfield of machine learning [ML]). The paper proposes that AI methods lead to an improved understanding of data. The authors state that AI, when used to its full extent, allows improved insight into the various processes, interactions, and characteristics of sample analysis data. The review highlights that AI-based techniques have been applied to chemical data since the 1970s and many AI approaches were being used by 2020. The paper specifically describes inverse modeling, preprocessing methods, and data modeling techniques as applied to spectral and image analysis for several analytical techniques.

Vibrational spectroscopy techniques, such as Fourier transform infrared (FT-IR) and Raman spectroscopy, have been demonstrated to be highly successful analytical methods and have proven to be low cost, require minimum sample preparation, be non-destructive, and yield highly valuable chemical information. There is a requirement for improvement and automation in data processing for classification and visualization of spectrochemical data measured from biological and medical samples. Multivariate classification is useful for assigning unknown samples to predefined groups and for information extraction from data. For this purpose, a tutorial article, representing the prerequisite algorithm requirements for machine learning approaches, has been published to demonstrate multivariate classification analysis using near-infrared (NIR), FT-IR, and Raman spectroscopy data (3). This paper highlights such steps as data preprocessing, data selection, feature extraction, classification, and model validation. These mathematical processes are essential for computation of practical spectrochemical classification models intended for analysis of biological samples.

A paper demonstrated that convolutional neural networks (CNNs), a basic AI and ML tool, can be used to effectively classify vibrational spectroscopy data as well as to identify important spectral regions (4). CNNs are capable of image classification, and can be used to learn or define various visual interpretations of spectroscopic data. CNNs are able to reduce the requirements for rigorous data preprocessing, and are able to identify the most important spectral regions for analyzing the desired spectral features of interest. Spectroscopic data is most often preprocessed as part of the chemometric modeling work. Such preprocessing steps include various types of baseline correction, scatter correction, noise reduction or removal, feature selection, derivative processing, and other steps, which are applied to the spectra prior to model building. As a standard practice, preprocessing has traditionally been considered essential for optimizing standard chemometric models. A simple CNN architecture, using a single convolutional layer (shallow CNN), has demonstrated improved performance as compared to standard classification algorithms such as partial least squares (PLS) regression. Classification accuracy for CNN vs. PLS was reported to be 86% vs 62%, respectively, for non-preprocessed test data, and 96% and 89%, respectively, for preprocessed data. The CNN algorithm was also able to identify important spectral regions, in order to assist in qualitative interpretation of classification results.

Early work in implementing AI for spectroscopic applications described the identification of polyatomic molecules from their molecular spectra (5). This research concluded that AI systems provided one of the most promising potential developments for the future of analytical molecular spectroscopy, noting that AI systems presented a more dynamic analytical solution set. It was reported that AI systems are able to compare measured spectra to computer-generated spectra, rather than just compare measured spectra to an archived spectral database. Such an AI approach offers more potential than manual methods to extract additional information from spectral analysis experiments.

In a series of papers, Luinge and colleagues have described the development of an AI expert system for spectral interpretation and for the elucidation of chemical structures from combined spectroscopic data (EXSPEC) (6–9). These papers describe the rationale and programming involved in creating such a system for interpretation of infrared, mass, and nuclear magnetic resonance (NMR) spectra for an earlier generation of computer systems.

In research to better connect Raman spectroscopy measurements of biomedical and clinical samples to medical diagnosis and decision-making, an AI system was designed as a way to understand the chemical pathways and the progression of disease (10). This approach has potential use for the future of personalized medicine. The initial stages of this AI system were designed to optimize spectral preprocessing in order to reduce noise, fluorescence and baseline effects, while employing the use of chemometrics. Specifically, this research analyzed breast cancer tissue samples using Raman spectroscopy combined with principal component (PCA) and linear discriminate analysis (LDA). For this work, tissue microarray (TMA) breast biopsies were classified into multiple groups, such as luminal A, luminal B, HER2, and triple negative subtypes. The classification accuracy results of applying the AI system to the Raman spectra, which contains chemical information on tissue lipid, collagen, and nucleic acid content, were 70%, 100%, 90%, and 96.7% for luminal A, luminal B, HER2, and triple negative subtypes, respectively. This result is quite encouraging for further disease tissue classification research.

Because vibrational spectroscopic techniques have shown significant progress in assessing biomedical samples, they are being further investigated using automated and AI data processing. When properly implemented, AI is capable of identifying and connecting meaningful chemical and diagnostic relationships for clinical samples measured using infrared or Raman spectroscopy. A special article assesses the current use of AI in biomedical vibrational spectroscopy applications (11).

ML, as a subfield of AI, is being used to extract, connect, and summarize information from large and complex analytical datasets for separation science, mass spectrometry, NMR, and atomic and molecular spectroscopy. The advancing of applications of AI technology is certainly the case in Raman and surface-enhanced Raman spectroscopy (SERS) techniques, which involve large databases of complex vibrational spectra. Applying chemometric and analysis methods using manual techniques is no longer satisfactory for biomedical or diagnostic work. To implement accurate and powerful data analysis for Raman and SERS experiments, AI systems have recently been explored and implemented. A 2020 review article delves into the current uses of AI or machine (deep) learning techniques employed in Raman spectroscopy, including SERS (12).

An AI approach was applied to classify biomedical samples using Raman spectroscopy for the purposes of medical diagnosis and decision-making (13). The functional aspects of the AI system included automation of noise filtering (via a fuzzy controller), fluorescence background identification and correction, baseline optimization (using genetic algorithms), spectral normalization and scattering compensation (using the standard normal variate [SNV] algorithm), multivariate statistical analysis, sample data clustering, and projected decision-making. The AI program useful for classification of biomedical samples includes integration of fuzzy control, genetic algorithms, and principal component analysis (PCA), as well as system identification. The main goal for implementing the AI system is to incorporate a Raman probe on the end effectors of medical robots, to provide real-time information during robotic surgery.

A review paper, written in 2022, summarizes the application of Raman spectroscopy for noninvasive identification of mixture composition, including spectral preprocessing, and a detailed discussion of the application of AI analysis in Raman methods (14).

A 2022 research paper discusses the use of ML in FT-Raman spectroscopy for trademark fingerprint assessment of fruit spirits (15). The combination of ML and Raman spectra was used in an attempt to compose reliable models for the classification of fruit spirits for trademark, geographical, and botanical origin. Raman spectroscopy is an excellent measurement technique for samples with high water content, such as alcoholic beverages, due to the relatively weak Raman water bending mode in the fingerprint region. The discriminant analysis for the optimized classification model was reported to be 96.2% accurate. Geographical origin prediction was also achieved, but classification of botanical origin recognition was not successful. It was noted that spectral data used for the discriminant models involved both Stokes and anti-Stokes spectra, as well as individual spectral regions (windows). Several ML algorithm approaches were attempted. The paper offers an excellent discussion of the aspects of ML and Raman.

Raman spectroscopy provides an analytical method for non-destructively measuring the molecular structure for biomedical samples. Because of these benefits, Raman provides a suitable analytical technique for the assessment of inflammatory skin diseases and skin inflammation. The combination of AI with Raman spectroscopy to predict skin inflammation with high sensitivity and specificity was tested (16). For this feasibility work, Raman spectra, using excitation at 785 nm, were collected on mouse ear tissue where inflammation had been chemically induced. Principal component analysis (PCA) and AI were evaluated using receiver operating characteristic (ROC) curves to assess inflammation. The accuracy rate without AI implementation was 80.0% with an area under the curve (AUC) of 0.864. With AI implementation added, the accuracy rate was improved to 93.1% with an AUC of 0.972. The current findings demonstrate AI and Raman are able to provide highly accurate information on the presence and pathology of skin inflammation.

Raman spectroscopy has become known for its speed of analysis, limited requirements for sample preparation, ability to measure samples with high water content, and cost effectiveness. With these features, Raman is quite suitable for analysis of foods and beverages, and has been studied relative to classification and authenticity of such products. In one study, vegetable cold-pressed oils, which are of interest for nutritional health benefits, were tested for authentication analysis (17). In this study, oils from sesame, hemp, walnut, linseed, pumpkin, and sea buckthorn were tested for authentication using Raman spectra combined with ML. By using Raman spectra and ML, both adulteration and the degree of adulteration were successfully detected.

Improvements in chemometrics and data analysis techniques for Raman spectroscopy have allowed Raman to be used in more applications than ever before. However, there are still important issues to overcome when applying Raman spectroscopy that still limit its ability to fulfill its full analytical potential. These limitations include standardization of large datasets, intensity and baseline correction, fluorescence interference, measurement noise, changes in spectra relative to sample presentation variation, and so forth. In order to improve the quality and repeatability of Raman spectra, deep learning has been implemented as a strategy for enhanced correction of Raman spectral measurements. A book chapter devoted to these issues has been published with discussions, illustrations, and proposed solutions (18).

Gasoline quality prediction was tested using analytical data from gas chromatography (GC) and FT-IR on 45 gasoline samples (19). The AI system approach included preprocessing implementation of principal component analysis (PCA) and fuzzy C means (FCM) algorithms. The FT-IR spectra were compressed and denoised using a discrete wavelet analysis method. The preprocessed data were analyzed using a hybrid neural network and support vector machines (SVM) classifier. The authors report an approximately 100% correct classification for six different categories of the gasoline using this workflow.

Lubricant oil efficiency is essential for maintenance of lubricated surfaces. The presence of water, aging (oxidation and decomposition), and contaminants can all seriously affect tribological (friction, wear, and lubrication) efficiency of oils. When oil efficiency is lost, longevity of metal parts and maintenance costs are a concern. In recent research, attenuated total reflectance (ATR) and Fourier transform infrared spectroscopy (FT-IR) were combined with AI techniques to assess lubricant oil status (20). The AI method consisted of using artificial neural networks (ANNs) with linear discriminant analysis (LDA) to derive analysis models for lubricant oils for the presence of water, as well as oil age and viscosity. The reported success for identifying the degree of oil deterioration was approximately 100% for the parameter of weeks of aging, and 97.7% for the parameters of viscosity and the presence of water.

Cereal quality is important for global human nutrition. The use of IR spectroscopy, hyperspectral imaging (HSI), and AI provides opportunities for quality monitoring of food and agricultural products. The combination of IR and HSI with AI provides high potential for improved measurements of food quality. A review article of these topics is given that describes the technical background, instrumentation, and data processing (preprocessing, feature extraction, and modeling) using AI approaches (21). The principles, advantages, challenges and future trends of IR and HSI integrated with AI are explored in this review.

FT-IR spectroscopy, in its traditional implementation, has shown great potential for food analysis applications—it is fast, non-destructive (noninvasive), and is capable of yielding a large quantity of chemical information. Recent successful applications include determination of spoilage in meat and the presence of microbial growth. For over two decades, ML has been used in combination with FT-IR to detect meat quality (22). In basic research, chicken breasts were purchased from a national retailer, comminuted for 10 s, and left to spoil at room temperature for 24 h. Every hour, ATR-FT-IR measurements were taken from the meat surface using the ATR technique. The IR spectral data were compared to total viable microbial counts determined using classical microbiology plating methods. The ATR-FT-IR spectra were processed using partial least squares (PLS) and genetic programming to provide accurate estimation of bacterial loads within 60 s. The genetic programming was used to predict bacterial counts (bacteria/g) and onset of proteolysis (spoilage).

Near-infrared (NIR) spectroscopy has had a long and established history of involvement with the fields of computer automation, computer networking, discriminant analysis, regression methods, spectral preprocessing, and ML in the form of neural networks (ANN and CNN). The field of chemometrics in spectroscopy was largely developed using NIR. In the late 1970s, all-possible-combinations (APC) search algorithms were in practice for optimizing multiple linear regression (MLR) calibration models. These APC models and techniques were referred to by Tomas B. Hirschfeld as “artificial intelligence” techniques, more or less—we would now refer to these as automated calibration programs. APC was capable of determining the optimized equation without additional human intervention. NIR techniques then progressed to the routine application of principal components regression (PCR), PLS, and ANN for quantitative calibrations and for classification problems. With this multi-decade experience in chemometrics, NIR applications have proliferated into many areas of commercial and research applications (23). Here, we give a few examples.

One study using NIR spectroscopy has applied AI DL to spectral analysis for cereal product composition measurements (24). For this research, NIR spectra were extracted using a deep learning-stacked sparse autoencoder (SSAE) method, with the prediction model constructed using an affine transformation (AT) and an extreme learning machine (ELM) algorithm. The authors report improved performance as compared to other classical (non-AI) NIR analysis calibration methods.

NIR spectroscopy is a prolific analytical technique applied for quantitative and qualitative analysis in the pharmaceutical, food, agricultural, materials, petrochemical, medical, and environmental sectors (23). The subject of mathematical preprocessing of NIR spectra prior to the modeling step has been addressed in many papers and books, and is an essential step in optimizing calibration models. Preprocessing steps can minimize the effects of particle size (for solids) and scattering (for liquids), as well as reduce spurious noise effects—all of which improve the performance of qualitative and quantitative calibrations.

A published paper describes the reduction of spectral noise using extreme learning machines, an approach which has demonstrated improvements in regression models and in large dataset classification work (25). This research reported the use of a parallel layer extreme learning machine (PL-ELM) algorithm, and constrained (C)- PL-ELM, which has an architecture based on a non-linear layer in parallel with another non-linear layer. The details of this algorithm structure are explained within the paper. The method was tested on six real-life datasets, and the results for both regression and classification problems indicated that the root mean square error (RMSE) and accuracy were improved by applying this technique.

NIR has been used in the brewing industry over the years for analysis of raw ingredients and finished beverages. The analysis of beer faults early in the fermentation process can improve final product quality and consumer acceptance. In one study, AI was integrated with NIR spectra and the signal from an electronic nose (e-nose) (26). For this work, a commercial beer was spiked with 18 common beer faults plus the control aroma. The 19 aroma profiles were tested for classification analysis using an ML model. In total, six different ML models were developed. A Bayesian regularization algorithm was selected as the optimum method with model classification performance ranging from 95.3% to 98.9% accuracy without statistical signs of under- or overfitting.

Ultraviolet-visible (UV-vis) spectroscopy is widely recognized as a workhorse for general analytical laboratory, process, and field measurements. Water quality monitoring is just one of the many applications where UV-vis exhibits its beneficial features, such as speed, environmental friendliness, non-destructive sample handling, and high analytical sensitivity.

Water quality measurements using UV-vis for process or field applications have been quite common over the years. The application of ML to UV-vis spectroscopy provides an automated and “smart” platform for routine water analysis and screening (27). In this work, ML in the form of ANNs has been evaluated for simplifying water quality monitoring using UV–vis. Field water samples were collected, and spectra measured from 200 to 800 nm in a laboratory setting. Optimized CNN and PLS models were compared for water parameters resulting in an 

 for total organic carbon (TOC) between predicted values and reference values to be 0.927 for PLS and 0.953 for the CNN model. Also, 

 between predicted values and true values for total suspended solids (TSS) concentrations was 0.827 with the PLS model and 0.915 with the CNN model. It was concluded that CNN would be the preferred approach as compared to PLS for online water quality monitoring using UV–vis spectroscopy.

AI is being integrated with quantum chemistry and statistical mechanics for spectroscopic studies. One paper discusses the investigation of medium-to-large size chromophores in condensed phases using UV-vis spectroscopy (28), where a prototypical nitroxide radical (TEMPO) was evaluated in different solvents. The paper explores the reliability, effectiveness, and robustness of the new AI platform for application to complex UV-vis spectroscopic studies.

From our brief survey of a few applications of AI to spectroscopic studies, we are able to discern that this advancing field will continue to change the development and capabilities of spectroscopic data analysis and implementation well into the future. AI has great positive potential when used properly with expert domain knowledge, and also great potential for misuse when blindly applied or when input data is inadequate or not well understood. Possibly we will be able to write future columns on this fascinating and rapidly emerging topic—we invite your feedback about this subject.

Ayres, L. B.; Gomez, F. J.; Linton, J. R.; Silva, M. F.; Garcia, C. D. Taking the Leap Between Analytical Chemistry and Artificial Intelligence: A Tutorial Review. 

(2) Houhou, R.; Bocklitz, T. Trends in Artificial Intelligence, Machine Learning, and Chemometrics Applied to Chemical Data. 

(3) Morais, C. L.; Lima, K. M.; Singh, M.; Martin, F. L. Tutorial: Multivariate Classification for Vibrational Spectroscopy in Biological Samples. 

(4) Acquarelli, J.; van Laarhoven, T.; Gerretzen, J.; Tran, T.N.; Buydens, L.M.; Marchiori, E. Convolutional Neural Networks for Vibrational Spectroscopic Data Analysis. 

(5) Gribov, L. A. Application of Artificial Intelligence Systems in Molecular Spectroscopy. 

(6) Luinge, H. J.; Van’t Klooster, H. A. Artificial Intelligence Used for the Interpretation of Combined Spectral Data. 

(7) Kleywegt, G. J.; Luinge, H. J.; Van’t Klooster, H. A. Artificial Intelligence used for the interpretation of combined spectral data: Part II. PEGASUS: A PROLOG program for the generation of acyclic molecular structures. 

(8) Luinge, H. J.; Kleywegt, G. J.; Van’t Klooster, H. A.; van der Maas, J. H. Artificial Intelligence Used for the Interpretation of Combined Spectral Data. 3. Automated Generation of Interpretation Rules for Infrared Spectral Data. 

(9) Luinge, H. J.; van der Maas, J. H. Artificial Intelligence for the Interpretation of Combined Spectral Data: Design and Development of a Spectrum Interpreter. 

(10) Talari, A. C.; Rehman, S.; Rehman, I. U. Advancing Cancer Diagnostics with Artificial Intelligence and Spectroscopy: Identifying Chemical Changes Associated with Breast Cancer. 

(11) Mantsch, H. H. Biomedical Vibrational Spectroscopy in the Era of Artificial Intelligence. 

(12) Lussier, F.; Thibault, V.; Charron, B.; Wallace, G. Q.; Masson, J. F. Deep Learning and Artificial Intelligence Methods for Raman and Surface-Enhanced Raman Scattering. 

(13) Ye, Z. Artificial-Intelligence Approach for Biomedical Sample Characterization Using Raman Spectroscopy. 

(14) Pan, L.; Zhang, P.; Daengngam, C.; Peng, S.; Chongcheawchamnan, M. A Review of Artificial Intelligence Methods Combined with Raman Spectroscopy to Identify the Composition of Substances. 

(15) Magdas, D. A.; David, M.; Berghian-Grosan, C. Fruit Spirits Fingerprint Pointed Out Through Artificial Intelligence and FT-Raman Spectroscopy. 

(16) Kanemura, Y.; Kanazawa, M.; Hashimoto, S.; et al. Assessment of Skin Inflammation Using Near-Infrared Raman Spectroscopy Combined with Artificial Intelligence Analysis in an Animal Model. 

(17) Berghian-Grosan, C.; Magdas, D. A. Raman Spectroscopy and Machine-Learning for Edible Oils Evaluation. 

(18) Jinadasa, M. W. N.; Kahawalage, A. C.; Halstensen, M.; Skeie, N. O.; Jens, K. J. Deep Learning Approach for Raman Spectroscopy. 

Recent Developments in Atomic Force Microscopy and Raman Spectroscopy for Materials Characterization

(19) Brudzewski, K.; Kesik, A.; Kołodziejczyk, K.; Zborowska, U.; Ulaczyk, J. Gasoline Quality Prediction Using Gas Chromatography and FTIR Spectroscopy: An Artificial Intelligence Approach. 

(20) Chimeno-Trinchet, C.; Murru, C.; Díaz-García, M. E.; Fernández-González, A.; Badía-Laíño, R. Artificial Intelligence and Fourier Transform Infrared Spectroscopy for Evaluating Water-Mediated Degradation of Lubricant Oils. 

(21) An, D.; Zhang, L.; Liu, Z.; Liu, J.; Wei, Y. Advances in Infrared Spectroscopy and Hyperspectral Imaging Combined with Artificial Intelligence for the Detection of Cereals Quality. 

(22) Ellis, D. I.; Broadhurst, D.; Kell, D. B.; Rowland, J. J.; Goodacre, R. Rapid and Quantitative Detection of the Microbial Spoilage of Meat by Fourier Transform Infrared Spectroscopy and Machine Learning. 

(23) Ciurczak, E. W.; Igne, B.; Workman J., Jr.; Burns, D. A., eds. 

(24) Le, B. T. Application of Deep Learning and Near-Infrared Spectroscopy in Cereal Analysis. 

(25) Henríquez, P. A.; Ruz, G. A. Noise Reduction for Near-Infrared Spectroscopy Data Using Extreme Learning Machines. 

(26) Gonzalez Viejo, C.; Fuentes, S.; Hernandez-Brenes, C. Smart Detection of Faults in Beers Using Near-Infrared Spectroscopy, a Low-Cost Electronic Nose and Artificial Intelligence. 

(27) Zhang, H.; Zhang, L.; Wang, S.; Zhang, L. Online Water Quality Monitoring Based on UV–Vis Spectrometry and Artificial Neural Networks in a River Confluence Near Sherfield-on-Loddon. 

(28) Falbo, E.; Fusè, M.; Lazzari, F.; Mancini, G.; Barone, V. Integration of Quantum Chemistry, Statistical Mechanics, and Artificial Intelligence for Computational Spectroscopy: The UV–vis Spectrum of TEMPO Radical in Different Solvents. 

Columns: Chemometrics in Spectroscopy

A sample library of selected references discussing the application of artificial intelligence (AI) in analytical chemistry and molecular spectroscopy is presented.

Artificial Intelligence in Analytical Spectroscopy, Part II: Examples in Spectroscopy

Because cells are heterogeneous materials appearing in many sizes and containing many different elements depending on the model, inductively coupled plasma–mass spectrometry (ICP-MS) is often used to conduct element analysis of single cells. Maria Montes-Bayón of the Faculty of Chemistry at the University of Oviedo (Asturias, Spain) has been working with single cell ICP-MS (sc-ICP-MS) to study the uptake and apoptotic status of nanoplatinum (IV) treated cells, specifically selenized yeast. Montes-Bayón recently spoke to Spectroscopy about this work.

What makes ICP-MS an effective technique in your experiments for single cell analysis?

Single cell -omics is a growing field of research, since cellular heterogeneity is what triggers the beginning of many diseases. Among single cell -omics techniques (such as genomics, trancriptomics, and proteomics), metallomics is also playing an important role, since metals are key regulators of cell functions. In addition, the monitoring of the differential uptake of drugs containing metals by individual cells might explain their effectiveness. The evaluation of the intracellular content of different elements in individual cells can be performed by ICP-MS using the previously established concept of single particle ICP-MS, considering some critical points (a cell is more sensitive to the analysis process than a particle). Through the years, we have been evaluating the suitability of different ICP-MS based strategies for single cell analysis and I would say we have found the most successful combination for the sensitive and sequential determination (with triple quadrupole-based mass analyzers) of single elements in single cells.

You’ve recently worked with selenized yeast as a reference standard. Why did you choose this standard, and what are the challenges involved in using these cells as a reference standard?

Different approaches are available for individual introduction into ICP-MS. To address the suitability of every system, different authors use different strategies. Reference materials are generally used to “validate” analytical results, but it is uncommon to have reference standards specifically for “cell-based” materials. We realized that selenium enriched yeast certified reference material, or SELM-1, was a yeast material that was certified in total Se content. While talking to microbiologists at our university, they noted that yeast cells were robust enough to be intact over time, and if preserved in the correct way, they should be alive. Thus, we started a multiple-techniques study on the suitability of the reference material (for example, using confocal microcopy of labeled cells, flow cytometry, and ICP-MS analysis). We also found it necessary to address the differential Se content intra- and extracellularly.

Are there any specific sample preparation and analysis conditions that are required to be successful in this analysis?

Indeed, we realized the importance of matrix effects when using ICP-MS. It seems strange that often something so well-known for so many years is not considered carefully when conducting this type of analysis. The paper we published establishes that the reference material can be used as such for sc-ICP-MS if carefully washed a few times and then dried before analysis. If this procedure is not followed, the accuracy of the results might be compromised when using external calibration.

Are there any sample cell types that are easier or more difficult to analyze, whether you learned this through testing or presume it from past experiences?

Yes. In our experience (our first paper is from 2017), we have researched yeast cells (

,and human cells (A2780, A2780cis, OV-CAR3, A549, MCF-7, MDA-MB 231, STR-747, HepG2 corresponding to ovarian, lung, breast, and liver cancer cells). The first two models (yeast and bacteria) are much more resistant to the nebulization and transport into the plasma so the chance for leaking of the intracellular content into the sample matrix is minimized. With human models, it is much more challenging due to the delicate composition of the cell membrane that can be easily disrupted (see our TRAC paper 2020). In some cases, we fix the cells to minimize disruption, but this can be risky if constitutive elements need to be analyzed.

How does this work differ from what has been previously done by yourself or others?

This work tries to address the possible use of an already commercialized reference material to validate single cell ICP-MS analysis data. I think no one has used it for this purpose before.

Were there any limitations or challenges you encountered in your work?

Not really. We just could not believe the results we were obtaining on the unwashed samples, and we thought we were doing something wrong. We had a lot of brainstorming with the two post-docs in my office trying to understand why the results were 30% lower than expected. We even asked a more experienced post-doc to run the same sample and some PhD students to measure the same cell suspension every day to address transport efficiencies. It turns out that it was due to some matrix effects! It took us a while to understand this.

Can you please summarize the feedback that you have received from others regarding this work?

Very positive, and we have gotten some great collaborations started out of this work.

What are the next steps in this research?

We have some ideas to minimize the matrix effects when we will have access to a time-of-flight (TOF) instrument, which should happen shortly. In addition, our research is uniquely focused on single cell and single particle analysis, and we are conducting many collaborative projects. Collaborations, particularly from outside of analytical chemistry, are very important for success and understanding in this kind of research.

Pereira, J. S. F.; Álvarez-Fernández García, R.; Corte-Rodríguez, M.; Manteca, A.; Bettmer, J.; LeBlanc, K. L.; Mester, Z.; Montes-Bayón, M. Towards Single Cell ICP-MS Normalized Quantitative Experiments Using Certified Selenized Yeast. 

Corte Rodriguez, M.; Álvarez-Fernández García, R.; Blanco, E.; Bettmer, J.; Montes-Bayón, M. Quantitative Evaluation of Cisplatin Uptake in Sensitive and Resistant Individual Cells by Single-Cell ICP-MS (SC-ICP-MS)

Corte-Rodríguez, M.; Álvarez-Fernández, R.; García-Cancela, P.; Montes-Bayón, M.; Bettmer, J. Single Cell ICP-MS Using Online Sample Introduction Systems: Current Developments and Remaining Challenges. 

Columns

Single-cell ICP-MS was used to study the uptake and apoptotic status of nanoplatinum (IV) treated cells, specifically selenized yeast, and the question of using commercialized reference material to validate single cell ICP-MS analysis is addressed.

Toward Normalization of Quantitative Single Cell ICP-MS Experiments

Laser-induced breakdown spectroscopy (LIBS) is currently one of the most popular techniques for direct element analysis of solid samples. However, when directly used for liquid sample analysis, there are disadvantages, including sample splashing, plasma quenching, and poor signal stability. These problems can be overcome through liquid-solid matrix conversion; at the same time, LIBS signal enhancement can be realized, and the sensitivity of detection of liquid samples can be improved. For this research, the authors used chitosan (CS) as a raw material, and introduced poly(vinyl alcohol) (PVA) and polyethyleneimine (PEI) to finally synthesize a new type of porous membrane material with better stability and more functional group content. The membrane was used as a liquid-solid conversion matrix material combined with LIBS technology to successfully achieve rapid separation and detection of Cu, Ag, Pb, and Cr, and the corresponding detection limits can reach 0.038, 0.069, 0.012, and 0.009 mg/L, respectively. This method further improves the sensitivity of the LIBS method. Combining it with membrane materials will replace inactive membranes and open up a new way for the rapid analysis of solution samples using LIBS technology.

With the rapid development of modern industry, industrial wastewater containing a large amount of heavy metal ions is continuously discharged into the environment, and bioaccumulation occurs in the food chain. Since most of the heavy metal ions are toxic, carcinogenic, non-biodegradable, nonthermally degradable, and stable, they will eventually cause long-term toxic and side effects in the environment that will pose a great threat to human health (1,2). Therefore, it is necessary to trace detect heavy metals in liquid to efficiently control water pollution.

Currently, laser-induced breakdown spectroscopy (LIBS) is one of the most popular elemental analysis technologies, and has been called one of the “future stars” in the field of elemental analysis (3). Because of its simple sample pre-processing, fast analysis speed, non-destructive analysis, micro-local analysis, multi-element simultaneous analysis, simple equipment configuration, and outstanding advantages that can realize online and remote monitoring, it has been widely used in biomedicine and food (4–6). However, because of the problems of liquid sputtering and plasma quenching in the process of liquid detection with LIBS technology, it is difficult to assay the elements in liquid samples directly. Membrane separation is an interesting pretreatment method suitable for LIBS analysis. Various materials, such as wood (7), ion-exchange membranes (8), and modified plasticized polyvinyl chloride (PVC) membranes (9) were used for liquid-solid matrix conversion of solution samples to facilitate direct analysis using LIBS.

Therefore, we propose a chitosan-based membrane separation method combined with LIBS technology to realize the detection of liquid samples. In this research, the choice of the solid phase matrix and the existence of the matrix are very important. Chitosan (CS) is a green and environmentally friendly material. Because of its abundant reserves and rich -NH

 and -OH binding sites, it has always attracted attention and is an ideal material for removing pollutants in wastewater (10).

At present, CS film has been widely used in biomedicine, drug delivery, hydrogels, water treatment, food packaging, and other fields due to its excellent hydrophilicity, high chemical reactivity, and availability (10). However, pure CS film has limited applications in complex environments, due to its weak mechanical strength and poor chemical stability. In order to overcome these limitations, CS is usually used with other polymers; in this regard, poly(vinyl alcohol) (PVA) is one of the most effective and compatible polymers. The polyelectrolyte material prepared by cross-linking CS and PVA by Gil and associates has shown good thermal oxidation (11). However, during the cross-linking process of CS and PVA, some -NH

 will be consumed to reduce the active sites. To solve this problem, Nadia and coauthors used polyethyleneimine (PEI) to modify the CS-PVA film, and successively prepared porous CS composite membranes through different methods (12,13). The CS composite membrane they prepared increased the active sites of the membrane, and increased the removal rate of Cu, Ni, and Cd by more than 40%. At the same time, due to the porous structure of the membrane, there are abundant reactive groups on its internal pores, so the adsorption is not limited to the surface of the adsorbent.

In this work, we used silica gel as a porogen, and used the characteristics of silica gel to dissolve in alkali but not acid, while CS is soluble in acid and insoluble in alkali; we synthesized a porous CS composite membrane. Subsequently, we used the porous CS composite membrane as the solid phase matrix, combined with the liquid-solid matrix conversion method, to convert the target elements Ag, Cu, Cr, and Pb from the solution into the porous membrane, and directly analyzed this membrane with the LIBS technology. Under the best instrument conditions, the analysis concentration of the target element was established.

The reagents used in the experiment are all analytical reagents. The target element stock solution was prepared by dissolving the corresponding analytically pure nitrate (1000 mg/L, AgNO

O) in ultrapure water (18.25 MΩ·cm). In order to avoid decomposition in light, all the target element solutions were stored in a dark environment for later use. The standard solution 1~100 mg/L was prepared by diluting the corresponding stock solution. The pH value of the solution was adjusted with 1 mol/L HNO

Chitosan (deacetylation degree ≥ 95%, viscosity 100–200 mPa) was purchased from Shanghai Aladdin Biochemical Technology Co., Ltd. The chitosan solution was prepared by dissolving chitosan powder in acetic acid aqueous solution (2 wt%). After the chitosan powder fully dissolved, the solution was filtered and defoamed by leaving it still for at least 12 h at room temperature. The silica gel (mSiO

O, p300–p400) was purchased from Shanghai Aladdin Biochemical Technology Co., Ltd. Poly(vinyl alcohol) (PVA, p160) was purchased from Shanghai Chenqi Chemical Technology Co., Ltd.

Preparation and Characterization of Porous CS Composite Membrane

In this experiment, the CS composite membrane was synthesized according to the method published by various authors (12,14). Then, 5 g PVA powder was dissolved in 50 mL distilled water and stirred at 70 

C for 24 h, then the 10 wt% homogeneous solution A was obtained. Next, 5 g PEI powder was dissolved in 50 mL distilled water to obtain 10 wt% homogeneous solution B. The obtained homogeneous solution A, B, and 1 wt% CS-acetic acid homogeneous solution were mixed, and then stirred for 2 h. The final preparation membrane solution was then prepared. Silica gel was added into the preparation solution and stirred vigorously to make it evenly dispersed. Then, the solution was poured into a 3×3 cm glass petri dish and placed in an oven at 60 °C to evaporate the solvent. The dried membrane was immersed in a 10 wt% NaOH aqueous solution and stored at 80 °C for 2 h to dissolve the silica particles and form a porous membrane. Finally, the porous membrane was rinsed with distilled water to remove the remaining NaOH, and dried for later use.

A Vetex70 spectrometer (Bruker Corporation) was employed to collect the Fourier-transform infrared (FT-IR) spectra, in KBr particles at room temperature over the range of 500–4000 cm

. The morphologies of the obtained nanocomposites were investigated using an Hitachi S-4800 field emission scanning electron microscope (SEM) equipped with an energy-dispersive X-ray analysis (EDAX) detector.

The LIBS-Tracer desktop laser-induced breakdown spectrometer was used for sample analysis, and was provided by Chengdu Ailiben Technology Co., Ltd. The instrument parameters were as follows: portable desktop instrument: 477 × 370 × 462 mm, 21 kg; laser wavelength: 1064 nm, wavelength range: full spectrum 185–850 nm (segment optional); resolution: 0.1 nm; pulse energy: 100 mJ; ablation diameter: 50 μm; power supply: 220/24 V. The diameter of the ablation point is approximately 0.2 mm. For the LIBS analysis, the sample was placed on an automated XY-axis panning table to ensure a new spot on the sample surface for each laser ablation event. Each analysis was performed by accumulating the signal from 100 ablation events. To create a standard curve, 10 replicate experiments were performed for each concentration of sample.

First, 100 μL of the solution containing the target element was added to a 1 mL centrifuge tube, and the pH of the solution was adjusted to 5–6 with 1 mol/L HNO

O solution. Then, 100 μL of the target element was transferred from the aqueous phase to the surface of the porous CS composite membrane with a pipette. After drying, a circle was formed with a diameter of 7.30 mm on the membrane. The analytical zone was located within the circle. The CS film can be directly analyzed by LIBS. The blank sample was the same volume of ultrapure water instead of the target element solution, and the same drying treatment was used to detect and analyze the LIBS.

Characterization of Porous CS Composite Membrane

The functional groups of the synthesized porous CS composite membrane were characterized by FT-IR, as shown in Figure 1. The peaks around 891.06 cm

 were the stretching vibration peaks of the glycosidic bond unique to CS (15,16). The absorption peak at 1650.59 cm

 characterizes the existence of C=0 of PVA and CS. The absorption peaks at 1588.80 cm

 were the deformation vibrations of NH (amide II) and -CH

 was the unique stretching vibration peak of the CH bond in PVP, and 3461.24 cm

 was infrared absorption peaks of OH in CS and PVP and NH in CS and PEI. The experimental results showed that the porous CS composite membrane was successfully synthesized. These two polar functional groups facilitate the interaction between the porous CS composite membrane and trace metals.

The surface structure of the porous CS composite membrane was characterized by SEM, as shown in Figure 2. Figure 2a was an electron micrograph of the silica gel porogen used in the experiment. It is obvious from the figure that there were uneven particles with a particle size of 37–48 μm. Figure 2b is made without silica gel CS composite membrane; it can be seen that the surface of the membrane is dense and smooth, which is not conducive to the storage of liquid solutions and the adsorption of target elements. Figure 2c was a CS composite membrane containing silica gel particles. The surface of the membrane was rough due to the presence of silica particles. Figure 2d showed the resultant membrane after immersing a CS composite membrane containing silica gel in a 10 wt% NaOH aqueous solution and storing it at 80 °C for 2 h. Compared with the CS composite membrane made without silica gel (Figure 2b), the membrane has a rich porous structure and the pore structure is uniform, indicating that the silica gel was evenly distributed in the membrane solution. The porous structure of the membrane was conducive to storing more target solutions, and, at the same time, it can provide more active sites and improve the adsorption efficiency.

Study on the Adsorption of Porous CS Composite Membrane

In order to further understand the adsorption effect of the prepared porous CS composite membrane on the target element, LIBS and energy-dispersive spectroscopy (EDS) X-ray analysis was performed on both the blank membrane and the membrane loaded with the target element. The results were shown in Figures 3 and 4.

In Figure 3, comparing the LIBS diagram (black line) of the blank sample and the LIBS diagram (red line) of the sample to be tested, it can be clearly seen that the synthesized CS composite film had no interference to the target element that was analyzed. This can be explained from Figure 4a. The blank CS composite film contains only C, H, O, and N elements, so there was no matrix interference. The LIBS analysis of the metal-loaded porous CS composite film showed that the characteristic spectral lines of the target elements Cu, Ag, Pb and Cr were all excited. These spectral lines were Cu(I) 324.754 nm, Cu(I) 327.396 nm, Ag(I) 328.068 nm, Ag(I) 338.288 nm, Cr(I) 425.433 nm, Cr(I) 427.481 nm, and Cr(I) 428.973 nm, respectively. The same results were obtained in the EDS analysis (Figure 4b), which proved that the prepared porous CS composite film has a good adsorption effect on trace metal elements, and can effectively solve the problems of droplet sputtering in the LIBS analysis of liquid samples. In the follow-up research, we used these spectral lines as analysis lines for quantitative calculations.

Optimization of Delay Time During LIBS Analysis

LIBS technology is based on time-resolved atomic emission spectroscopy technology. In order to avoid the interference of continuous spectrum and obtain a better LIBS signal value, it is generally necessary to select a suitable delay time in the research. Figure 5 shows the changes of LIBS signal strength and signal-to-noise ratio (S/N) under different delay times. In the S/N calculation, the noise was defined as the standard deviation of the background value in the range of 2 nm adjacent to the emission line peak, and the signal was the peak height. It can be seen from Figure 5a that the signal strength of Cr does not change much with the delay time, while the signal strengths of Pb, Ag, and Cu were the strongest at 3 μs. In Figure 5b, the S/N of Cr decreases first with the increase of the delay time, and remains basically unchanged after 3 μs. The change trend of Ag and Pb S/N with delay time is not obvious, and the S/N of Cu is the maximum at 3 μs. Comprehensively, 3 μs was selected as the best delay time.

Quantitative Analysis and Detection Limit

Through the calibration curve method, within the concentration of 0–100 mg/L, the relationship between the signal and the concentration of metal ions in the water environment was analyzed by LIBS, and the quantitative curves of Ag, Cr, Cu, and Pb were established. Each point in the calibration curve corresponds to the average of ten standard samples, and the error bars reflect the standard deviation of every ten measured values. The established standard curve was shown in Figure 6. In the concentration range of 0~100 mg/L, the target element under study has a good linear relationship (

 > 0.90), and the curve can be described as 

 is the intensity of the emission spectrum, and 

 is the concentration of the target element in the solution sample.

According to the definition of the International Union of Pure and Applied Chemistry (IUPAC), the limits of detection (LODs) and limits of quantification (LOQs) of target elements are calculated (17,18):

 was the standard deviation of the intensity of the target element in 11 blank sample analyses at the selected quantitative wavelength, and 

 was the slope of the established calibration curve. The experimental results were shown in Table I.

Considering the feasibility of the proposed method in practical application, the accuracy and reproducibility of the proposed method were evaluated by analyzing actual samples, including environmental spiked samples and standard reagents. The environmental samples were obtained from an artificial lake (Yulan Lake, Northwest University). The spiked concentrations of samples were 5, 7, and 9 mg/L, respectively. All parallel samples (

 = 5) were analyzed under the optimal conditions, and the results are listed in Table II. The results show that the mean recoveries of the actual samples were in the range of 85.11%~109.45%, and the relative standard deviations (RSDs) were lower than 10.94%, which is satisfactory in terms of accuracy and repeatability for LIBS analysis of liquids.

Using the opposite solubility of silica gel and CS, a porous CS composite membrane was synthesized and applied to the analysis of heavy metal elements in liquid samples by LIBS. In the study, the porous CS composite membrane was characterized by FT-IR and SEM, and the results showed that the porous CS composite membrane was rich in N and O binding sites and porous structure, which was helpful for storing more target solutions. At the same time, the LIBS spectrum and EDS analysis results further confirmed that the synthesized porous film has a good adsorption effect on Ag, Cu, Cr, and Pb. Compared with the direct analysis of liquid samples by LIBS technology, the method proposed in this research avoids the problems of liquid sample sputtering and plasma quenching, and the detection limit of the method is further improved. At the same time, the raw materials of the synthesized porous CS composite membrane are easy to obtain, and the entire sample pretreatment process is simple. This provides a new idea for LIBS technology to analyze liquid samples. The membrane material is expected to be widely used.

This article does not contain any studies with human participants or animals performed by any of the authors.

All authors declare that there are no conflicts of interest.

(1) Bera, A.; Trivedi, J. S.; Kumar, S. B.; Chandel, A. K. S.; Haldar, S.; Jewrajka, S. K. Anti-organic Fouling and Anti-biofouling Poly(piperazineamide) Thin Film Nanocomposite Membranes for Low Pressure Removal of Heavy Metal Ions. 

(2) Zhao, Y.; Tian, G.; Duan, X.; Liang, X.; Meng, J.; Liang, J. Environmental Applications of Diatomite Minerals in Removing Heavy Metals from Water. 

(3) Winefordner, J. D.; Gornushkin, I. B.; Correll, T.; Gibb, E.; Smith, B. W.; Omenetto, N. Comparing Several Atomic Spectrometric Methods to the Super Stars: Special Emphasis on Laser Induced Breakdown Spectrometry, LIBS, a Future Super Star. 

(4) Wang, W.; Liu, Y.; Chu, Y.; Xiao, S.; Nie, J.; Zhang, J.; Qi, Q.; Guo, L. Stable Sensing Platform for Diagnosing Electrolyte Disturbance Using Laser-Induced Breakdown Spectroscopy. 

(5) Samek, O.; Beddows, D. C. S.; Telle, H. H.; Kaiser, J.; Liška, M.; Cáceres, J. O.; Gonzáles Ureña, A. Quantitative Laser-Induced Breakdown Spectroscopy Analysis of Calcified Tissue Samples. 

(6) Sezer, B.; Bjelak, A.; Velioglu, H. M.; Boyaci, I. H. Protein Based Evaluation of Meat Species by Using Laser Induced Breakdown Spectroscopy. 

(7) Chen, Z.; Godwal, Y.; Tsui, Y. Y.; Fedosejevs, R. Sensitive Detection of Metals in Water Using Laser-Induced Breakdown Spectroscopy on Wood Sample Substrates. 

(8) Dockery, C. R.; Pender, J. E.; Goode, S. R. Speciation of Chromium via Laser-Induced Breakdown Spectroscopy of Ion Exchange Polymer Membranes. 

(9) Santos, K. M.; Cortez, J.; Raimundo, I. M.; Pasquini, C.; Boa Morte, E. S.; Korn, M. G. A. An Assessment of the Applicability of the Use of a Plasticised PVC Membrane Containing Pyrochatecol Violet Complexing Reagent for the Determination of Cu

(10) Brião, G. d. V.; De Andrade, J. R.; Da Silva, M. G. C.; Vieira, M. G. A. Removal of Toxic Metals from Water Using Chitosan-Based Magnetic Adsorbents. A Review. 

(11) Gil-Castell, O.; Teruel-Juanes, R.; Arenga, F.; Salaberria, A. M.; Baschetti, M. G.; Labidi, J.; Badia, J. D.; Ribes-Greus, A. Crosslinked Chitosan/poly(vinyl alcohol)-Based Polyelectrolytes for Proton Exchange Membranes. 

(12) Sahebjamee, N.; Soltanieh, M.; Mousavi, S. M.; Heydarinasab, A. Removal of Cu

 Ions from Aqueous Solution Using a Novel Chitosan/polyvinyl Alcohol Adsorptive Membrane. 

(13) Sahebjamee, N.; Soltanieh, M.; Mousavi, S. M.; Heydarinasab, A. Preparation and Characterization of Porous Chitosan–Based Membrane with Enhanced Copper Ion Adsorption Performance. 

(14) Zeng, X.; Ruckenstein, E. Control of Pore Sizes in Macroporous Chitosan and Chitin Membranes. 

(15) Kong, Q.; Ji, Y.; Ma, H.; Song, Y.; He, B.; Li, J. A Novel Route for the Removal of Cu(II) and Ni(II) Ions via Homogeneous Adsorption by Chitosan Solution.

(16) Wang, X. H.; Wang, C. Y. Chitosan-Poly(vinyl alcohol)/attapulgite Nanocomposites for Copper(II) Ions Removal: pH Dependence and Adsorption Mechanisms. 

(17) Analytical Methods Committee. Recommendations for the Definition, Estimation and Use of the Detection Limit.

(18) Long, G. L.; Winefordner, J. D. Limit of Detection: A Closer Look at the IUPAC Definition. 

 is with the Shaanxi Key Laboratory of Earth Surface System and Environmental Carrying Capacity at the College of Urban and Environmental Sciences at Northwest University, in Xi’an, China. 

 are with the Shenzhen Agricultural Products Quality and Safety Inspection and Testing Center, in Shenzhen, China. 

 have contributed equally to this article. Direct correspondence to: 

Peer-Reviewed Research

Detecting metal elements in liquid samples cannot be done efficiently by only using LIBS, but when the technique is combined with appropriate membrane materials, rapid analysis of solution samples can be realized.

Porous Chitosan Composite Membrane Tandem Laser-Induced Breakdown Spectroscopy for Detection of Metal Elements in Liquid Samples

Shrimp tends to deteriorate during the refrigeration process. To monitor the freshness of shrimp during refrigeration, near-infrared (NIR) hyperspectral imaging was utilized to non-destructively identify the freshness of shrimp. In the process, three preprocessing methods (multivariate scatter correction [MSC], standard normal variate [SNV], and direct orthogonal signal correction [DOSC]) were employed to preprocess the full-wavelength spectral data, and three characteristic wavelength extraction algorithms (competitive adaptive reweighted sampling [CARS], and random forest [RF] simulated annealing [SA]) were used to extract the best-pre-processed data. Because extreme learning machine (ELM) and kernel extreme learning machine (KELM) are easily affected by parameters, ELM (based on teaching-learning-based optimization [TLBO]) and KELM (based on teaching-learning-based optimization [TLBO]) were proposed. In this study, four discriminant models (ELM, TLBO– ELM, KELM, and TLBO–KELM) were used for the full wavelength modeling analysis and the characteristic wavelength modeling analysis. In this work, the results of the final selected models are presented.

Shrimp is a food with high protein content (about 20%, which is higher than that of many fish, eggs, and milk [1,2]). Compared to fish, shrimp contains less essential amino acids and valine, making it a nutritionally balanced protein source. Shrimp is rich in glycine, which makes it taste sweet (3). Shrimp also has certain adjuvant therapeutic effects on skin ulcers, hand and foot convulsions, and neurasthenia; therefore, it potentially may have high medicinal value. Compared with poultry, shrimp has less fat content and almost no animal sugar as an energy source, making it highly nutritious (4). Shrimp is rich in taurine (which can lower the serum cholesterol level in the human body [5]), as well as potassium, iodine, magnesium, phosphorus, vitamin A, and vitamin B12. For these reasons, as well as many others, shrimp is very popular with consumers.

Shrimp are often processed into various seafood products (6); however, because of the action of microorganisms and their enzymes, they are vulnerable to quality damage during storage and transport (7). As a result, the short shelf life of shrimp severely affects their sale and distribution (8). Therefore, it is especially essential to identify the shelf life of shrimp in an accurate and timely matter (9,10).

The freshness of shrimp was traditionally identified using instruments for analysis and manual organoleptic evaluation (11–14). The instrument analysis method offers the inspector an approach to determine the chemical index of shrimp, such as pH value (15); unfortunately, this method has poor reliability and cannot be repeated. The artificial sensory evaluation necessitates the inspector to identify the odor and meat quality of the shrimp (16), so the method can be time-consuming and costly.

In recent years, many researchers have introduced hyperspectral imaging techniques with low cost and nondestructive testing characteristics to inspect a variety of products. Thus, it is extensively utilized in the area of freshness detection of food (17–19). Yu and associates used hyperspectral imaging technology to estimate the total volatile basic nitrogen (TVB-N) content of white shrimp, and obtained high precision (20). Wu and coauthors proposed a new method utilizing the fusion of spectral and image information to finally discriminate the freshness of salmon (21). Yao and associates achieved three freshness levels for egg sample classification by using a hyperspectral image acquisition system (22). Zhu and coauthors utilized two hyperspectral imaging systems to achieve freshness identification of spinach under different temperatures and storage time conditions (23). The near-infrared (NIR) spectral range makes the rapid imaging of samples particularly convenient, and the chemical composition difference within each pixel is shown by high-contrast images (24). Hyperspectral imaging allows simultaneous acquisition of distinct spectra and image information for every pixel on the surface of an object (25). The hyperspectral imaging (NIR band), which combines the above two advantages, is a good development prospect for the quality detection of aquatic products (26).

According to this investigation, it is feasible to utilize hyperspectral imaging to accurately identify refrigerated storage days for shrimp. The detail objectives of this study are as follows: (a) analyzing the peak distribution of the full spectral curves and the standard deviation of the average spectral curves for different refrigerated storage days of shrimp; (b) determining the best data preprocessing method to optimize the accuracy of full wavelength modeling; (c) extracting the characteristic wavelengths of the best preprocessed hyperspectral data using diverse algorithms; (d) using teaching-learning based optimization (TLBO) to optimize the parameters of the extreme learning machine (ELM) and kernel extreme learning machine (KELM) discriminant models; (e) modeling of characteristic wavelengths by using discriminant models to obtain the best-combined model for identifying refrigerated storage days for shrimp; and (f) visualizing classification results to provide more intuitive prediction results.

The experimental samples were 420 white shrimp with an average weight of about 15 g. All the shrimp were placed individually in sterile bags, and then transported in foam incubators with fresh crushed ice by special vehicles to the laboratory. The preparation of samples was performed in the laboratory, where the shrimp were individually removed from the incubator, their heads and tails were removed, and what remained was washed in ice water. Then all samples were placed in a 4 

C incubator for different storage times (1d, 2d, 3d, 4d, 5d, 6d, and 7d). A total of 60 samples were labeled for each storage time (1–7 days), and we tested 60 samples/day for 7 d.

The hyperspectral imaging system is the “Gaia Sorter” near-infrared high spectrometer. As shown in Figure 1, the working mechanism of the “Gaia Sorter” near-infrared high spectrometer works as follows: A uniform light source illuminates a sample placed on an electronically controlled mobile platform, and the emitted light of the sample is captured by a spectral camera through the lens to obtain one-dimensional (1D) images and spectral information. With the continuous movement of the sample, it is possible to obtain continuous 1D images and realtime spectral information, and eventually a three-dimensional (3D) data cube containing image information and spectral information (27).

Hyperspectral Image Acquisition and Correction

Prior to using the NIR spectrometer, the light source was adjusted for maximum brightness. The electronically controlled mobile platform was stabilized (the standard whiteboard had been removed), and all electrical cables were connected. The original hyperspectral images of white shrimps were collected in groups of 4 samples, and 15 original hyperspectral images were collected per day for a total of 105 original hyperspectral images collected over the 7 days. To minimize the impact of the dark current of the CCD camera and the uneven intensity distribution of the light sources with different wavelengths, the original hyperspectral image required reflection correction (28). The correction formula is:

 represents the corrected hyperspectral image, 

 represents the all-black reference calibration image, and 

 represents the all-white reference calibration image.

Spectral Data Extraction and Preprocessing

Because the original hyperspectral data is a hypercuube, a series of processes was required to finally obtain the spectral data (29). The process of spectral data extraction is shown in Figure 2. The specific steps are as follows: (a) opening the hyperspectral image corrected for reflectivity in ENVI 5.3 (ITT Visual Information Solutions); (b) selecting the entire area of each white shrimp as the region of interest (ROI); and (c) calculating the average reflectance for each wavelength of all pixels in the ROI, using them as spectral values of the sample; and then using Matlab2019a (The Math Works) to read the data, and to organize and add sample labels.

Because various noise sources are introduced in the original hyperspectral data acquisition process, it is necessary to preprocess the original spectral data. In this study, three spectral preprocessing methods (MSC, SNV, and DOSC) were respectively employed to preprocess the original spectral data. Table I shows the detailed steps and corresponding equations for the three preprocessing methods used.

The highly redundant character of hyperspectral data will affect the robustness of subsequent models (30). However, this can be improved by using appropriate methods to extract the characteristic wavelengths. Competitive adaptive reweighted sampling (CARS), random frog (RF), and simulated annealing (SA) were used to extract characteristic wavelengths.

CARS is a variable selection method proposed in imitation of Darwin’s evolution theory of “natural selection.” First, a large number of samples are deleted based on Monte Carlo (MC) sampling. Second, the selection is made from the remaining wavelength variables by adaptive reweighted sampling (ARS) combined with PLSR. CARS can serve the purpose of efficiently finding the optimal combination of variables by selecting variables with large absolute regression coefficients in multiple Monte Carlo operations (31).

RF is a characteristic selection algorithm that jumps different models in model space, and does not require an a priori distribution (32). Its core idea is to construct an initial set of variables and a subset of candidate variables, and then calculate the probability of selecting an acceptable set of candidate variables; the higher the probability of selection, the more important the variables are to the model.

Various chemometric approaches, including four discriminant models (ELM, TLBO–ELM, KELM, and TLBO–KELM), were used to detect shrimp freshness based on near-infrared hyperspectral imaging.

June 2023

Calibrating the Composition of a Copolymer

Artificial Intelligence in Analytical Spectroscopy, Part II: Examples in Spectroscopy

Toward Normalization of Quantitative Single Cell ICP-MS Experiments

Porous Chitosan Composite Membrane Tandem Laser-Induced Breakdown Spectroscopy for Detection of Metal Elements in Liquid Samples

Identifying Freshness of Shrimp Following Refrigeration Using Near-Infrared Hyperspectral Imaging