*Article* **Multivariate Analysis Coupled with M-SVM Classification for Lard Adulteration Detection in Meat Mixtures of Beef, Lamb, and Chicken Using FTIR Spectroscopy**

**Muhammad Aadil Siddiqui 1,\* , Mohd Haris Md Khir <sup>1</sup> , Gunawan Witjaksono <sup>2</sup> , Ali Shaan Manzoor Ghumman <sup>3</sup> , Muhammad Junaid <sup>4</sup> , Saeed Ahmed Magsi <sup>1</sup> and Abdul Saboor <sup>5</sup>**


**Abstract:** Adulteration of meat products is a delicate issue for people around the globe. The mixing of lard in meat causes a significant problem for end users who are sensitive to halal meat consumption. Due to the highly similar lipid profiles of meat species, the identification of adulteration becomes more difficult. Therefore, a comprehensive spectral detailing of meat species is required, which can boost the adulteration detection process. The experiment was conducted by distributing samples labeled as "Pure (80 samples)" and "Adulterated (90 samples)". Lard was mixed with the ratio of 10–50% *v*/*v* with beef, lamb, and chicken samples to obtain adulterated samples. Functional groups were discovered for pure pork, and two regions of difference (RoD) at wavenumbers 1700–1800 cm−<sup>1</sup> and 2800–3000 cm−<sup>1</sup> were identified using absorbance values from the FTIR spectrum for all samples. The principal component analysis (PCA) described the studied adulteration using three principal components with an explained variance of 97.31%. The multiclass support vector machine (M-SVM) was trained to identify the sample class values as pure and adulterated clusters. The acquired overall classification accuracy for a cluster of pure samples was 81.25%, whereas when the adulteration ratio was above 10%, 71.21% overall accuracy was achieved for a group of adulterated samples. Beef and lamb samples for both adulterated and pure classes had the highest classification accuracy value of 85%, whereas chicken had the lowest value of 78% for each category. This paper introduces a comprehensive spectrum analysis for pure and adulterated samples of beef, chicken, lamb, and lard. Moreover, we present a rapid M-SVM model for an accurate classification of lard adulteration in different samples despite its low-level presence.

**Keywords:** food adulteration; halal authentication; Fourier transform infrared (FTIR) spectroscopy; principal component analysis (PCA); chemometric methods; multiclass support vector machine (M-SVM)

#### **1. Introduction**

The verification of authenticity and the detection of adulterants are critical aspects of food control, particularly in high-value items. As a measure of food quality and authenticity, laboratory data as well as chemical, physical, and visual pictures of foodstuffs are employed. The authenticity of the food is a major concern in the worldwide food industry; with the abundance of packaged food with a lengthy supply chain on the market, food

**Citation:** Siddiqui, M.A.; Khir, M.H.M.; Witjaksono, G.; Ghumman, A.S.M.; Junaid, M.; Magsi, S.A.; Saboor, A. Multivariate Analysis Coupled with M-SVM Classification for Lard Adulteration Detection in Meat Mixtures of Beef, Lamb, and Chicken Using FTIR Spectroscopy. *Foods* **2021**, *10*, 2405. https://doi.org/ 10.3390/foods10102405

Academic Editor: Theodoros Varzakas

Received: 16 August 2021 Accepted: 21 September 2021 Published: 11 October 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

authenticity is still an issue, as introduced by Spink and Mayor [1]. Nowadays, manual inspection, which is highly impacted by subjective variables, is nevertheless used frequently in quality evaluation. As a result, detecting pork in a variety of food items has become a major research topic in many countries, particularly in those where religious laws restrict the eating of pig products. Food adulterations may only financially impact a part of the population, but others may be more seriously affected [2–4] due to food poisoning, their religious views [5,6], etc. Some of the food tampering has been poisonous, for instance, such as the addition of sawdust to make white bread [7,8], the melamine adulteration of formula milk [7,9,10], the mixing of oil for engines with oil for human consumption in Spain [11]; some cases also involved the misrepresentation of food ingredients such as the UK horse meat issue in 2013 [12–15]. There are several ways of determining the provenance of animal species in meat products that are based on nucleic acid resources, commonly known as molecular techniques, which include DNA finger printing, PCR assays and PCR simple sequence repeat (PCR-SSR) [16,17], chromatographic techniques, isotopic techniques, vibrational and fluorescence spectroscopy, elemental techniques, nuclear magnetic resonance spectroscopy, sensory analysis, non-chromatographic mass spectrometry, immunological techniques, along with chemometrics and bioinformatics [18]. However, each methodology has its own set of drawbacks such as being costly, time-consuming, and inefficient, as well as requiring a wide range of equipment and making it difficult to understand the acquired data; moreover, most of these methods often require extensive sample preparation or are very susceptible to impurities. Unless all the protocols are strictly followed, they may lead to unpredictable outcomes. As a result, establishing a quick and reliable identification procedure to recognize meat species is critical. To address these restrictions, individuals have increasingly turned to spectroscopic methods in recent years. Fourier transform infrared spectroscopy (FTIR) has been widely used in the identification of agricultural commodities such as wine, olive oil, tea, and meat due to its quick and easy operation [19–22]. Research into food-authentication vibrational spectroscopy technologies today has been growing [22–26], partly because the sample preparation using the FTIR technique is relatively simple, results are relatively rapid, and this process is non-destructive in nature. The FTIR spectroscopic methods are thus fast becoming popular [27–33]. Some researchers have started to veer to Near Infrared (NIR) spectroscopy, mainly because its feasibility would open the possibilities of making the food authentication instrumentation set-up portable [28–30]. FTIR is quick and relatively inexpensive, with an easier sample preparation and a non-destructive process [18,19,24,34]. FTIR spectroscopy can distinguish meat and lard in meatball broth quickly and with high accuracy [19,21]; it has also been used with chocolate [24,34] and vegetable oils [22]. Table 1 presents the summary of methods and adulterants used in the literature, along with the multivariate techniques used for detecting the adulteration in different meat species. Therefore, the aim of this study was to utilize in-depth FTIR spectral analysis to improve the accuracy of lard adulteration detection by employing the classification of pure and adulterated samples combined with an M-SVM analysis for lard adulterated in mixtures of beef, lamb, and chicken.

**Table 1.** Summary of food analyses using multivariate techniques with infrared spectroscopy for the detection of meat species adulteration [35–40].


#### **2. Materials and Methods**

#### *2.1. Meat Sample Collection*

All meat samples were obtained from the local market at Seri Iskander in Malaysia. After that, the meat was washed with purified water and cut into small parts (1 cm × 1 cm) and held at −10 ◦C. Total samples were then divided into two different classes, as pure and adulterated. There were 80 pure and 90 adulterated samples produced for the spectral analysis. The sample preparation was designed to be straightforward, with no extra chemical substances used. Beef, lamb, and chicken loin cuts were used, and all pork was lean meat taken from chops.

#### *2.2. Extraction Procedure and Sample Distribution*

Lard and other animal body fats from meat such as chicken fat, beef fat, and mutton fat were extracted according to the method stated by [34], with little variation. All samples were gradually heated from 50 ◦C to 150 ◦C for 45 min until the fat was extracted from all the samples on the petri dish. The discharged fat was then filtered as the concentration contained solid minute particles. Moreover, samples were centrifuged at 3000 rpm for 20 min and filtered through Whatman filter paper. Pure fats produced by the extraction process were then used to make adulterated samples. All the chemicals used in this experiment were of analytical consistency. Pure and adulterated fats were then analyzed using FTIR spectroscopy. The instrument used was Frontier FT-IR by PerkinElmer. The optical system with KBr beam splitter was used to enable quality data collection over a range of 8300–350 cm−<sup>1</sup> at a best resolution of 0.4 cm. The resulting spectrum contained 2500 continuous values for one sample, with intervals of 0.8 cm−<sup>1</sup> . To guarantee that there was no major fluctuation between each spectra scanned, each spectrum was recorded at the same temperature. This procedure was required to remove any uncontrolled ambient influences on the instrument and the sample.

#### *2.3. Spectral Data Pre-Processing*

Smoothing and normal variate transformation (SNV) were used as spectrum preprocessing approaches in this investigation. The reflectance spectra were smoothed by Savitzky-Golay smoothing using a second-order polynomial and a 5-point window to eliminate the random disturbances caused by the system's internal components. SNV was used to adjust for scatter effects and reduce slope variation. The Savitzky-Golay smoothing filter was used to increase the precision of the data without distorting the signal tendency.

#### *2.4. Preparing Mixture Samples*

Lard was mixed with body fats of lamb, beef, and chicken to obtain a series of standard or trained sets of 80 pure and 90 adulterated samples containing 10–50% *v*/*v* of lard in lamb, beef, and chicken samples, as shown in Table 2. The following method is according to Rohman et al. [23]. We prepared six pieces for each combination of lard mixed with a defined percentage of lamb, chicken, and beef, with pork in the proportion of 10, 20, 30, 40, and 50%, whereas B-50%, L-50%, and C-50% represent a 50-50 ratio of pork with beef, lamb, and chicken, respectively; meanwhile, B-90%, L-90%, and C-90% indicate 10% lard with 90% of the respective species. The detailed distribution of samples is presented in Table 3.

**Table 2.** Distribution of adulterated and pure samples along with the number of pieces produced and spectra obtained for individual species.



**Table 3.** Composition of adulterated samples with the ratio of lard mixed with samples of beef, lamb, and chicken, represented by their initials (Lamb: L-90% to L50%, Beef: B-90% to B-50%, Chicken: C-90% to C-50%).

#### **3. Results and Discussion**

After a careful process of sample-making and data pre-processing, the obtained spectrum for both pure and adulterated samples was analyzed separately. The developed workflow for further investigating the lard adulteration was carried out using a three-stage process. In the first stage, identification of functional groups in lard samples without any contamination was made. Secondly, pure spectral samples of beef, lamb, chicken, and lard were analyzed by overlapping the spectrums and identifying the region of difference (RoD) for highly significant regions. Moreover, the profiling of adulterated samples with the percentage difference for beef, lamb, and chicken was also carried out. After spectral analysis, the third and final stage combined the multivariate analysis with M-SVM classification for both pure and adulterated samples separately. Samples were divided into two classes, 'Haram (lard)' and 'Halal (chicken, lamb, and beef)', for M-SVM classification.

#### *3.1. FTIR Spectra Analysis of Pure Samples*

Amid the four different meat fats, the pure lard used in this study was evaluated and analyzed separately using FTIR spectroscopy. The peak is shown in Figure 1 approximately at wavenumber 2921 cm−<sup>1</sup> , which was due to the tensile vibration of C-H (Sp<sup>3</sup> ) in = C-H cis. The functional group-CH<sup>2</sup> provided peaks at wavenumber 2853 cm−<sup>1</sup> consecutively as result of asymmetrical and symmetrical vibration. The peak showed the triglyceride ester carbonyl (C=O) group at wavenumber 1750 cm−<sup>1</sup> .

In the fingerprint region, vibrations of the stretching mode from the C-O group in esters were detected at wavenumber 1155 cm−<sup>1</sup> , while at wavenumber 1467 cm−<sup>1</sup> the bending vibrations of the CH<sup>2</sup> and CH<sup>3</sup> aliphatic groups were detected, as shown in Figure 1. Table 4 shows the details of wavenumber and the associated vibration of functional groups for the pure lard sample.

Figure 2 below shows the FTIR spectra of pure samples overlapped for the identification of wavenumbers, with associated peaks identified as the region of difference (RoD) along with the fingerprint region. This spectrum can be divided into three regions to make the analysis convenient: the first region range is at wavenumber 3000–2500 cm−<sup>1</sup> , the second region range is 2000–2500 cm−<sup>1</sup> , the third region range is 1500–2000 cm−<sup>1</sup> , and to conclude, the fingerprint region range is at wavenumber 1500–500 cm−<sup>1</sup> . Two separate regions are highlighted with dotted lines (a and b), with the overlapping of pure samples

for all species, as indicated in Figure 2, where the change in absorbance values is highly prominent; wavenumbers associated with these two regions are in the spectrum ranges of 1700–1800 cm−<sup>1</sup> for RoD(a) and 2800–3000 cm−<sup>1</sup> for RoD(b) respectively as shown in Figure 3. The FTIR spectra of all the lipids obtained from different species were combined and overlapped. triacylglycerol <sup>2921</sup> Asymmetrical or symmetrical stretching methylene (- CH2) band vibration

*Foods* **2021**, *10*, x FOR PEER REVIEW 5 of 14

eride ester carbonyl (C=O) group at wavenumber 1750 cm−1.

**Table 4.** Functional group and associated mode of vibration for pure lard.

groups for the pure lard sample.

Amid the four different meat fats, the pure lard used in this study was evaluated and

In the fingerprint region, vibrations of the stretching mode from the C-O group in

1155 Vibrations of stretching mode from the C-O group in es-

1467 Bending vibrations of the CH2 and CH3 aliphatic groups

<sup>1750</sup>Carbonyl (C=O) functional group of the ester linkage of

ters

esters were detected at wavenumber 1155 cm−1, while at wavenumber 1467 cm−1 the bending vibrations of the CH2 and CH3 aliphatic groups were detected, as shown in Figure 1. Table 4 shows the details of wavenumber and the associated vibration of functional

**Frequency (cm−1) Functional Group Vibration** 

analyzed separately using FTIR spectroscopy. The peak is shown in Figure 1 approximately at wavenumber 2921 cm−1, which was due to the tensile vibration of C-H (Sp3) in = C-H cis. The functional group-CH2 provided peaks at wavenumber 2853 cm−1 consecutively as result of asymmetrical and symmetrical vibration. The peak showed the triglyc-

**Figure 1.** Spectrum analysis of pure pork identifying the frequencies for functional group vibrations.

**Figure 1.** Spectrum analysis of pure pork identifying the frequencies for functional group vibrations.

ond region range is 2000–2500 cm−1, the third region range is 1500–2000 cm−1, and to con-


**Table 4.** Functional group and associated mode of vibration for pure lard.

**Figure 2.** Overlapped spectrum from FTIR covering 3500–650 cm<sup>−</sup>1, representing the fingerprint and functional group regions for pure samples of beef, lamb, lard, and chicken, with identification **Figure 2.** Overlapped spectrum from FTIR covering 3500–650 cm−<sup>1</sup> , representing the fingerprint and functional group regions for pure samples of beef, lamb, lard, and chicken, with identification of potential regions of difference (RoD).

**Figure 3.** RoD(a) Region of difference peaks zoomed in at wavenumber 1700–1800 cm<sup>−</sup>1 showing the absorbance value for pure samples denoted by a; lard has the lowest value among all samples. RoD(b) Zoomed-in peaks at wavenumber 2800–3000 cm−1 where peaks denoted as a, b, and c rep-

As the value for the adulteration of lard increases for both beef and chicken, the absorbance values merge with the lard, showing high contrast compared to lamb samples, which indicates negligible change when lard is mixed. This is clearly visible in the spectral analysis shown in Figure 4 for all the adulterated samples. The absorbance values in the region of RoD(b) are carefully analyzed, where the adulteration of lard can potentially be detected. This is shown in Table 5. On the other hand, beef samples are highly prone, and lard is detectable because of the significant change in absorbance value at the region of 2800–3000 cm−1 in the spectrum, specifically at RoD(b) a and b, which represent regions at 2840–2860 and 2900–2940 cm−1, respectively. Table 5 lists all the absorbance values at the peaks of RoD(b) in Figure 2; the percentage difference is calculated with respect to lard

resent potential regions with difference in absorbance values for all samples.

for peak absorbance in regions with high significance.

chicken in the region of RoD(b) at the highly significant region of 2800–3000 cm<sup>−</sup>1.

**RoD(b)-a** 

**Species Type Sample Absorbance Value at** 

**Table 5.** Absorbance values and percentage difference with respect to lard for adulterated samples of beef, lamb, and

**Pure Lard** Pork-100% 1.5963 1.75306 **RoD(b)-a RoD(b)-b** 

**Adulterated Beef** B-50% 1.6580 1.9154 3.79% 8.85%

**Absorbance Value at** 

B-60% 1.8357 2.1793 13.95% 21.67%

**RoD(b)-b Percentage Difference w.r.t Pork** 

of potential regions of difference (RoD).

1.2

1.6

Absorbance

2.0

2.4

 Cow fat Chicken fat Lamb fat Lard

**Fingerprint Region**

of potential regions of difference (RoD).

**Figure 2.** Overlapped spectrum from FTIR covering 3500–650 cm<sup>−</sup>1, representing the fingerprint

690 1380 2070 2760

Wavenumber (1/cm)

**Region of Difference (RoD)**

a b

**Figure 3.** RoD(a) Region of difference peaks zoomed in at wavenumber 1700–1800 cm<sup>−</sup>1 showing the absorbance value for pure samples denoted by a; lard has the lowest value among all samples. RoD(b) Zoomed-in peaks at wavenumber 2800–3000 cm−1 where peaks denoted as a, b, and c represent potential regions with difference in absorbance values for all samples. **Figure 3.** RoD(a) Region of difference peaks zoomed in at wavenumber 1700–1800 cm−<sup>1</sup> showing the absorbance value for pure samples denoted by a; lard has the lowest value among all samples. RoD(b) Zoomed-in peaks at wavenumber 2800–3000 cm−<sup>1</sup> where peaks denoted as a, b, and c represent potential regions with difference in absorbance values for all samples.

As the value for the adulteration of lard increases for both beef and chicken, the absorbance values merge with the lard, showing high contrast compared to lamb samples, which indicates negligible change when lard is mixed. This is clearly visible in the spectral analysis shown in Figure 4 for all the adulterated samples. The absorbance values in the region of RoD(b) are carefully analyzed, where the adulteration of lard can potentially be detected. This is shown in Table 5. On the other hand, beef samples are highly prone, and lard is detectable because of the significant change in absorbance value at the region of 2800–3000 cm−1 in the spectrum, specifically at RoD(b) a and b, which represent regions at As the value for the adulteration of lard increases for both beef and chicken, the absorbance values merge with the lard, showing high contrast compared to lamb samples, which indicates negligible change when lard is mixed. This is clearly visible in the spectral analysis shown in Figure 4 for all the adulterated samples. The absorbance values in the region of RoD(b) are carefully analyzed, where the adulteration of lard can potentially be detected. This is shown in Table 5. On the other hand, beef samples are highly prone, and lard is detectable because of the significant change in absorbance value at the region of 2800–3000 cm−<sup>1</sup> in the spectrum, specifically at RoD(b) a and b, which represent regions at 2840–2860 and 2900–2940 cm−<sup>1</sup> , respectively. Table 5 lists all the absorbance values at the peaks of RoD(b) in Figure 2; the percentage difference is calculated with respect to lard for peak absorbance in regions with high significance.

2840–2860 and 2900–2940 cm−1, respectively. Table 5 lists all the absorbance values at the peaks of RoD(b) in Figure 2; the percentage difference is calculated with respect to lard for peak absorbance in regions with high significance. **Table 5.** Absorbance values and percentage difference with respect to lard for adulterated samples of beef, lamb, and chicken in the region of RoD(b) at the highly significant region of 2800–3000 cm<sup>−</sup>1. **Species Type Sample Absorbance Value at RoD(b)-a Absorbance Value at RoD(b)-b Percentage Difference w.r.t Pork Pure Lard** Pork-100% 1.5963 1.75306 **RoD(b)-a RoD(b)-b Adulterated Beef** B-50% 1.6580 1.9154 3.79% 8.85% The highest proximity of absorbance values to pure lard can be seen in the samples of B-50%, C-90%, C-80%, and C-50%, for both regions RoD(b)-a and RoD(b)-b. At the same time, adulterated beef shows a pattern of variation according to the adulteration percentage of lard. Beef samples with 10% adulteration (B-90%) have an approximate percentage difference of 7–14%, while beef with 50% adulteration (B-50%) shows approximately 3–8% change for both regions. All samples containing adulterated chicken from C-50% to C-90% show the lowest percentage difference as compared to lamb and beef. This reveals the highest similarity to be between chicken and lard, which could present some difficulty in detecting the adulteration of lard in chicken irrespective of the percentage mixing. Moreover, adulterated lamb samples depict minor variation in absorbance values throughout the mixing samples (L-50% to L-90%) and have the highest percentage difference as compared to pure lard.

#### B-60% 1.8357 2.1793 13.95% 21.67% *3.2. Results of Principal Component Analysis*

Pure lard, along with other samples of beef, chicken, and lamb, was classified using the chemometric of PCA. PCA is used to reduce the dimension of the spectral signal. The wavenumber regions for PCA were also optimized. To confirm the separation based on adulterant type, the raw data (eigenvectors of the covariance matrix) was subjected to principal component analysis (PCA). Further explanation on PCA is at Appendix A.1

**Adulterated Chicken** 

**Adulterated Lamb** 

B-70% 1.8310 2.1784 13.69% 21.63% B-80% 1.7611 2.0906 9.81% 17.56% B-90% 1.7262 2.0227 7.81% 14.28%

C-50% 1.5256 1.8577 4.52% 5.79% C-60% 1.5289 1.8737 4.31% 6.65% C-70% 1.5312 1.8868 4.16% 7.34% 1.5358 1.8995 3.86% 8.01% C-90% 1.5358 1.8995 3.86% 8.01%

L-50% 1.8739 2.2576 15.99% 25.15% L-60% 1.8739 2.2576 15.99% 25.15% L-70% 1.8739 2.2576 15.99% 25.15%

L-90% 1.8710 2.2396 15.84% 24.37%

**Figure 4.** Peaks zoomed-in for adulterated samples overlapped with pure pork, for beef (B-50% to B-90%), lamb (L-50% to L-90%), and chicken (C-50% to C-90%), with associated adulteration ratio **Figure 4.** Peaks zoomed-in for adulterated samples overlapped with pure pork, for beef (B-50% to B-90%), lamb (L-50% to L-90%), and chicken (C-50% to C-90%), with associated adulteration ratio and associated peaks for RoD(a) and RoD(b).


and associated peaks for RoD(a) and RoD(b). **Table 5.** Absorbance values and percentage difference with respect to lard for adulterated samples of beef, lamb, and chicken in the region of RoD(b) at the highly significant region of 2800–3000 cm−<sup>1</sup> .

> It is possible to observe a distinct split depending on the level of adulteration by showing the scores of the first two main components (Figure 5), which represent 99.36 percent of data variance. Only a little amount of overlap exists between the chicken samples that have been tainted with pork. The selection of wavenumbers was based on their ability to

provide a useful classification between samples, as seen in Figure 5. The PCA plot showed clusters of samples based on their similarity with the first main component (PC1) and the second main component (PC2), which provided a good separation between the lamb, beef, and pork groups but was unable to separate pork and chicken. The percentage (%) variability of PC1 and PC2 was 97.31% and 2.05%, respectively. PC1 comprised the most variation of the data, as shown in Table 6. (PC1) and the second main component (PC2), which provided a good separation between the lamb, beef, and pork groups but was unable to separate pork and chicken. The percentage (%) variability of PC1 and PC2 was 97.31% and 2.05%, respectively. PC1 comprised the most variation of the data, as shown in Table 6.

Pure lard, along with other samples of beef, chicken, and lamb, was classified using the chemometric of PCA. PCA is used to reduce the dimension of the spectral signal. The wavenumber regions for PCA were also optimized. To confirm the separation based on adulterant type, the raw data (eigenvectors of the covariance matrix) was subjected to principal component analysis (PCA). Further explanation on PCA is at Appendix A.1

It is possible to observe a distinct split depending on the level of adulteration by

showing the scores of the first two main components (Figure 5), which represent 99.36 percent of data variance. Only a little amount of overlap exists between the chicken samples that have been tainted with pork. The selection of wavenumbers was based on their ability to provide a useful classification between samples, as seen in Figure 5. The PCA plot showed clusters of samples based on their similarity with the first main component

50% Pork).

*Foods* **2021**, *10*, x FOR PEER REVIEW 8 of 14

*3.2. Results of Principal Component Analysis* 

**Figure 5.** Principal component analysis plot showing the similarity between pork, chicken, lamb, and beef samples with adulterated mixtures. C1-C5(10–50% Pork), B1-B5(10–50% Pork), L1-L5(10– **Figure 5.** Principal component analysis plot showing the similarity between pork, chicken, lamb, and beef samples with adulterated mixtures. C1–C5 (10–50% Pork), B1–B5 (10–50% Pork), L1–L5 (10–50% Pork).

**Table 6.** Percentage of variance for each PCA component contributing to the variation of the classification.


PC2 2.05% PC3 0.64% The FTIR spectra of the pure pork sample were compared with those of adulterated beef, chicken, and lamb. Three dimensional plots are shown in Figure 6. The PCA analysis shows the PCA projection divided into three dimensions for better analysis.

The FTIR spectra of the pure pork sample were compared with those of adulterated beef, chicken, and lamb. Three dimensional plots are shown in Figure 6. The PCA analysis shows the PCA projection divided into three dimensions for better analysis. Figure 6a shows the distribution of samples across the first principal component using 1D spectra of the pure samples for beef, lamb, chicken, and pork, where chicken and pork samples overlap and correlate highly coupled values of absorbance with similar wavenumbers. At the same time, Figure 6b depicts the samples at PC1 and PC2 using 2D representation for all the adulterated species. Figure 6c combines all the three principal components using 3D for all the adulterated samples. The regions in these figures are separated based on the adulteration quantity, starting with slightly mixed, i.e., 10%, to highly adulterated, i.e., 50%. In the first projection, the plotted points representing the samples of chicken, beef, and lamb are scattered, and they are far from the pork group. The closer the dots of chicken, beef, and lamb are to the pork samples, the more significant the quantity of lard is in pure samples.

#### *3.3. Multiclass Support Vector Machine Classification*

The data obtained from the previous processes were divided into testing data (30%) and training data (70%), and subsequently evaluated with the classification model. The data acquired from the FTIR spectroscope was analyzed using the scikit-learn machine learning library in Python. The radial basis function (RBF) was used as the kernel function of SVM using the grid search method. To add an extra validation step to our model, we used the confusion matrix for both multiclass datasets, as shown in Tables 7 and 8. The confusion

matrix projects the true data against predicted data. In our study, we divided the problem into two different sections: one identified pure samples correctly, and the other predicted the adulterated samples. The learning rate was 0.0001, and the regularization parameter λ was set to 1/epochs. Table 7 illustrates the user, producer, and overall accuracy of the pure samples data set. Details of the SVM is explained at Appendix A.2. Pure samples of beef and lamb using optimal parameters produced the highest accuracy (85%) among all the samples. Furthermore, pure samples of chicken had the lowest accuracy of 75%, whereas pure pork was significantly better than chicken, with 80% accuracy. Moreover, Figure 7 shows a confusion matrix using a 10-fold cross-validation for the pure samples where the a, b, and c rows represent the true label; meanwhile, according to the model prediction, the a, b, and c columns represent the number of predicted sets for each respective class. *Foods* **2021**, *10*, x FOR PEER REVIEW 9 of 14

**Figure 6.** (**a**) Pure samples of all the species using a one-dimensional projection for first principal component. (**b**,**c**) Principal component analysis of the two and three-dimensional projections of adulterated samples of pork (**\***), chicken (×), lamb (**+**), and beef (O), showing the clustering. **Figure 6.** (**a**) Pure samples of all the species using a one-dimensional projection for first principal component. (**b**,**c**) Principal component analysis of the two and three-dimensional projections of adulterated samples of pork (**\***), chicken (×), lamb (+), and beef (O), showing the clustering.

adulterated, i.e., 50%. In the first projection, the plotted points representing the samples


Figure 6a shows the distribution of samples across the first principal component us-**Table 7.** Sensitivity, precision, and classification accuracy for pure samples of beef, lamb, chicken, and pork.

of chicken, beef, and lamb are scattered, and they are far from the pork group. The closer the dots of chicken, beef, and lamb are to the pork samples, the more significant the quan-**Table 8.** Sensitivity, precision, and classification accuracy for adulterated samples of beef, chicken, and lamb.


respective class.

and training data (70%), and subsequently evaluated with the classification model. The data acquired from the FTIR spectroscope was analyzed using the scikit-learn machine learning library in Python. The radial basis function (RBF) was used as the kernel function

used the confusion matrix for both multiclass datasets, as shown in Tables 7 and 8. The confusion matrix projects the true data against predicted data. In our study, we divided the problem into two different sections: one identified pure samples correctly, and the other predicted the adulterated samples. The learning rate was 0.0001, and the regularization parameter λ was set to 1/epochs. Table 7 illustrates the user, producer, and overall accuracy of the pure samples data set. Details of the SVM is explained at Appendix A.2. Pure samples of beef and lamb using optimal parameters produced the highest accuracy (85%) among all the samples. Furthermore, pure samples of chicken had the lowest accuracy of 75%, whereas pure pork was significantly better than chicken, with 80% accuracy. Moreover, Figure 7 shows a confusion matrix using a 10-fold cross-validation for the pure samples where the a, b, and c rows represent the true label; meanwhile, according to the model prediction, the a, b, and c columns represent the number of predicted sets for each

**Figure 7.** Heatmap confusion matrix of multiclass classification for pure samples of beef, chicken, lamb, and pork showing the predicted and true labels. **Figure 7.** Heatmap confusion matrix of multiclass classification for pure samples of beef, chicken, lamb, and pork showing the predicted and true labels.

**Table 7.** Sensitivity, precision, and classification accuracy for pure samples of beef, lamb, chicken, and pork. The predicted labels for pure samples shown in Figure 7 misclassified three samples of pure chicken as pure pork, while two samples of pure pork were falsely labeled as chicken. Moreover, beef and lamb both had three label misclassifications, one for each species of meat.

**Classified as User Accuracy (Sensitivity) Producer Accuracy (Precision) Overall Accuracy**  Beef 85% 85.00% 81.25% Lamb 85% 85.00% Chicken 78% 75.00% Pork 76% 80.00% Table 8 shows the confusion matrix for the multiclass SVM of adulterated data samples. The adulterated data set contained all the samples that were adulterated with different proportions of lard. The AdulteratedBeef sample included samples with a *v*/*v* ratio from B-50% to B-90%. The producer accuracy was highest for AdulteratedLamb at 76.6%, whereas AdulteratedBeef had the second-highest value of 73.3%. The spectrum of lamb had no change in absorbance value when it was adulterated, irrespective of the adulteration ratio, which was also validated by the SVM classifier by getting the maximum number of correctly classified labels, as shown in Figure 8. *Foods* **2021**, *10*, x FOR PEER REVIEW 11 of 14

species of meat. Table 8 shows the confusion matrix for the multiclass SVM of adulterated data sam-**Figure 8.** Heatmap confusion matrix of the multiclass SVM classifier for adulterated samples of beef, chicken, and lamb. **Figure 8.** Heatmap confusion matrix of the multiclass SVM classifier for adulterated samples of beef, chicken, and lamb.

ples. The adulterated data set contained all the samples that were adulterated with different proportions of lard. The AdulteratedBeef sample included samples with a *v*/*v* ratio from B-50% to B-90%. The producer accuracy was highest for AdulteratedLamb at 76.6%, AdulteratedChicken samples, with 20 correctly classified samples, produced the lowest precision accuracy of 66% due to its high variation in absorbance values, as shown in AdulteratedChicken samples, with 20 correctly classified samples, produced the lowest precision accuracy of 66% due to its high variation in absorbance values, as shown in Figure 8.

#### whereas AdulteratedBeef had the second-highest value of 73.3%. The spectrum of lamb Figure 8. **4. Conclusions**

had no change in absorbance value when it was adulterated, irrespective of the adulteration ratio, which was also validated by the SVM classifier by getting the maximum number of correctly classified labels, as shown in Figure 8. **4. Conclusions**  FTIR spectroscopy, coupled with the multivariate and M-SVM methods, seems to be an efficient and rapid technique for the discrimination of lard from other meat samples. FTIR spectroscopy, coupled with the multivariate and M-SVM methods, seems to be an efficient and rapid technique for the discrimination of lard from other meat samples. In this paper, we demonstrated the identification and discrimination of lard from beef, chicken, and lamb fats in meat mixtures. FTIR spectral analysis in combination with Principal

may be applied for detecting an adulteration quantity of less than 10%.

tute for smart mobility, University Technology Petronas, Perak, Malaysia.

**Institutional Review Board Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflicts of interest.

**Informed Consent Statement:** Not applicable.

*A.1. Principal Component Analysis* 

sponding author.

**Appendix A** 

In this paper, we demonstrated the identification and discrimination of lard from beef,

cipal Component Analysis (PCA) and M-SVM have shown that pure lard fat has unique peaks that can distinguish the pork from beef, chicken, and lamb meat at wavenumbers 1155 cm−1, 1467 cm−1, 1750 cm−1, and 2921 cm−1. The absorbance values indicate a direct correlation between lard and other species. The PCA results show that adulteration in chicken meat is positively correlated with pork meat, while lamb is negatively correlated with respect to lard. The SVM model produced an overall prediction accuracy of 81.25% for pure samples, and for adulterated samples, it showed a 72.2% prediction accuracy. The overall accuracy was computed using the sensitivity and precision values. The model accurately classified the pure samples better than the adulterated samples due to a smaller number of samples and the minimalistic difference in absorbance values of the spectrum. Thus, this study has the potential to establish as a rapid method for halal authentication and could revolutionize the in-line quality control in the meat industry. For future work, the FTIR profiles for pure and adulterated samples can be increased, and deep learning

**Author Contributions:** M.A.S. and G.W. conceived the project; M.H.M.K. assisted and supported the experiment; A.S.M.G. supported the sample preparation; and M.J., S.A.M. and A.S. reviewed the manuscript. All authors have read and agreed to the published version of the manuscript.

**Funding:** This study is funded by Centre of Graduate Studies, UTP in collaboration with ITI insti-

**Data Availability Statement:** Data presented in this study is available at request from the corre-

Component Analysis (PCA) and M-SVM have shown that pure lard fat has unique peaks that can distinguish the pork from beef, chicken, and lamb meat at wavenumbers 1155 cm−<sup>1</sup> , 1467 cm−<sup>1</sup> , 1750 cm−<sup>1</sup> , and 2921 cm−<sup>1</sup> . The absorbance values indicate a direct correlation between lard and other species. The PCA results show that adulteration in chicken meat is positively correlated with pork meat, while lamb is negatively correlated with respect to lard. The SVM model produced an overall prediction accuracy of 81.25% for pure samples, and for adulterated samples, it showed a 72.2% prediction accuracy. The overall accuracy was computed using the sensitivity and precision values. The model accurately classified the pure samples better than the adulterated samples due to a smaller number of samples and the minimalistic difference in absorbance values of the spectrum. Thus, this study has the potential to establish as a rapid method for halal authentication and could revolutionize the in-line quality control in the meat industry. For future work, the FTIR profiles for pure and adulterated samples can be increased, and deep learning may be applied for detecting an adulteration quantity of less than 10%.

**Author Contributions:** M.A.S. and G.W. conceived the project; M.H.M.K. assisted and supported the experiment; A.S.M.G. supported the sample preparation; and M.J., S.A.M. and A.S. reviewed the manuscript. All authors have read and agreed to the published version of the manuscript.

**Funding:** This study is funded by Centre of Graduate Studies, UTP in collaboration with ITI institute for smart mobility, University Technology Petronas, Perak, Malaysia.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data presented in this study is available at request from the corresponding author.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Appendix A**

#### *Appendix A.1. Principal Component Analysis*

Principal Component Analysis (PCA) is a statistical technique that is particularly useful in reducing observations that have many dimensions. This technique consists of transforming dimensions of a dataset into a new but smaller set of uncorrelated dimensions called principal components (PCs). An array of (*qij*) values can be normalized using the equation below:

$$X\_{i\bar{j}} = q\_{i\bar{j}} - q\_{\bar{j}} \tag{A1}$$

The data given to us is the array element data corresponding to the variable *Xij*, and the mean value of the variable *q<sup>j</sup>* . Then, using the new dataset array, a correlation matrix is constructed so that information about how the variables in the dataset are correlated can be obtained. To create our new correlation matrix *X* with the new correlation coefficients *Xij*, the following formula is used:

$$\mathcal{R} = X^T \bullet X \tag{A2}$$

Only the principal components that explain the greatest amount of data in the original are determined using the equation below:

$$S = V \bullet Q \tag{A3}$$

where *S* is the matrix data, known as Score; *V* is the eigenvectors; and *Q* is the original data array. The matrix *S* (Score) will now represent the data in a way that each column represents the projection of the initial data *Q*.

#### *Appendix A.2. Support Vector Machine Classification*

Most machine learning techniques have been created and statistically verified for linearly separable data. For the reduction of dimensionality, linear classifiers such as Support Vector Machines (SVMs) or the (conventional) Principal Component Analysis (PCA) are common examples. However, to efficiently accomplish tasks involving pattern analysis and discovery, most real-world data require non-linear approaches. By incorporating the kernel trick, the SVM approach has improved over time. To detect a pattern in non-linear separable data, the kernel method effectively translates the input data to higher dimensions. When the training data has many variables in comparison to the number of observations, SVMs are an excellent classification approach. In SVM, every sample x that consists of n variables is treated as an n-dimensional vector. Prediction performance can be assessed using the following three indicators: sensitivity (User Accuracy), precision (Producer Accuracy), and overall accuracy. Precision is the proportion of appropriately positive labels produced by our software to all positive labels produced. The ratio of the exactly positive labels identified by our algorithm to all positive labels is referred to as sensitivity. Accuracy is the proportion of correctly categorized topics to the total number of issues. Equations (A4)–(A6) present the formula for Precision, Accuracy, and Sensitivity.

$$Sensitivity = \frac{True\text{ Positive}}{Predicted\text{Results}}\tag{A4}$$

$$Precision = \frac{\text{True Positive}}{\text{Actual Results}} \tag{A5}$$

$$\text{Overall Accuracy} = \frac{\text{True Positive} + \text{True Negative}}{\text{Total}} \tag{A6}$$

#### **References**


**Sumaiya Shomaji \*,†, Naren Vikram Raj Masna † , David Ariando, Shubhra Deb Paul , Kelsey Horace-Herron , Domenic Forte, Soumyajit Mandal and Swarup Bhunia**

> Department of Electrical and Computer Engineering, University of Florida, 216 Larsen Hall, P.O. Box 116200, Gainesville, FL 32611, USA; nmasna@ufl.edu (N.V.R.M.); dariando@ufl.edu (D.A.); shubhra.paul@ufl.edu (S.D.P.); khoraceherron@ufl.edu (K.H.-H.); dforte@ece.ufl.edu (D.F.); soumyajit@ece.ufl.edu (S.M.); swarup@ece.ufl.edu (S.B.)

**\*** Correspondence: shomaji@ufl.edu

† These authors contributed equally to this work.

**Abstract:** Dyeing vegetables with harmful compounds has become an alarming public health issue over the past few years. Excessive consumption of these dyed vegetables can cause severe health hazards, including cancer. Copper sulfate, malachite green, and Sudan red are some of the nonfood-grade dyes widely used on vegetables by untrusted entities in the food supply chain to make them look fresh and vibrant. In this study, the presence and quantity of dye-based adulteration in vegetables are determined by applying <sup>1</sup>H-nuclear magnetic resonance (NMR) relaxometry. The proposed technique was validated by treating some vegetables in-house with different dyes and then soaking them in various solvents. The resulting solutions were collected and analyzed using NMR relaxometry. Specifically, the effective transverse relaxation time constant, *T*2,*eff* , of each solution was estimated using a Carr–Purcell–Meiboom–Gill (CPMG) pulse sequence. Finally, the estimated time constants (i.e., measured signatures) were compared with a library of existing *T*2,*eff* data to detect and quantify the presence of unwanted dyes. The latter consists of data-driven models of transverse decay times for various concentrations of each water-soluble dye. The time required to analyze each sample using the proposed approach is dye-dependent but typically no longer than a few minutes. The analysis results can be used to generate warning flags if the detected dye concentrations violate widely accepted standards for food dyes. The proposed low-cost detection approach can be used in various stages of a produce supply chain, including consumer household.

**Keywords:** food adulteration; dye additives; nuclear magnetic resonance; relaxometry

#### **1. Introduction**

Food adulteration has reportedly increased over the last few years because of the complex supply chain of food from producer to consumer. Due to urbanization, consumers rely on growth, processing, transportation, and supply of food by multiple entities in the supply chain [1]. An untrusted entity can cause adulteration of food at any of these stages. Adulteration can take numerous forms, e.g., deliberate addition of substances with adverse health outcomes, not meeting desired product quality metrics, imitating other food substances, and using false labels on food packaging [2]. Human health is sensitive to food and thus can be affected by acute or chronic exposure to adulterated products. Even major health hazards, involving liver, vision, skin, and stomach disorders, are directly associated to adulterated food intake [3]. Foods like vegetables, fruits, fish, or meat adulterated with formalin have been found to be responsible for asthma and cancer [3]. Use of chemical pesticides has been linked to severe health problems, such as nerve damage and cancer [4]. There is also evidence that dye additives are responsible for genotoxicity, hypersensitivity, and carcinogenicity [5].

**Citation:** Shomaji, S.; Masna, N.V.R.; Ariando, D.; Deb Paul, S.; Horace-Herron, K.; Forte, D.; Mandal, S.; Bhunia, S. Detecting Dye-Contaminated Vegetables Using Low-Field NMR Relaxometry. *Foods* **2021**, *10*, 2232. https://doi.org/ 10.3390/foods10092232

Academic Editor: Theodoros Varzakas

Received: 6 August 2021 Accepted: 2 September 2021 Published: 21 September 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

Synthetic dyes are added to many foods to provide them with a fresh look and compensate for natural color variations. These dyes are often harmful for the health and may even be carcinogenic [6]. Therefore, it is very important to understand the ingredients of food items before consuming them. This information is generally available for packaged foods since genuine product labels include the names of any dyes within the list of ingredients. However, fresh fruits and vegetables are generally not labeled. Dishonest entities in the food supply chain can exploit this lack of information to add toxic dyes to fruits and vegetables that make them appear fresh and vibrant to customers. Some real-world examples of this practice are shown in Figure 1. Existing methods to detect many of these dyes have been thoroughly reviewed in [6]. For example, chromatographic, physiochemical, sensory, spectroscopy, and DNA-based detection methods have been combined with chemometrics for a wide range of adulteration-detection applications [7]. Detection approaches that are particularly suitable for dyes include capillary electrophoresis, electrochemical voltametric analysis, and amperometry [6–8]. To illustrate, carcinogenic compounds, like malachite green [9] and Sudan red [10], can be easily detected by liquid chromatography, gas chromatography, capillary electrophoresis, amperometry, and plasmon resonance light scattering [6]. Synthetic dyes are added to many foods to provide them with a fresh look and compensate for natural color variations. These dyes are often harmful for the health and may even be carcinogenic [6]. Therefore, it is very important to understand the ingredients of food items before consuming them. This information is generally available for packaged foods since genuine product labels include the names of any dyes within the list of ingredients. However, fresh fruits and vegetables are generally not labeled. Dishonest entities in the food supply chain can exploit this lack of information to add toxic dyes to fruits and vegetables that make them appear fresh and vibrant to customers. Some real-world examples of this practice are shown in Figure 1. Existing methods to detect many of these dyes have been thoroughly reviewed in [6]. For example, chromatographic, physiochemical, sensory, spectroscopy, and DNA-based detection methods have been combined with chemometrics for a wide range of adulteration-detection applications [7]. Detection approaches that are particularly suitable for dyes include capillary electrophoresis, electrochemical voltametric analysis, and amperometry [6–8]. To illustrate, carcinogenic compounds, like malachite green [9] and Sudan red [10], can be easily detected by liquid chromatography, gas chromatography, capillary electrophoresis, amperometry, and plasmon resonance light scattering [6].

vegetables, fruits, fish, or meat adulterated with formalin have been found to be responsible for asthma and cancer [3]. Use of chemical pesticides has been linked to severe health problems, such as nerve damage and cancer [4]. There is also evidence that dye

additives are responsible for genotoxicity, hypersensitivity, and carcinogenicity [5].

Foods 2021, 10, x FOR PEER REVIEW 2 of 11

Figure 1. Various instances of vegetables and other consumables being adulterated with harmful chemicals. In most cases, cheap, industrial-grade dyes are used instead of food colors to maximize profits [11–14]. **Figure 1.** Various instances of vegetables and other consumables being adulterated with harmful chemicals. In most cases, cheap, industrial-grade dyes are used instead of food colors to maximize profits [11–14].

Traditional methods have shown promising results in detecting food dyes with very high accuracy [6]. However, they have some limitations. Firstly, they require a labor-intensive set of tasks that ranges from sample preparation to analysis. Therefore, the experiments require a large expenditure of time and human effort, making them unsuitable for at-home and field applications. Secondly, some of these methods often require expensive instrumentation that is often unavailable in the low- and middleincome countries where dye-based adulteration is most common [7]. For example, NMR spectroscopy requires highly uniform magnets, which are bulky and expensive [15,16]. Thirdly, low-cost methods generally detect adulteration by observing anomalies in basic physical or chemical properties of the suspect substance (e.g., viscosity, pH, or electrical conductivity) [6,7]. However, modern "smart" adulteration techniques can bypass such simple detection methods [17]. Therefore, to confront the food adulteration issues, i.e., the deliberate or accidental contamination of food items with banned substances, the food industry, government bodies, and consumers need sensitive, rapid, reliable, inexpensive, widely applicable, and difficult-to-attack methods to detect adulterated foods. Spectroscopy meets many of these criteria and is promising for detecting adulteration. During spectroscopy-based analysis, the chemical composition of a food product is investigated by measuring its frequency-dependent absorption or Traditional methods have shown promising results in detecting food dyes with very high accuracy [6]. However, they have some limitations. Firstly, they require a laborintensive set of tasks that ranges from sample preparation to analysis. Therefore, the experiments require a large expenditure of time and human effort, making them unsuitable for at-home and field applications. Secondly, some of these methods often require expensive instrumentation that is often unavailable in the low- and middle-income countries where dye-based adulteration is most common [7]. For example, NMR spectroscopy requires highly uniform magnets, which are bulky and expensive [15,16]. Thirdly, low-cost methods generally detect adulteration by observing anomalies in basic physical or chemical properties of the suspect substance (e.g., viscosity, pH, or electrical conductivity) [6,7]. However, modern "smart" adulteration techniques can bypass such simple detection methods [17]. Therefore, to confront the food adulteration issues, i.e., the deliberate or accidental contamination of food items with banned substances, the food industry, government bodies, and consumers need sensitive, rapid, reliable, inexpensive, widely applicable, and difficult-to-attack methods to detect adulterated foods. Spectroscopy meets many of these criteria and is promising for detecting adulteration. During spectroscopy-based analysis, the chemical composition of a food product is investigated by measuring its frequency-dependent absorption or reflection spectra. Absorbance-based spectroscopy is mostly used for liquids, whereas reflection-based spectroscopy is used to identify fillers and adulterants, such as low-cost spices and dyes used to mask ageing. A variety of spectroscopic techniques, including near-infrared (IR), mid-IR, Raman, nuclear quadrupole resonance (NQR), and nuclear magnetic resonance (NMR), have been successfully used for

monitoring food quality [18]. Each technique has its own advantages and disadvantages, which makes the optimum choice strongly application dependent.

NMR is rapidly emerging as an important analytical technique for food analysis and screening [19]. NMR-based methods can be grouped into three major measurement categories: imaging, spectroscopy, and relaxometry. NMR spectroscopy has many applications in food analysis and adulterant detection. For example, it has been used to detect Sudan red in paprika powder with higher sensitivity than Raman or IR spectroscopy [10]. Nevertheless, NMR is intrinsically a bulk measurement method, so detecting adulterants at extremely low concentrations (e.g., parts per billion) remains challenging [10]. Moreover, high-resolution NMR spectroscopy requires a strong and highly uniform static magnetic field (known as *B*0). Such fields are typically generated using large cryogenically cooled superconducting coils, thus resulting in very high installation and maintenance costs. A recent work proves that cryogen-free, desktop-sized permanent magnets can provide a lower-cost alternative [20]. Nevertheless, such magnets must be temperature-stabilized and manually-calibrated, so costs are still quite high (typically at least \$20,000) [15,16]. As a result, complete NMR spectrometers (which combine the magnet with sample interrogation and readout electronics) cost \$50,000 or more. Thus, there is a need for lower-cost alternatives for analyzing food samples.

NMR relaxometry provides such an alternative since it can be performed in a relatively weak and inhomogeneous *B*<sup>0</sup> field, which in turn allows the size, complexity, and cost of the magnet to be greatly reduced [21,22]. Relaxometry focuses on measuring the nuclear spin relaxation times of specific substances present in a sample, namely the spin- lattice (*T*1) and spin-spin (*T*2) time constants; the translational diffusion coefficient (*D*) can also be measured. In a semi-classical picture, atomic nuclei with non-zero spin can be modeled as rotating magnetic dipoles. The static *B*<sup>0</sup> field tends to align these dipoles (by convention, along the *z*-axis) much like compass needles in the Earth's magnetic field, thus resulting in non-zero magnetization of the sample in thermal equilibrium. A second, time-varying magnetic field (known as *B*1) can be applied to perturb the magnetization away from equilibrium. Once *B*<sup>1</sup> is removed, the sample gradually returns to equilibrium; this process is known as relaxation [23]. Specifically, *T*<sup>1</sup> is the time constant for re-establishment of the equilibrium "longitudinal" magnetization, while *T*<sup>2</sup> is the time constant for decay of the non-equilibrium "transverse" magnetization.

The two parameters are generally not equal to each other (in almost all cases, *T*<sup>1</sup> ≥ *T*2) and also exhibit different dependencies on *B*<sup>0</sup> field strength and temperature [21].

Several studies have used <sup>1</sup>H-NMR and <sup>13</sup>C-NMR NMR spectroscopy to detect food dyes (e.g., azo dyes) in solution [24,25]. Azo dyes are water-soluble, organic compounds that contain a functional group of the form R−N = N−R', where R and R' are typically aromatic groups. These dyes are widely used in some foods and also in the textile industry; common examples include Sudan red, metanil yellow, and malachite green. However, the NMR relaxation properties of aqueous solutions of azo dyes have not been carefully studied. This paper seeks to use the *T*<sup>1</sup> and *T*<sup>2</sup> relaxation time constants to detect these dies in food samples. To the best of our knowledge, it is the first to show that NMR relaxometry can be used for rapid and low-cost detection of multiple dyes (including malachite green and Sudan red) present within common vegetables.

NMR relaxometry can be used to determine the presence and quantity of a target compound with the help of a reference sample and chemometric analysis. In this approach, relaxometry was first performed on a reference sample and its relaxation time recorded. Next, relaxometry was performed on the test sample, and the relaxation time was again recorded. Finally, the relaxation times were compared to detect the presence and quantity of the target compound. Several methods, including linear regression, comparison with internal and external standards, and comparison of relaxation spectra, were used to quantitatively analyze the resulting data [26,27]. Linear or nonlinear regression on *T*<sup>1</sup> and/or *T*<sup>2</sup> values is simple to implement and numerically stable, while finding and preparing an appropriate reference compound (i.e., internal or external standard) is sometimes

troublesome. However, both regression- and standards-based methods tend to fail for complex mixtures due to overlap between the *T*<sup>1</sup> and/or *T*<sup>2</sup> values of different components. Comparison of relaxation spectra generated using Laplace inversion is well-suited for such complex samples but suffers from limited resolution due to the numerically ill-conditioned nature of the inverse Laplace transform [27]. In this study, a simple and practical approach was developed for quantification of multiple food dyes by combining an external reference with nonlinear regression.

#### **2. Materials and Methods**

To simplify sample preparation, deionized (DI) water was used as the reference sample for all dyes, which is acceptable when only a single dye is present in a given test sample. The latter is a reasonable assumption since the goal of most dye-based adulteration is to impart a single color (e.g., green, orange, or red) to the vegetable or fruit in question. Finally, a general nonlinear regression method for quantitative analysis of the acquired relaxation data was used [10]. The details of this process are described next.

#### *2.1. Dyes and Vegetables*

A large number of chemical dyes have been used to make vegetables look fresh and vibrant [6], many of which are inedible and harmful to human health. For this study, three widely-used dyes were chosen: copper sulfate, malachite green, and Sudan red [28]. The first dye, copper (II) sulfate (CuSO4), is an inorganic compound that dissolves in water to produce a dark blueish-green solution. When dipped in this solution, green vegetables, like bitter gourds, peas, and cucumbers, turn dark or vibrant green. Unfortunately, CuSO<sup>4</sup> is poisonous if ingested in large quantities (>1 gm) [29], with symptoms ranging from slight nausea to severe gastrointestinal infections and other diseases [28]. For this study, three different green vegetables, namely bitter gourd, okra, and pointed gourd (also known as parwal), were purchased from a local store and dyed using copper sulfate. The second dye, malachite green, is the monochloride salt of an aromatic cation (a triarylmethane) with formula C23H25N<sup>2</sup> + [30]. It is generally used to color materials like leather or silk but because of its green hue is also illegally used to color vegetables, like peas and green chilies [28]. However, it is moderately toxic (even at concentrations as low as 0.1 µg/mL) and may also be carcinogenic [29]. In this study, yellow and green peas were dyed using malachite green. The third dye, Sudan red, is a reddish-orange lysochrome azo dye with formula C17H14N2O<sup>2</sup> [31]. This chemical is known to be carcinogenic and banned in food items but nevertheless continues to be illegally used to color red chilies, red chili powder, red capsicum fruits, red pepper, chili jam, and tomatoes [32,33]. In this study, red chilies were dyed using Sudan red. All the dyes were purchased from Sigma-Aldrich (St. Louis, MO, USA), while the vegetables were obtained from local grocery stores (Gainesville, FL, USA).

#### *2.2. NMR Relaxometry Instrumentation*

A block diagram of the overall experimental setup is illustrated in Figure 2a. The setup uses a benchtop permanent magnet (Spincore Technologies Inc., Gainesville, FL, USA) with a measured field strength of 0.5266 T at room temperature, resulting in a nominal <sup>1</sup>H-NMR resonance frequency of 22.6 MHz. A 3D-printed holder containing the solenoid probe coil and NMR sample tube is centered between the magnetic poles [34]. The holder is coupled to a commercial benchtop NMR spectrometer (Kea<sup>2</sup> , Magritek Inc., Malvern, PA, USA) through a two-capacitor impedance matching network [35]. The spectrometer is powered by two 12-V, sealed lead-acid (SLA) batteries with a capacity of 18 Ah (not shown in the figure) and connected to a personal computer using a USB interface. A proprietary graphical user interface (GUI)-based software, Prospa, is used to control the spectrometer and acquire experimental data.

Inc., Malvern, PA, USA) through a two-capacitor impedance matching network [35].

pacity of 18 Ah (not shown in the figure) and connected to a personal computer using

as a Faraday cage. Figure 2c shows the internal layout of this enclosure, while Figure

Figure 2b shows a photograph of the experimental setup. The permanent magnet,

The spectrometer is powered by two 12-V, sealed lead-acid (SLA) batteries with a ca-

a USB interface. A proprietary graphical user interface (GUI)-based software, Prospa, is used to control the spectrometer and acquire experimental data.

matching network, and sample holder are placed within a metallic enclosure that provides electromagnetic shielding from external radio frequency (RF) interference by acting

2d shows a more detailed view of the sample holder with a 10-mm thin-wall precision NMR tube (Wilmad-LabGlass, Vineland, NJ, USA) inserted into it.

Figure 2. (a) A block diagram of the experimental setup; (b) a picture of the actual measurement setup; (c) an inside view of the magnet enclosure; and (d) a picture of the 3D-printed sample holder parts with the coil and an NMR tube inserted. **Figure 2.** (**a**) A block diagram of the experimental setup; (**b**) a picture of the actual measurement setup; (**c**) an inside view of the magnet enclosure; and (**d**) a picture of the 3D-printed sample holder parts with the coil and an NMR tube inserted.

The probe coil was hand-wound using AWG 22 copper wire. The signal-to-noise ratio (SNR) of the NMR measurements [36] was maximized by iteratively optimizing the coil geometry to maximize its quality factor (Q) at the 1H-NMR resonant frequency (f0 ≈ 22.6 MHz). The final design consisted of a tightly-packed solenoid with a relatively short length-to-diameter ratio (L ≈ 2 cm and d ≈ 10 mm, resulting in L/d ≈ 2) but a relatively large number of turns (N = 13). Coil properties around f0 were measured Figure 2b shows a photograph of the experimental setup. The permanent magnet, matching network, and sample holder are placed within a metallic enclosure that provides electromagnetic shielding from external radio frequency (RF) interference by acting as a Faraday cage. Figure 2c shows the internal layout of this enclosure, while Figure 2d shows a more detailed view of the sample holder with a 10-mm thin-wall precision NMR tube (Wilmad-LabGlass, Vineland, NJ, USA) inserted into it.

using a vector network analyzer (E5071C, Agilent Technologies). The results (inductance = 840 nH, series resistance = 415 mΩ) confirm adequately high quality factor (Q 287) and self-resonant frequency (fSRF ≈ 130 MHz). The estimated position of the coil within the sample holder is shown in Figure 2d. 2.3. Methodology Instead of measuring the adulterant in situ, it was first washed out into solution. For this purpose, the sample (fruit or vegetable) was soaked in a solvent with known properties (e.g., DI water or brine) for a few minutes. The concentration of adulterant in the solvent was then measured using NMR relaxometry. This process has several The probe coil was hand-wound using AWG 22 copper wire. The signal-to-noise ratio (SNR) of the NMR measurements [36] was maximized by iteratively optimizing the coil geometry to maximize its quality factor (*Q*) at the <sup>1</sup>H-NMR resonant frequency (*f* <sup>0</sup> ≈ 22.6 MHz). The final design consisted of a tightly-packed solenoid with a relatively short length-to-diameter ratio (*L* ≈ 2 cm and *d* ≈ 10 mm, resulting in *L*/*d* ≈ 2) but a relatively large number of turns (*N* = 13). Coil properties around *f* <sup>0</sup> were measured using a vector network analyzer (E5071C, Agilent Technologies). The results (inductance = 840 nH, series resistance = 415 mΩ) confirm adequately high quality factor (*Q* 287) and self-resonant frequency (*fSRF* ≈ 130 MHz). The estimated position of the coil within the sample holder is shown in Figure 2d.

#### advantages, including (i) eliminating the effect of sample heterogeneity from the T<sup>1</sup> *2.3. Methodology*

and T2 measurements and (ii) greatly simplifying sample preparation. The acquired relaxation data were further analyzed in two steps: (i) library creation (Figure 3a) and (ii) quantifying the concentration of adulterant (Figure 3b). Our current implementation of both steps focused on T2 since it can be rapidly and accurately measured using Instead of measuring the adulterant in situ, it was first washed out into solution. For this purpose, the sample (fruit or vegetable) was soaked in a solvent with known properties (e.g., DI water or brine) for a few minutes. The concentration of adulterant in the solvent was then measured using NMR relaxometry. This process has several advantages, including (i) eliminating the effect of sample heterogeneity from the *T*<sup>1</sup> and *T*<sup>2</sup> measurements and (ii) greatly simplifying sample preparation. The acquired relaxation data were further analyzed in two steps: (i) library creation (Figure 3a) and (ii) quantifying the concentration of adulterant (Figure 3b). Our current implementation of both steps focused on *T*<sup>2</sup> since it can be rapidly and accurately measured using the well-known Carr–Purcell–Meiboom–Gill (CPMG) pulse sequence [37,38], but the procedure can be readily extended to include *T*<sup>1</sup> data (e.g., from an inversion recovery (IR) pulse sequence).

Figure 3. (a) Flowchart showing the process of creating a library; (b) flowchart showing the process of determining unknown concentrations; (c) measured variation of T1 and T2 values for the reference sample (DI water) over 10 experiments. **Figure 3.** (**a**) Flowchart showing the process of creating a library; (**b**) flowchart showing the process of determining unknown concentrations; (**c**) measured variation of *T*<sup>1</sup> and *T*<sup>2</sup> values for the reference sample (DI water) over 10 experiments.

#### **3. Results and Discussion**

#### 3. Results and Discussion *3.1. Library Creation*

(IR) pulse sequence).

3.1. Library Creation Calibration was carried out by using IRand CPMG pulse sequences to measure the T1 and T2 values of the reference sample, which is typically 12 mL of DI water. The measured values are T1 ≈ 2370 ms and T2 ≈ 2200 ms at room temperature (see Figure 3c). This value of T1 ≈ 2400 ms is in good agreement with earlier studies [39], while T2 is similar Calibration was carried out by using IRand CPMG pulse sequences to measure the *T*<sup>1</sup> and *T*<sup>2</sup> values of the reference sample, which is typically 12 mL of DI water. The measured values are *T*<sup>1</sup> ≈ 2370 ms and *T*<sup>2</sup> ≈ 2200 ms at room temperature (see Figure 3c). This value of *T*<sup>1</sup> ≈ 2400 ms is in good agreement with earlier studies [39], while *T*<sup>2</sup> is similar to *T*1, as expected for water [40].

the well-known Carr–Purcell–Meiboom–Gill (CPMG) pulse sequence [37,38], but the procedure can be readily extended to include T1 data (e.g., from an inversion recovery

to T1, as expected for water[40]. The next goal was to confirm that aqueous solutions of all three dyes under study exhibited T2 contrast, i.e., a reproducible dependence of T2 on dye concentration. For this, known quantities of each dye were dissolved in a fixed amount (100 mL) of reference sample (either DI water or 0.5% NaCl solution) to create a library of solutions. For convenience, a solution containing x gram of a particular dye was referred as "x% solution". Next, 12 mL of each solution was placed in an NMR sample tube and analyzed using a CPMG pulse sequence. The measured relaxation time constant is denoted by T2,eff to distinguish it from that of the reference sample (DI water). In each case, the CPMG echo spacing (tE) was kept small enough to ensure that molecular diffusion did not sig-The next goal was to confirm that aqueous solutions of all three dyes under study exhibited *T*<sup>2</sup> contrast, i.e., a reproducible dependence of *T*<sup>2</sup> on dye concentration. For this, known quantities of each dye were dissolved in a fixed amount (100 mL) of reference sample (either DI water or 0.5% NaCl solution) to create a library of solutions. For convenience, a solution containing *x* gram of a particular dye was referred as "*x*% solution". Next, 12 mL of each solution was placed in an NMR sample tube and analyzed using a CPMG pulse sequence. The measured relaxation time constant is denoted by *T*2,*eff* to distinguish it from that of the reference sample (DI water). In each case, the CPMG echo spacing (*tE*) was kept small enough to ensure that molecular diffusion did not significantly affect the value of *T*2,*eff* [37].

nificantly affect the value of T2,eff [37]. The smallest value of x (i.e., the sample weight) used within the proposed library was experimentally adjusted for each dye to ensure that the resulting change in T2,eff could be accurately estimated within a few scans. For this, the measured CPMG echo decay curves were fit to mono-exponential functions of the form Ae−ntE/T2,eff using leastsquares function minimization; here, A is the initial signal amplitude, and n = 1, 2,... is the echo number. Figure 4 shows the measured dependence of T2,eff on concentration for all three dyes. In each case, a monotonic decrease of T2,eff with concentration was The smallest value of *x* (i.e., the sample weight) used within the proposed library was experimentally adjusted for each dye to ensure that the resulting change in *T*2,*eff* could be accurately estimated within a few scans. For this, the measured CPMG echo decay curves were fit to mono-exponential functions of the form *Ae*−*ntE/T*2,*eff* using least-squares function minimization; here, *A* is the initial signal amplitude, and *n* = 1, 2,... is the echo number. Figure 4 shows the measured dependence of *T*2,*eff* on concentration for all three dyes. In each case, a monotonic decrease of *T*2,*eff* with concentration was observed; the effect is particularly strong for CuSO4. As a result, sample concentration can be unambiguously estimated from the measured value of *T*2,*eff* .

observed; the effect is particularly strong for CuSO4. As a result, sample concentration can be unambiguously estimated from the measured value of T2,eff. The underlying cause for the observed decrease in *T*2,*eff* with concentration is increased inter-molecular dipole-dipole (D-D) relaxation of the water molecules. Inter- molecular D-D relaxation is typically the dominant relaxation mechanism in dilute aqueous solutions [21]. It arises from time-varying fluctuations in the *B*<sup>0</sup> field seen by each nucleus due to random thermal motion of other molecules or ions in the solution (which act like miniature dipole field sources). In the case of CuSO4, the effect is dominated by random motion of the added Cu2+ ions, which contain unpaired electrons and are thus paramagnetic [41]. In the case of the organic dyes, the effect is likely dominated by slower motion (and thus increased D-D relaxation rates) [21] of the loosely-organized shell of water molecules that surrounds each dye molecule due to mutual electrostatic attraction. Each shell is in rapid chemical exchange with bulk water molecules, thus explaining the observed mono-exponential echo decay curves.

Foods 2021, 10, x FOR PEER REVIEW 7 of 11

Foods 2021, 10, x FOR PEER REVIEW 7 of 11

Figure 4. A library of vegetable dyes was created using the relationship between T2,eff and its concentration. This library can be used to quantify the amount of dye used in vegetable adulteration. The libraries exhibiting these trends are shown for 3 different dyes: (a) copper sulfate, (b) malachite green, and (c) Sudan red. Separate library functions are shown for two different experimental solvents, namely DI water and 0.5% NaCl solution. **Figure 4.** A library of vegetable dyes was created using the relationship between *T*2,*eff* and its concentration. This library can be used to quantify the amount of dye used in vegetable adulteration. The libraries exhibiting these trends are shown for 3 different dyes: (**a**) copper sulfate, (**b**) malachite green, and (**c**) Sudan red. Separate library functions are shown for two different experimental solvents, namely DI water and 0.5% NaCl solution. and are thus paramagnetic [41]. In the case of the organic dyes, the effect is likely dominated by slower motion (and thus increased D-D relaxation rates) [21] of the loosely-organized shell of water molecules that surrounds each dye molecule due to mutual electrostatic attraction. Each shell is in rapid chemical exchange with bulk water

The underlying cause for the observed decrease in T2,eff with concentration is increased inter-molecular dipole-dipole (D-D) relaxation of the water molecules. Intermolecular D-D relaxation is typically the dominant relaxation mechanism in dilute aqueous solutions [21]. It arises from time-varying fluctuations in the B0 field seen by The observed relationship between *T*2,*eff* and concentration for each dye was quantified using nonlinear regression, i.e., least-squares curve fitting. The resulting functions can be inverted to estimate unknown dye concentrations, as described in the next section. molecules, thus explaining the observed mono-exponential echo decay curves. The observed relationship between T2,eff and concentration for each dye was quantified using nonlinear regression, i.e., least-squares curve fitting. The resulting functions can be inverted to estimate unknown dye concentrations, as described in the next section.

#### each nucleus due to random thermal motion of other molecules or ions in the solution *3.2. Detection of Unknown Concentrations* 3.2. Detection of Unknown Concentrations

(which act like miniature dipole field sources). In the case of CuSO4, the effect is dominated by random motion of the added Cu2+ ions, which contain unpaired electrons and are thus paramagnetic [41]. In the case of the organic dyes, the effect is likely dominated by slower motion (and thus increased D-D relaxation rates) [21] of the loosely-organized shell of water molecules that surrounds each dye molecule due to mutual electrostatic attraction. Each shell is in rapid chemical exchange with bulk water The calibration curves described in the previous section were used to estimate the concentration of dye washed out from adulterated vegetables. For this purpose, nonadulterated vegetables were purchased from a local market, dyed by immersing them in the appropriate solution, and air-dried to remove extra liquid. Finally, the adulterated vegetables were soaked in the reference solvent (typically DI water) to wash out the dye. The *T*2,*eff* value of the solution was then analyzed using a CPMG pulse sequence. The calibration curves described in the previous section were used to estimate the concentration of dye washed out from adulterated vegetables. For this purpose, nonadulterated vegetables were purchased from a local market, dyed by immersing them in the appropriate solution, and air-dried to remove extra liquid. Finally, the adulterated vegetables were soaked in the reference solvent (typically DI water) to wash out the dye. The T2,eff value of the solution was then analyzed using a CPMG pulse sequence.

molecules, thus explaining the observed mono-exponential echo decay curves. The observed relationship between T2,eff and concentration for each dye was quantified using nonlinear regression, i.e., least-squares curve fitting. The resulting functions can be inverted to estimate unknown dye concentrations, as described in the next section. 3.2. Detection of Unknown Concentrations The calibration curves described in the previous section were used to estimate the A careful set of experiments was performed to determine the optimum samplepreparation procedure. Firstly, the optimum solution concentration for dyeing vegetables was determined. Figure 4 shows that the NMR setup can reliably detect concentrations as low as 0.1–0.3%. Thus, a higher concentration (1%) was used to dye each vegetable. Specifically, 1% CuSO<sup>4</sup> was used for pointed gourd, bitter gourd, and okra; 1% malachite green for peas; and 1% Sudan red for red dried chilies. The original (raw) and adulterated (dyed) vegetable samples are visually compared in Figure 5. A careful set of experiments was performed to determine the optimum sample-preparation procedure. Firstly, the optimum solution concentration for dyeing vegetables was determined. Figure 4 shows that the NMR setup can reliably detect concentrations as low as 0.1–0.3%. Thus, a higher concentration (1%) was used to dye each vegetable. Specifically, 1% CuSO4 was used for pointed gourd, bitter gourd, and okra; 1% malachite green for peas; and 1% Sudan red for red dried chilies. The original (raw) and adulterated (dyed) vegetable samples are visually compared in Figure 5.

aration procedure. Firstly, the optimum solution concentration for dyeing vegetables was determined. Figure 4 shows that the NMR setup can reliably detect concentrations as low Figure 5. Comparison between the raw and dyed vegetables: (a) raw okra, (b) okra dyed with copper sulfate, (c) raw peas, (d) peas dyed with malachite green, (e) raw red chilies, and (f) red chilies dyed with Sudan red. **Figure 5.** Comparison between the raw and dyed vegetables: (**a**) raw okra, (**b**) okra dyed with copper sulfate, (**c**) raw peas, (**d**) peas dyed with malachite green, (**e**) raw red chilies, and (**f**) red chilies dyed with Sudan red.

as 0.1–0.3%. Thus, a higher concentration (1%) was used to dye each vegetable. Specifically, 1% CuSO4 was used for pointed gourd, bitter gourd, and okra; 1% malachite green for peas; and 1% Sudan red for red dried chilies. The original (raw) and adulterated (dyed) vegetable samples are visually compared in Figure 5. Figure 5. Comparison between the raw and dyed vegetables: (a) raw okra, (b) okra dyed with copper sulfate, (c) raw The vegetables were soaked in the corresponding dye solutions for 3 h and then air-dried for 12 h in room temperature. Then, it was determined the optimum combination of reference solvent, temperature, and soaking time, *tsoak*, for washing out each dye. Firstly, both DI water and 0.5% NaCl solution were studied as reference solvents; the results were similar, so DI water was chosen for convenience. Secondly, the solvent temperature and soak time were varied. For water at room temperature, *tsoak* = 5, 60, and 180 min were used. For warm water at 60 ◦C, *tsoak* = 1, 2, and 5 min were used since the wash-out process (which is driven by diffusion) was expected to be significantly faster. Figure 6a–c show that *T*2,*eff* values decreased with time as more dye (CuSO4 in this case) washed out into solution; the rate of change was significantly higher for warm water, as expected. Similarly,

peas, (d) peas dyed with malachite green, (e) raw red chilies, and (f) red chilies dyed with Sudan red.

Foods 2021, 10, x FOR PEER REVIEW 8 of 11

Figure 6d,f confirm that (i) the estimated dye concentrations increased with time, and (ii) warm water could extract most of the dye within *tsoak* = 2 min, while much longer soak times were required at room temperature. concentration from T2,eff measurements) was repeated 10 times for each sample to ensure that the results are repeatable and consistent. The experiments confirm that both the presence of the chosen dyes and their extracted concentrations can be reliably estimated (with typical error < 4%) using the proposed technique.

air-dried for 12 h in room temperature. Then, it was determined the optimum combi-

dye. Firstly, both DI water and 0.5% NaCl solution were studied as reference solvents; the results were similar, so DI water was chosen for convenience. Secondly, the solvent temperature and soak time were varied. For water at room temperature, tsoak = 5, 60, and 180 min were used. For warm water at 60 °C, tsoak = 1, 2, and 5 min were used since the wash-out process (which is driven by diffusion) was expected to be significantly faster.

expected. Similarly, Figure 6d,f confirm that (i) the estimated dye concentrations increased with time, and (ii) warm water could extract most of the dye within tsoak = 2 min, while much longer soak times were required at room temperature. During the experiments, the optimized procedure described above (dyeing with 1% solution, drying for 12 h, soaking in warm water for 2 min, and finally estimating dye

Figure 6a–c show that T2,eff values decreased with time as more dye (CuSO<sup>4</sup>

The vegetables were soaked in the corresponding dye solutions for 3 h and then

in this case)

nation of reference solvent, temperature, and soaking time, tsoak, for washing out each

washed out into solution; the rate of change was significantly higher for warm water, as

Figure 6. Estimated values of T2,eff (top row) and concentration of extracted dye in solution (CuSO<sup>4</sup> , bottom row) as a function of time using water at room temperature and warm water at 60 °C: (a), (d) pointed gourd; (b), (e) bitter gourd; and (c), (f) okra. **Figure 6.** Estimated values of *T*2,*eff* (top row) and concentration of extracted dye in solution (CuSO4, bottom row) as a function of time using water at room temperature and warm water at 60 ◦C: (**a**,**d**) pointed gourd; (**b**,**e**) bitter gourd; and (**c**,**f**) okra.

3.3. Discussion While the experiments in the paper were focused on three common dyes, the proposed method can be extended to any dye that exhibits NMR relaxation contrast (in T1 and/or T2) while in aqueous solution. Compounds containing paramagnetic ions (such as Cu2+ or Ni2+) fall into this category since they result in increased intermolecular D-D relaxation rates. Compounds with permanent electric dipole moments, such as During the experiments, the optimized procedure described above (dyeing with 1% solution, drying for 12 h, soaking in warm water for 2 min, and finally estimating dye concentration from *T*2,*eff* measurements) was repeated 10 times for each sample to ensure that the results are repeatable and consistent. The experiments confirm that both the presence of the chosen dyes and their extracted concentrations can be reliably estimated (with typical error < 4%) using the proposed technique.

#### most azo and aryl dyes, may also exhibit a small amount of relaxation contrast due to *3.3. Discussion*

While the experiments in the paper were focused on three common dyes, the proposed method can be extended to any dye that exhibits NMR relaxation contrast (in *T*<sup>1</sup> and/or *T*2) while in aqueous solution. Compounds containing paramagnetic ions (such as Cu2+ or Ni2+) fall into this category since they result in increased intermolecular D-D relaxation rates. Compounds with permanent electric dipole moments, such as most azo and aryl dyes, may also exhibit a small amount of relaxation contrast due to the reduced mobility of water molecules in their associated hydration shells. Additional relaxation contrast can be obtained by performing *T*<sup>1</sup> measurements at different field strengths (e.g., by using an electromagnet to generate *B*0); this process is known as field-cycling relaxometry [42].

Besides generality, additional desirable features for the proposed food-adulteration detection platform include portability and cost-effectiveness. As noted earlier, NMR spectroscopy is expensive because of the need to generate a strong and highly uniform *B*<sup>0</sup> field. While the magnet size and cost requirements can be significantly reduced by focusing on relaxometry, the large size and power consumption of the spectrometer electronics (which includes an analog front-end and a digital back-end) remains a barrier for portable and low-cost applications. Fortunately, recent work has demonstrated miniaturized and low-power versions of both the front- and back-ends. For example, a portable NMR spectrometer based on a custom front-end and a low-cost system-on-chip (SoC) back-end has been developed [43]. Such miniaturized and low-cost devices can be used to replace the benchtop spectrometer used in the current setup.

#### **4. Conclusions**

This paper has demonstrated, for the first time to our knowledge, a simple, low-cost, yet powerful technique that combines NMR relaxometry with nonlinear regression-based trend modeling to detect and quantify harmful dyes in vegetables. Our experimental results show that the proposed technique can reliably quantify the presence of three commonly used illegal dyes, namely copper sulfate, malachite green, and Sudan red, at concentrations as low as 1 g/L (0.1%). The proposed technique can be used for detecting and potentially quantifying chemical dye-based produce adulteration in various stages of a supply chain, including retail facilities and consumer households. Future work will focus on extending our approach to a wider range of chemical dyes and food items as well as further enhancing the detection sensitivity.

**Author Contributions:** Conceptualization, S.M. and S.B.; data curation, S.S., N.V.R.M., S.D.P. and K.H.-H.; formal analysis, N.V.R.M., S.D.P., D.F., S.M. and S.B.; funding acquisition, S.M. and S.B.; investigation, S.S., N.V.R.M., D.A., K.H.-H., D.F., S.M. and S.B.; methodology, S.S., N.V.R.M., S.M. and S.B.; project administration, S.M. and S.B.; resources, D.F., S.M. and S.B.; supervision, D.F., S.M. and S.B.; validation, D.A. and K.H.-H.; writing—original draft, S.S., N.V.R.M., D.A., S.D.P., K.H.-H., D.F., S.M. and S.B.; writing—review and editing, S.S., N.V.R.M., D.A., S.D.P., K.H.-H., D.F., S.M. and S.B. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the National Science Foundation (grant no. 1563924).

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**

