Abstract
The use of cosmetic products for skin colour lightening or bleaching is a common practice across the globe. However human health effects associated with mercury in these products such as skin cancer and kidney failure have been reported. Conventional techniques used in the analysis of mercury such as atomic absorption spectroscopy and mass spectroscopy are destructive, time consuming and expensive. Although the energy dispersive X-ray fluorescence (EDXRF) spectroscopy method is rapid, non-destructive and requires minimal or no sample preparations, it has a high detection limit for mercury and therefore quantification of trace levels of mercury below 1ppm is challenging. This is mainly due to spectral overlaps, weak mercury fluorescence signals and extreme matrix effects. In this work, a novel chemometrics-assisted EDXRF spectroscopy method was utilized to realize rapid, direct detection and quantification of both low (< 1 ppm) and high mercury levels in skin whitening creams and lotions. 50 simulate (mixture of distilled water and pure glycerol) samples spiked with mercury concentrations ranging from a blank sample to 500 ppm were used for method development. The samples were then analyzed in triplicates for 900 seconds using the NeX EDXRF spectrometer set-up. Two chemometric techniques namely, principal component analysis (PCA) and artificial neural networks (ANNs) were used to perform exploratory analysis of the measured EDXRF spectra and quantification of the mercury levels. From PCA, it was found that the spectral data forms distinct clusters for both the low and high mercury concentration levels in the 9.6-10.4 keV Hg Lα and 11.2-12.4 keV Hg Lβ regions. Two ANN Hg concentration models were developed, one for ppb concentrations (0-1000 ppb; 30 samples) and the other for ppm concentrations (0-500 ppm; 17 samples). The R2, RMSEP, LOD and LOQ values for the two models were 0.72, 20.6%, 527 ppb, 819 ppb and 0.98, 4.6%, 3 ppm and 11 ppm respectively. The ppm model was used to ascertain sample results acquired by utilizing the conventional EDXRF method, most of which were closely matching while the ppb model established that two of the real samples registered a Hg content equal to 731±151 ppb. It may therefore be concluded that the chemometrics – EDXRF approach has potential for rapid, non-destructive and trace quantitative analysis of mercury compared to traditional measurement approaches. The technique is recommended for quality control and assurance of consumer products by the relevant regulatory authorities such as the Kenya Bureau of standards.
TABLE OF CONTENT
Declaration i
Acknowledgements ii
Abstract iii
TABLE OF CONTENT iv
LIST OF FIGURES vii
LIST OF TABLES viii
List of Abbreviations ix
CHAPTER ONE
INTRODUCTION
1.1 Background to the Study 1
1.2 Statement of the Problem 3
1.3 Objectives of the study 3
1.3.1 General objective 3
1.3.2 Specific objectives: 3
1.4 Justification and Significance of the Study 4
CHAPTER TWO
LITERATURE REVIEW
2.1 Overview 5
2.2 Assessment of mercury levels in skin lighteners around the world 5
2.3 Spectroscopy techniques used in the analysis of skin lighteners 7
2.3.1 A Survey of conventional spectroscopy methods 7
2.3.2 X-ray fluorescence spectroscopy 9
2.4 Utility of chemometrics in analytical spectroscopy 10
2.4.1 Exploratory analysis of spectral data 11
2.4.2 Preprocessing of spectral data 11
2.4.3 Quantification modeling of Spectral data 12
CHAPTER THREE
THEORETICAL BACKGROUND
3.1 Overview 14
3.2 The Sim Base Matrix Identification 14
3.3 Utility of chemometrics in EDXRF spectroscopy 14
3.3.1 Principal Component Analysis (PCA) 15
3.3.2 Artificial Neural Networks (ANNs) 16
3.4 The X-Ray Fluorescence (XRF) Spectrometer 21
3.5 Energy Dispersive X-Ray Fluorescence Spectrometry 22
3.6 Interaction of X-rays with matter 24
3.6.1 Radiation scattering 25
CHAPTER FOUR
MATERIALS AND METHODS
4.1 Overview 28
4.2 Instrumentation of EDXRF Spectroscopy 28
4.3 Base Matrix candidate identification 29
4.3.1 Spectra analysis for candidate skin lightening base matrices 31
4.4 Simulate samples preparation and analysis 33
4.5 Sample volume for analysis 36
4.6 Real Samples Preparation and Analysis 36
4.7 Multivariate Chemometric analysis of EDXRF spectra 37
4.7.1 Data Preprocessing 37
4.7.2 Spectra Analysis by PCA 38
4.7.3 Training ANN Models for Hg concentrations in ppb and ppm 38
CHAPTER FIVE
RESULTS AND DISCUSSION
5.1 Overview 40
5.2 EDXRF Spectra Analysis Results for Simulate Samples 40
5.2.1 Hg ROI Results for simulate samples 40
5.3 Results of EDXRF Spectra Data Analysis and Modelling Using Multivariate Chemometric Techniques 42
5.3.1 PCA Results for Hg concentrations (ppb) 42
5.3.2 The ANN modeling results for the ppb Data 43
5.3.3 PCA analysis Results for Hg concentrations (ppm) 46
5.3.4 The ANN modeling results for the ppm dataset 47
5.3.5 ANN Model Results for Real Samples 49
5.4 EDXRF Results for Simulate and Real Samples 50
5.4.1 Analysis results of the simulate samples 50
5.4.2 Analysis results of real samples (pastes) 50
5.4.3 Analysis results of real samples (liquids) 51
5.5 Comparison Between Conventional EDXRF and ANN model Results 53
CHAPTER SIX
CONCLUSIONS AND RECOMMENDATIONS
6.1 Conclusions 55
6.2 Recommendations 56
REFERENCES 57
APPENDICES 62
Appendix 1: Simulate sample preparation pictograms 62
Appendix 2: Real samples 63
Appendix 3: R-Scripts for the Multivariate Analysis Techniques 64
R-Script for PCA analysis of Hg concentrations (ppb) 64
Training Algorithm for the ANN Model with Hg concentrations (ppb) 67
R Script for PCA analysis of Hg Concentrations (ppm) 72
Training Algorithm for ANN model with Hg concentrations (ppm) 75
R script for calculating the LoD and LoQ for the ANN model with Hg concentrations (ppb) 78
LIST OF FIGURES
Fig. 3.1: Flow chart showing modeling steps… 18
Fig. 3.2: Model architecture 19
Fig. 4.1: EDXRF Schematic Instrumentation… 30
Fig. 4.2: The base matrix samples… 32
Fig. 4.3: Glycerol structure 33
Fig. 4.4: EDXRF scatter plot for the base matrices 33
Fig. 4.5: PCA analysis results for candidate matrices… 34
Fig. 4.6: Spectral data acquisition time… 37
Fig. 4.7: Curve for optimum sample volume 38
Fig. 5.1: Spectra plot for simulate samples 42
Fig. 5.2: Plot for spectral intensity Vs Hg concentration (ppm)… 43
Fig. 5.3: PC scores plot for Hg concentration (ppb)… 44
Fig. 5.4: The loadings plot for Hg concentrations (ppb) 45
Fig. 5.5: Trained model architecture used in this work… 46
Fig. 5.6: The ANN model used for quantification of Hg concentrations (ppb) 46
Fig. 5.7: PC scores plot for Hg concentrations (ppm)… 48
Fig. 5.8: The loadings plot for Hg concentrations (ppm) 48
Fig. 5.9: Trained model architecture for the ppm concentrations… 49
Fig. 5.10: The ANN model used for quantification of Hg concentrations (ppm) 50
Fig. 5.11: PC scores plot for real samples… 53
Fig. 5.12: Loadings plot for real samples… 54
LIST OF TABLES
Table 2.1: Conventional techniques for mercury identification and quantification… 9
Table 4.1: Simulate sample concentrations 36
Table 4.2: Classed Hg concentrations (ppb) 39
Table 4.3: Classed Hg concentrations (ppm) 40
Table 5.1: EDXRF Simulate samples results (ppm) 47
Table 5.2: Sample results by using the developed ANN model for Hg concentrations (ppb) 47
Table 5.3: ANN model statistics for Hg concentrations (ppb) 50
Table 5.4: ANN model statistics for Hg concentrations (ppm)… 51
Table 5.5: Sample results by using the developed ANN model for Hg concentrations (ppm)…51 Table 5.6: EDXRF paste samples results… 52
Table5.7: Real samples analysis results by using the ANN model for Hg conc. (ppb) 53
Table 5.8: Comparison between results by conventional EDXRF and ANN model for sim samples… 55
Table 5.9: Comparison between conventional EDXRF and ANN model results for real samples… 56
List of Abbreviations
AAS Atomic Absorption Spectroscopy
AFS Atomic Fluorescence Spectroscopy
ANNs Artificial Neural Networks
CV AAS Cold Vapour Atomic Absorption Spectroscopy
EDA Exploratory Data Analysis
EDXRF Energy Dispersive X-Ray Fluorescence
EXAFS Extended X-ray Absorption Fine Structure
FIAS Flow Injection Atomic Spectroscopy
GEXRF Grazing Emission X-Ray Fluorescence
GI XRF Grazing Incident X-Ray Fluorescence
ICP OES Inductively Coupled Plasma Optical Emission Spectroscopy
ICP MS Inductively Coupled Plasma Mass Spectroscopy
LCLS Linac Coherent Light Sources
PCA Principle Component Analysis
PLSR Partial Least Squares Regression
SVMs Support Vector Machines
SDD Silicon Drifted Detector
TXRF Total Reflection X-Ray Fluorescence
WDXRF Wavelength dispersive X-Ray Fluorescence
WHO World Health Organization
XRFEL X-Ray Free Electron Lasers
XANES X-Ray Absorption Near Edge Spectroscopy
XRF X-Ray Fluorescence
LOD Limit of detection
LOQ Limit of quantification
LLD Lower limit of detection
RMSEP Root mean square error of prediction
R2 Square of the regression value
CHAPTER ONE
INTRODUCTION
1.1 Background to the Study
The use of skin lightening products in the African continent is common because a lot of the women crave for a fair complexion free from spots, specks and imperfection, which is associated with beauty and youthfulness (Kamakshi, 2011). Most of these skin lighteners contain mercury as one of the ingredients but mostly with unspecified quantification and labeling (Al-Saleh, 2016). The presence of mercury in skin lighteners has been found to suppress the enzyme that produces melanin in the human body (Murphy et al., 2009). Thus, prolonged and extensive exposure to mercury poses a health risk to consumers and has been found to cause skin disorders, and in some cases infection to the brain, nervous system and kidneys (Mahé et al., 2007).
A WHO publication on mercury in skin lighteners showed that mercury concentrations ranged from 1% to 10%, while other beauty products such as facial lightening creams contained concentrations of up to 33%, without the companies selling the products listing mercury as one of the ingredients (Campbell et al., 2003; Dlova et al., 2014). The CV AAS method has been used to detect and quantify mercury with a detection limit of a few ppt (Višnjevec et al., 2014). However, positive interference of spectral signals remains a challenge to this technique. ICP MS having an LOD of up to 1 ppt has also been used for analysis of mercury but requires special sample preparation and is also expensive (Bailey et al., 2003). ICP OES has also been used to detect and quantify mercury in skin lightening lotions (Nguyen et al., 1998). However, the method is destructive, has poor detection limit and low sensitivity (Eschnauer et al., 1989).
The XRF spectroscopy method in its various modalities i.e. the EDXRF and the WDXRF is a rapid, non-destructive method with minimal or without preparation of sample and can be applied in a wide ranging concentration of elements (solid, powder and liquid). In XRF spectrometry, the spectral lines are used to identify the elements. Such lines, grouped in the series K, L, M, N, etc are the characteristic line series for every element (Antwerpen et al., 2006). Light elements produce K line series, middle-ranging elements give rise to K as well as L lines, with the heavy ones emitting K, L and M line series (Bote et al., 2009).
The choice of an analysis line depends on the sample type, elements present in the sample, range of elemental concentration and the conditions for excitation. The L shell emits an electron which fills a vacancy in the K shell and thereby producing a Kα line radiation. When an electron from the M shell fills the same vacancy in the K shell, a Kβ line radiation is obtained, while Kγ line radiation is given out when an electron from the N shell fills the vacancy.
X-ray fluorescence is capable of analyzing quantitatively multi-elemental composition of samples in which the detection limits are in ppb range, depending on the sample form and spectrometer excitation conditions such as tube conditions and strength of the radioisotope source. During analysis, high energy photons strike the target material, exciting electrons in the core levels of atoms in the material (Shibata et al., 2009). This causes de-excitation through characteristic fluorescent radiation whose energy is used for identification based on intensity and elemental concentration in the sample. In XRF spectroscopy, incident X-ray photons produce scatter radiation when they interact with electrons in atoms of the target element (Wobrauschek et al., 2010). This radiation is either Rayleigh (coherent) scattering or Compton (incoherent) scattering. Coherent scattering occurs when energy is conserved during collision between the incident beam and sample while incoherent scattering is produced when some energy is lost by the scattered photons (Marguí et al., 2009).
However, detection limit, spectral overlaps produced by the L and M series as a result of inter- element effects (matrix effects) and weak fluorescence signals for light elements with atomic number below sodium (i.e. Z<11) still remain a challenge in this technique (Nguyen et al., 1998).
Spectral complexity caused by matrix effects in the XRF analysis of complex samples make spectrum evaluation difficult as well as deconvolution of resultant fluorescence intensities into respective concentrations (Wobrauschek et al., 2010). In classical EDXRF, the concentration of all elements in the sample must be known in order to deal with matrix effects challenge. The use of chemometrics-assisted EDXRF spectroscopy method in this work overcame the challenges and attained a direct detection and quantification of trace level mercury in varying sample concentrations. Chemometric techniques such as PCA and ANNs are robust analytical tools which utilize mathematical, statistical and computational methods to reveal hidden relationships in data sets (Luo, 2006; Reinholds et al., 2015; Worley et al., 2013).
1.2 Statement of the Problem
Prolonged exposure to unregulated and unquantified mercury levels in creams and lotions used for skin lightening poses a health hazard to the product consumers. There is therefore need to perform quality assurance tests of the skin lighteners in order to ascertain that the mercury content in them does not exceed the World Health Organization limit of 1 ppm, (Bose-O’Reilly et al., 2010). Current spectrometric techniques that have been used in the detection and quantification of mercury include AAS, AFS and ICP- MS. They have had limitations in that they involve wet – chemistry, are destructive to the samples, have poor limit of detection and high cost (von Burg, 1995). EDXRF spectroscopy which is rapid and nondestructive to the samples has also been used, though challenged by spectral overlaps associated with Kα and Lβ lines of trace level elements as well as a LLD (Melquiades et al., 2015), (Antwerpen et al., 2006). In this research a chemometric assisted EDXRF technique will be used to overcome the aforementioned challenges and also address the problem of unspecified and unquantified mercury levels in skin lighteners.
1.3 Objectives of the study
1.3.1 General objective:
To perform rapid detection and quantification of mercury levels in skin lighteners utilizing a chemometric-assisted EDXRF spectroscopy approach.
1.3.2 Specific objectives:
(i) Identify an appropriate skin lightening base matrix material and acquire EDXRF spectra from simulate samples spiked with a wide range of mercury concentrations.
(ii) Develop a multivariate chemometric calibration model for mercury quantification using the EDXRF spectra obtained from specific objective (i) above.
(iii) Perform the analysis of a wide variety of skin lighteners obtained from the local markets using the chemometric-assisted EDXRF technique to determine their mercury content.
(iv) Compare the predicted mercury concentrations in both the simulate and real samples using the developed model with those obtained using conventional EDXRF.
1.4 Justification and Significance of the Study
Negative effects caused by the wide spread use of skin lightening products containing unspecified and unquantified mercury levels has raised a global health concern (Hamann et al., 2014). Poor quality control and lack of proper labeling of the contents in these products has been the major challenge. Although a number of spectrometric techniques including EDXRF spectroscopy have been used in the determination of mercury, rapid and direct analysis of low concentrations i.e. less than 0.1 ppm is still a challenge (Orisakwe et al., 2013). In the case of EDXRF, the spectral overlap of the Hg energies (9.6 – 12.4 keV) and other trace elements such as Pb and As, complicate the quantification of low mercury levels.
This research work utilizes chemometric techniques namely, the principal component analysis (PCA) and artificial neural networks (ANNs) to perform EDXRF spectral deconvolution and exploratory data analysis in the model development for rapid, direct detection and quantification of mercury levels of both high (> 3 ppm) and low (< 1 ppm) mercury levels in skin lighteners. The technique has also overcome the challenge of extreme matrix effects and spectral overlaps associated with the conventional EDXRF spectroscopy.
Login To Comment