Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Rapid Assessment of Exercise State through Athlete’s Urine Using Temperature-Dependent NIRS Technology

Rapid Assessment of Exercise State through Athlete’s Urine Using Temperature-Dependent NIRS... Hindawi Journal of Analytical Methods in Chemistry Volume 2020, Article ID 8828213, 7 pages https://doi.org/10.1155/2020/8828213 Research Article Rapid Assessment of Exercise State through Athlete’s Urine Using Temperature-Dependent NIRS Technology 1 2 2 2 1 1 Lihe Ding, Lei-ming Yuan , Yiye Sun, Xia Zhang, Jianpeng Li, and Zou Yan School of Physical Education & Sport Science, Wenzhou Medical University, Wenzhou 325035, China College of Electric & Electronic Engineering, Wenzhou University, Wenzhou 325035, China Correspondence should be addressed to Lei-ming Yuan; yuan@wzu.edu.cn Received 9 April 2020; Accepted 18 August 2020; Published 29 August 2020 Academic Editor: Alessandro Buccolieri Copyright © 2020 Lihe Ding et al. )is is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Athletes usually take nutritional supplements and perform the specialized training to improve the performance of sport. A quick assessment of their athletic status will help to understand the current physical function of athletes’ status and the effect of nutritional supplementation. Human urine, as one of the most important body indicators, is composed of many metabolites, which can provide effective monitoring information for physical conditions. In this study, temperature-dependent near-infrared spectroscopy (NIRS) technology was used to collect the spectra of athlete’s urine for evaluating the feasibility of rapidly detecting the exercise state of the basketball player. To obtain the detection results accurately, several chemometrics methods including principal component analysis (PCA), variables selection method of variable importance in projection (VIP), continuous 1D wavelet transform (CWT), and partial least square-discriminant analysis (PLS-DA) were employed to develop a classifier to distinguish the physical status of athletes. )e optimal classifying results were obtained by wavelet-PLS-DA classifier, whose average precision, sensitivity, and specificity are all above 0.95, and the overall accuracy of all samples is 0.97. )ese results demonstrate that temperature-dependent NIRS can be used to rapidly assess the physical function of athlete’s status and the effect of nutritional supplementation is feasible. It can be believed that temperature- dependent NIR spectroscopy will obtain applications more widely in the future. human body is one of the most important outputs composed 1. Introduction of many metabolites. It has been widely used for assessment To achieve a high level of athletic performance, hard of human health and disease diagnosis, and it will help to trainings and reasonable dietary nutritional supplements quickly understand the body’s physical function and nu- are necessary. However, up to now, there is no rapid tritional diet [3, 4]. Creatine, as an auxiliary nutritional monitoring method to assess the exercise state of athlete supplement, will be metabolized and decomposed after and the effect of nutritional supplementation. Conse- consumption and eventually be eliminated from the body by quently, it usually leads to the over or under training for urination [5]. )erefore, urine can be used as an analytical athlete and affects the performance and physical health of test to assess whether the athlete takes the sport’s supple- athlete [1]. )erefore, it is necessary to develop a new ment creatine. Traditional methods of detecting the sup- technology to accurately assess the exercise state and plement creatine usually have the disadvantages of complex physical function of athletes. )is will provide scientific sample preparation, being time-consuming, and the need to training guidance and scientific information of nutritional use toxic and harmful reagents [6–8], which cannot meet the supplements for athletes. needs of rapid and accurate detection. )erefore, it is Body metabolites contain a variety of primary and necessary to develop a new technology to rapid detect the secondary metabolites, which can reflect athlete’s physical status of basketball player before and after taking creatine. function and state and provide the most intuitive infor- Near infrared spectroscopy (NIRS) is known as a fast mation for human health and exercise states [2]. Urine of and nondestructive analysis technology in the wavelength 2 Journal of Analytical Methods in Chemistry range of 780–2500 nm. NIRS has been widely applied in 2. Materials and Methods food, agriculture, biology, and chemistry fields [9–14]. At 2.1. Sample Collection and Preparation. 15 male basketball present, many researchers have used NIRS to qualita- players at the age range of 18 to 23 (weight range of 75± 5 kg) tively and quantitatively analyze the metabolites of ath- from Wenzhou Medical University were convened for this letes in sports medicine fields, such as quantitative experimental trials. Participants were interviewed to obtain determination of glucose content in urine [15, 16], rapid body information including drug intake, nutritional sup- detection, and analysis of trace protein in urine [17]. plements, past medical diseases, and anthropometric data. However, the lack of the spectral information greatly All participants agreed to participate in the study and signed limits the application broad of NIRS [18]. To solve this the informed consent. problem, temperature-dependent near-infrared spec- Before the urine collection, all athletes are required to troscopy technology was proposed [19], which added the write down the dietary information on the day of urine information of temperature into spectral dataset. )us, collection and to drink 100 ml of water 3 hours before more useful spectral information about temperature urine collection and underwent 48 hours without exercise. influence on analytes can be used to construct the In this experiment, all athletes will have their urine col- multivariate model. As a new type of spectroscopy lected twice before and after the specialized exercise. 15 ml technology, a few applications have been applied and urine sample was first collected before training. All ath- good analytical results were also obtained [20, 21]. letes will take 1.5 g creatine before training and then train Compared with traditional NIRS technology, tempera- for 120 minutes, and the other 15 ml urine sample will be ture-dependent NIRS technology can effectively increase collected after 5 minutes of rest. When the collection of the amount of spectral information and help improve the urine was finished, all urine samples were bottled into accuracy and stability of calibration model. 15 ml centrifuge tubes and immediately refrigerated at Spectral data, especially temperature-dependent near- −20 C for future use. infrared (NIR) spectra data, is a kind of high-dimensional data, containing mass spectral information. )erefore, development of multivariate calibration model is usually 2.2. Collection Temperature-Dependent NIRS Data. )e required for dimensional reduction, denoising operation, collection of temperature-dependent NIRS data from 4000 variables selection, and vector projection [22]. )e −1 to 12000 cm was performed on a Vertex 70 spectrometer commonly used method for dimensional reduction is (Bruker Optics Inc., Ettlingen, Germany). )e temperature principal component analysis (PCA) method [23], which control equipment used in this study is the 2216e temper- can be effective in extracting several principal components ature controller (Bruker Optics Inc., Ettlingen, Germany), from high-dimensional spectra data and reducing the which can provide a precision temperature (±0.1 C). In this dimension of spectra data. In this paper, temperature- study, the temperature range of this experiment is from 20 C dependent NIRS with a temperature dimension that has ° ° to 50 C with a step of 5 C, and the urine will be kept in the one more dimension than that of traditional spectral condition of 7 temperature points orderly from low to high. dataset. Hence, suitable chemometrics method is neces- To increase the ratio of signal to noise and reduce the sary to process these high-dimensional spectral data. In random errors, three spectra of each urine samples with scan addition to the PCA, the classic partial least squares number 64 were collected at each temperature. Finally, the discriminant analysis (PLS-DA) model was also consid- NIR matrix 210 × with 2074 columns and 210 rows were ered to discriminate the pre- and postexcise group [24]. obtained for analysis. Additionally, another two analysis methods of variable importance in projection (VIP) [25] and continuous 2.3. Multivariate Analysis Methods and Model Evaluation wavelet transform (CWT) [26, 27] were also introduced to further improve the classifying performance of the cali- 2.3.1. Principal Component Analysis (PCA). Principal bration model. component analysis (PCA) is one of the most used methods )e main research objective of this study is to verify in chemometrics. It is usually implemented to reduce the the feasibility of applying temperature-controlled NIRS dimensionality of dataset and provide the score plot for technology to quickly discriminant analysis of pre- and visualizing the distribution of samples [28]. As a commonly postexercise states of basketball players after eating cre- used exploratory method, PCA can make the primary atine. )e specific goals are as follows: (1) collecting NIR evaluation of similarity between samples’ classes. )e main data of urine samples at a series of temperature condi- principal of PCA is to convert a set of correlated variables tions; (2) making the preliminary spectral exploration into several linearly independent principal components with varying temperatures and sample’s visualization of (PCs) using orthogonal transformation [29]. As a result, the spectral data using PCA algorithm; (3) constructing the dimensionality of dataset is greatly reduced through PCA multivariate calibration classifiers between NIRS dataset analysis. and exercise state using PLS-DA algorithm; (4) identifying the optimal variables and enhancing the resolution of spectra using VIP and CWT algorithms; (5) comparing 2.3.2. Classification Algorithm. )e classical linear classi- performances of all classifiers and identifying the best fication method of PLS-DA algorithm was applied in this detection classifier. study. )e main principle of PLS-DA is to extract several Journal of Analytical Methods in Chemistry 3 latent variables (LVs), which are the linear combination intensity, it can be found that the spectral intensity of the original variables from the independent variables in generally shows a downward trend with increasing the format of matrix X. )en the relationship between temperature except for the spectra of 30 C. Besides, with independent variables X and dependent variables Y is increasing temperature of urine sample, the spectral in- −1 established and the prediction value of each sample is tensity at 11250 cm is bizarre and decreases with big ° ° ° obtained by this developed relationship. To achieve jump at low temperature (20 C, 25 C, and 30 C) but small sample classification, class of each sample is determined slope at high. )e possible reason is that the substances in ° ° according to the threshold value of class which is cal- urine from 25 C to 30 C have been changed. On the other culated and identified by Bayesian statistics [24]. Finally, hand, although the spectral intensity decreases with the the class of all samples is identified using PLS-DA temperature, it is difficult to fit using linear functions. classifier. Based on the above analysis, it can be found that it is hard to directly distinguish the samples before and after training only by observing the intensity of spectral curve. 2.3.3. CWT and VIP Methods. To enhance the resolution of )erefore, multivariate calibration model will be applied spectral, remove the noise/uninformative variables, and in the subsequent analysis to further analyze the spectral improve the performance of calibration model, continuous dataset. 1D wavelet transform (CWT) was applied in this paper [30]. During the implement, “Sym2” wavelet filter and the scale parameter 20 were used to process the NIRS data. As known 3.2. PCA of Temperature-Dependent NIRS Data. Prior to the that there are many redundant and uninformative variables calibrating analysis, it is recommended to explore the involved in the spectral matrix that will lead to the poor structure of spectral dataset. In this study, the effective performance of the calibration model. To solve this issue, statistical method of PCA was used to explore and vi- variable selection was commonly used to identify the op- sualize the space distribution of samples by extracting timal variables prior to establishing the multivariate model. several new principal components from high-dimen- In this study, variable importance in projection (VIP) was sional dataset. PCA was firstly performed on the tem- considered to select the optimal variables. During the perature-dependent NIRS dataset to plot the score scatter processing stage of VIP, V value that reflects the contri- of samples and observe the sample’s distribution. bution of each variable was calculated based on the PLS Figure 2(a) shows the PCA score plots of 210 urine regression to identify the optimal variable [31]. Finally, those spectral samples collected from before and after training. optimal variables with V value larger than 1 were eventually It is clear that PC1 and PC2 account for 92.94% and 6.48% selected. contributions to the original spectra, respectively, and the total cumulative variance contributes close to 99.42%. However, the samples from before and after training were 2.3.4. Model’s Evaluation. In this study, four evaluation still overlapped in these two-dimensional space, which parameters, namely, sensitivity, specificity, precision, and indicated that the urine samples from before and after accuracy, are used to evaluate the performance of the training were difficult to separate based on first two PCs. classification model. )ey can accurately and objectively To further explore the distribution of samples, 210 evaluate the performance of PLS-DA classification model. A ° ° samples coming from different temperature of 20 C–50 C classification model with good performance should have the were considered. high value of sensitivity, specificity, precision, and accuracy [24, 32]. All algorithms and calculations are carried out in Taking a close observation on Figure 2(b), it shows the the MATLAB 2015b environment ()e Math Works, Natick, distribution of samples with some linear relationship at USA). the same temperature. In addition, with rising of the temperature of urine sample, samples are distributed 3. Results and Discussion from the lower right to the upper left except for 30 C. Moreover, there is no sample overlapped among different 3.1. Spectral Profile Analysis. In this study, temperature- temperatures. )ese indicate that the temperature will dependent NIRS data of basketball players’ urine were significantly affect the NIRS response signals of urine collected in temperature ranges from 20 to 50 degrees samples. But PCA is an unsupervised method and cannot Celsius in steps of 5. )e corresponding spectral profiles accurately distinguish the categories of urine samples. are shown in Figure 1, where the average spectral curves )e possible reason is that the change of substances of the seven temperatures are collected before and after composition in urine before and after exercise is small, the basketball players’ exercise. It can be found that there −1 and the unsupervised principal component analysis are several obvious absorption peaks at 7800 cm , −1 −1 method cannot fully exhibit this change only using 9300 cm , and 10800 cm , which may be caused by several PCs. As a result, PCA failed to accurately dis- function groups of C-H, H-O, N-H, and C-O band in tinguish these samples that belong to two classes. To solve water, glucose, and protein from the urine [2]. Obviously, this issue, more information needed to be mined from there is no significant difference in spectral profiles be- NIRS data, and the classic PLS-DA classification model tween before and after exercise of basketball player by will be performed to construct the classification model. eye-naked judgement. On the consideration of spectral 4 Journal of Analytical Methods in Chemistry 1.0 1.2 1.0 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0.0 4500 6000 7500 9000 10500 12000 4500 6000 7500 9000 10500 12000 –1 –1 Wavenumber (cm ) Wavenumber (cm ) 20°C 40°C 20°C 40°C 25°C 45°C 25°C 45°C 30°C 50°C 30°C 50°C 35°C 35°C (a) (b) Figure 1: Average spectrum of basketball player’s urine with different temperature measurements. (a) Before exercise; (b) after exercise. 2 2 1.5 1.5 1 1 0.5 0.5 0 0 –0.5 –0.5 –1 –1 –1.5 –1.5 –2 –2 –2.5 –2.5 –6 –4 –2 0 26 4 –6 –4 –2 0 24 6 PC1 (92.94%) PC1 (92.94%) Before 20°C 40°C After 25°C 45°C 30°C 50°C 35°C (a) (b) Figure 2: Two-dimensional PCA and 2D-PCA analysis maps for the two classes of urine samples. (a) PCA plots of 210 urine spectral samples at 20 C. (b) PCA plots of urine samples at different temperature. 3.3. PLS-DA Model Based on Full and CWT Variables. First of all, PLS-DA classification model was established Based on the above analysis, the unsupervised PCA method based on the full variables of raw spectra to distinguish the urine cannot directly distinguish urine samples from the class of sample. In Table 1, it can be seen that the values of sensitivity, before and after training. )erefore the multivariate mod- specificity, and precision in calibration set are all 100%. eling method of PLS-DA was used to create the classification However, the result of validation and prediction set is lower than that in the calibration set. For urine samples from the model. Prior to establishing the PLS-DA model, 210 samples were randomly divided into the calibration set and pre- before-training group, the precision, sensitivity, and specificity diction set with the ratio of 2 :1. )en the classification in prediction set are 0.78, 0.82, and 0.71, respectively; mean- model was established based on the calibration set, and the while, the after-training group is 0.76, 0.71, and 0.82, and the number of latent variables (LVs) involved in PLS-DA model overall classification accuracy of all samples is 0.77. )is result was optimized using 10-fold cross-validation and was de- indicates that it is possible to rapidly detect the athletic state of termined at the lowest root mean square error of cross- athlete using NIRS technology. To further improve the per- validation (RMSECV). )e specifically calculated results are formance of classification model, continuous 1D wavelet shown in Table 1. transform was proposed to transform the temperature- PC2 (6.48%) Intensity PC2 (6.48%) Intensity Journal of Analytical Methods in Chemistry 5 Table 1: Classification results of PLS-DA model based on raw variables, optimal variables selected from raw spectra, and optimal variables from CWT spectra. Calibration Validation Prediction Model Methods LVs Class 2 3 4 Pre Sen Spe Pre Sen Spe Pre Sen Spe 1 1.00 1.00 1.00 0.71 0.83 0.69 0.78 0.82 0.71 — 7 2 1.00 1.00 1.00 0.82 0.69 0.83 0.76 0.71 0.82 1 1.00 1.00 1.00 0.88 0.91 0.89 0.88 0.90 0.83 VIP 5 2 1.00 1.00 1.00 0.92 0.89 0.91 0.86 0.83 0.90 PLS-DA 1 1.00 1.00 1.00 0.99 1.00 0.99 0.95 1.00 0.94 Wavelet 8 2 1.00 1.00 1.00 1.00 0.99 1.00 1.00 0.94 1.00 1 1.00 1.00 1.00 0.86 0.86 0.82 0.66 0.81 0.75 Wavelet-VIP 9 2 1.00 1.00 1.00 0.82 0.82 0.86 0.87 0.75 0.81 Notes: 1: latent variables; 2: precision; 3: sensitivity; 4: specificity. dependent NIRS dataset, and this would help improve the related to urine of basketball player were used to construct the PLS-DA classification model and the corresponding resolution of spectra. )en the PLS-DA classification model was built as in the previous step, and its performance was compared results were shown in Table 1. It can be seen that the with the full-variables-based PLS-DA model. In Table 1, it can performance of VIP-PLS-DA model in terms of the model be found that the precision, sensitivity, and specificity of class 1 precision (0.78 versus 0.88 and 0.76 versus 0.86), sensi- (labelled “before training”) and class 2 (labelled “after training”) tivity (0.82 versus 0.90 and 0.71 versus 0.83), and speci- in prediction set are 0.95, 1.00, and 0.94 and 1.00, 0.94, and 1.00, ficity (0.71 versus 0.83 and 0.82 versus 0.90) has been and the overall accuracy reached 0.97. Obviously, the perfor- improved when compared with FULL-PLS-DA model. )e possible reason for this improvement is that the mance of PLS-DA classification model has been significantly improved compared to that of full-variables-based PLS-DA unrelated variables may be deleted by VIP algorithm and the model becomes more stable and accurate. However, model. )e result shows that the continuous 1D wavelet transform (CWT) is an effective way to enhance the resolution the opposite result was obtained when the wavelet-VIP- PLS-DA model was considered compared with FULL- and further improve the accuracy of model. PLS-DA model. Specifically, there are no significant differences between 3.4. PLS-DA Model Based on Optimal Variables. Although a performances of these PLS-DA models in class 2, but the high classification accuracy has been obtained by the CWT- performance for class 1 is worse than FULL-PLS-DA model. PLS-DA model, the calculating process of these models is It demonstrates that although the continuous 1D wavelet can complex due to too many variables involved in the calcu- improve the performance of PLS-DA, the subsequent var- lation model. Moreover, it is known that there are many iable selection may not be suitable when the spectral data irrelevant, collinear, and redundant variables in spectral data was transformed by continuous 1D wavelet. )erefore, only which will lead to poor performance of classification model. CWTpretreatment is the better way to analyze NIRS data. In )erefore, it is necessary to hunt suitable variable selection addition, it can be found that there are many noise variables algorithms to identify a few important variables before −1 −1 in the range of 4000–5400 cm and 6500–7200 cm in establishing PLS-DA classification model [33]. Employ- Figure 3(a), which are contrary to the characteristic of ments of variable selections not only help reduce the di- continuous spectrum. To eliminate the effect of those noise mensionality of spectral data and the complexity of variables, we also performed the calculation only using the computation but also improve the accuracy and robustness −1 variables from the range of 11600–12000 cm and of the model. For this reason, classic variable selection al- −1 −1 5400 cm –6500 cm , and the overall accuracy is about gorithm VIP was applied on raw spectra data and the 0.85, which is consistent with the VIP-PLS-DA model. )is spectral data processed by continuous 1D wavelet transform means those unstable variables in the range of algorithm. −1 −1 4000–5400 cm and 6500–7200 cm do not work or even Results of variables selection through VIP are shown negatively affect the PLS-DA classification model. in Figure 3. According to the criterion of VIP algorithm, When all classification models, including FULL-PLS- those variables will be selected as the optimal variables DA, VIP-PLS-DA, wavelet-PLS-DA, and wavelet-VIP-PLS- whose VIP is larger than 1. )erefore, variables above the DA, are considered and compared, the best classification bold black line in Figure 3 are selected as optimal vari- model is wavelet-PLS-DA, whose the overall accuracy rea- ables. Specifically, 604 and 483 feature variables were ches 0.97, and the better one is VIP-PLS-DA model. )e identified from raw spectrum and wavelet-transformed worst one is the wavelet-VIP-PLS-DA model with accuracy spectrum. Compared with 2074 full variables, the number of 0.77. )ese results demonstrate that it is feasible to use of variables was decreased by 70.87% and 76.71%, and the multivariate calibration model and NIRS data to determine dimension of the spectra had been greatly decreased. the exercise status of athlete. In this primary study, there is When the variables selection was completed, those feature still a lot of work that needs to be further improved and variables that contained the most useful information 6 Journal of Analytical Methods in Chemistry 3.5 3.5 2.5 2.5 1.5 1.5 0.5 0 0.5 4000 6000 8000 10000 12000 4000 6000 8000 10000 12000 –1 –1 Wavenumber (cm ) Wavenumber (cm ) (a) (b) Figure 3: Optimal variables selected by VIP from raw spectrum (a) and wavelet transformed spectrum (b). Prevention, diagnosis, and treatment of the overtraining supplemented, such as more exercise types, more nutritional syndrome: joint consensus statement of the European college supplements, more reasonable experimental designs, and of sport science and the American college of sports medicine,” more effective analysis methods. Medicine & Science in Sports & Exercise, vol. 45, no. 1, pp. 186–205, 2013. 4. Conclusion [2] L. P. Moreira, L. Silveira Jr., M. T. T. Pacheco, and D. D. F. M. Rocco, “Detecting urine metabolites related to In this study, the urine samples of basketball players were training performance in swimming athletes by means of collected from before- and after-training groups and were Raman spectroscopy and principal component analysis,” measured using NIRS technology, coupled with the newly Journal of Photochemistry and Photobiology B: Biology, proposed temperature-dependent approach in the temperature vol. 185, pp. 223–234, 2018. ° ° ° range of 20 C to 50 C with step of 5 C to collect the NIRS data. [3] W. K. Han, S. S. Waikar, A. Johnson et al., “Urinary bio- To distinguish the exercise state of athletes, the classic linear markers in the early diagnosis of acute kidney injury,” Kidney classification method PLS-DA was established based on the International, vol. 73, no. 7, pp. 863–869, 2008. [4] S. D. Smith, M. A. Wheeler, J. Plescia, J. W. Colberg, processed variables that were preprocessed by CWT, VIP, and R. M. Weiss, and D. C. Altieri, “Urine detection of survivin their combinations. Comprehensively, comparing perfor- and diagnosis of bladder cancer,” JAMA, vol. 285, no. 3, mances of all PLS-DA models, CWT-PLS-DA has the best pp. 324–328, 2001. performance whose average precision, sensitivity, and specificity [5] T. Guinovart, D. Hernandez-Alonso, ´ L. Adriaenssens et al., in prediction set are 0.98, 0.97, and 0.97, respectively. )e result “Characterization of a new ionophore-based ion-selective indicates that temperature-dependent NIRS is a potential electrode for the potentiometric determination of creatinine technique to accurately assess the exercise status of athletes and in urine,” Biosensors and Bioelectronics, vol. 87, pp. 587–592, will help optimize the amount of training and nutritional supplements for athletes in the further. [6] S. Sununta, P. Rattanarat, O. Chailapakul, and N. Praphairaksit, “Microfluidic paper-based analytical devices for determination of creatinine in urine samples,” Analytical Data Availability Sciences, vol. 34, no. 1, pp. 109–113, 2018. [7] M. M. Erenas, I. Ortiz-Gomez, ´ I. de Orbe-Payá et al., “Ion- All the data supporting the current findings reported in this ophore-based optical sensor for urine creatinine determina- manuscript are available from the corresponding author tion,” ACS Sensors, vol. 4, no. 2, pp. 421–426, 2019. upon request. [8] H. Jurdakov ´ a, ´ R. Gorov ´ a, ´ G. Addova, ´ A. Salingova, ´ and I. Ostrovsky, ´ “FIA-MS/MS determination of creatinine in Conflicts of Interest urine samples undergoing butylation,” Analytical Biochem- istry, vol. 549, pp. 113–118, 2018. )e authors declare that they have no conflicts of interest. [9] C. Pasquini, “Near infrared spectroscopy: a mature analytical technique with new perspectives—a review,” Analytica Chi- Acknowledgments mica Acta, vol. 1026, pp. 8–36, 2018. [10] L. Wang, D.-W. Sun, H. Pu, and J.-H. Cheng, “Quality )is work was supported by the National Natural Science analysis, classification, and authentication of liquid foods by Foundation of China (61705168) and Wenzhou Municipal near-infrared spectroscopy: a review of recent research de- Science and Technology Bureau (G20190024). velopments,” Critical Reviews in Food Science and Nutrition, vol. 57, no. 7, pp. 1524–1538, 2017. [11] B. Grassi and V. Quaresima, “Near-infrared spectroscopy and References skeletal muscle oxidative functionin vivoin health and disease: [1] R. Meeusen, M. Duclos, C. Foster et al., “European college of a review from an exercise physiology perspective,” Journal of sport science; American college of sports medicine. Biomedical Optics, vol. 21, no. 9, Article ID 091313, 2016. VIP VIP Journal of Analytical Methods in Chemistry 7 [12] L.-M. Yuan, F. Mao, X. Chen, L. Li, and G. Huang, “Non- [28] X. Chen, H. Ding, L.-M. Yuan, J.-R. Cai, X. Chen, and Y. Lin, invasive measurements of “Yunhe” pears by vis-NIRS tech- “New approach of simultaneous, multi-perspective imaging for quantitative assessment of the compactness of grape nology coupled with deviation fusion modeling approach,” bunches,” Australian Journal of Grape and Wine Research, Postharvest Biology and Technology, vol. 160, p. 111067, 2020. vol. 24, no. 4, pp. 413–420, 2018. [13] M. Zareef, F. Y. H. Kutsanedzie, A. A. Agyekum et al., “An [29] M. Ringner, ´ “What is principal component analysis?” Nature overview on the applications of typical non-linear algorithms Biotechnology, vol. 26, no. 3, pp. 303-304, 2008. coupled with nir spectroscopy in food analysis,” Food Engi- [30] X. Cui, X. Liu, X. Yu, W. Cai, and X. Shao, “Water can be a neering Reviews, vol. 12, no. 2, pp. 173–190, 2020. probe for sensing glucose in aqueous solutions by temperature [14] R. Sheng, W. Cheng, H. Li, S. Ali, A. Akomeah Agyekum, and dependent near infrared spectra,” Analytica Chimica Acta, Q. Chen, “Model development for soluble solids and lycopene vol. 957, pp. 47–54, 2017. contents of cherry tomato at different temperatures using [31] M. Nie, L. Meng, X. Chen et al., “Tuning parameter identi- near-infrared spectroscopy,” Postharvest Biology and Tech- fication for variable selection algorithm using the sum of nology, vol. 156, p. 110952, 2019. ranking differences algorithm,” Journal of Chemometrics, [15] H. Kim and D. G. Allen, “Using digital filters to obtain ac- vol. 33, no. 4, p. e3113, 2019. curate trended urine glucose levels from toilet-deployable [32] X. Chen, Y. Xu, L. Meng et al., “Non-parametric partial least near-infrared spectrometers,” Journal of Analytical & Bio- squares-discriminant analysis model based on sum of ranking analytical Techniques, vol. 7, no. 5, pp. 5–8, 2016. difference algorithm for tea grade identification using elec- [16] N. Yamamoto, N. Kawashima, T. Kitazaki et al., “Ultrasonic tronic tongue data,” Sensors and Actuators B: Chemical, standing wave preparation of a liquid cell for glucose mea- vol. 311, Article ID 127924, 2020. surements in urine by midinfrared spectroscopy and potential [33] L.-M. Yuan, J.-R. Cai, L. Sun, E. Han, and T. Ernest, application to smart toilets,” Journal of Biomedical Optics, “Nondestructive measurement of soluble solids content in vol. 23, no. 5, Article ID 050503, 2018. apples by a portable fruit analyzer,” Food Analytical Methods, [17] R. Raghavachari, “Biomedical applications of near-infrared vol. 9, no. 3, pp. 785–794, 2016. spectroscopy,” in Near-Infrared Applications in Biotechnology, pp. 327–351, CRC Press, Boca Raton, FL, USA, 2019. [18] K. Liu, X. Chen, L. Li, H. Chen, X. Ruan, and W. Liu, “A consensus successive projections algorithm—multiple linear regression method for analyzing near infrared spectra,” Analytica Chimica Acta, vol. 858, pp. 16–23, 2015. [19] Y. Sun, X. Cui, W. Cai, and X. Shao, “Understanding the complexity of the structures in alcohol solutions by tem- perature-dependent near-infrared spectroscopy,” Spectrochi- mica Acta Part A: Molecular and Biomolecular Spectroscopy, vol. 229, Article ID 117864, 2020. [20] X.-W. Liu, X.-Y. Cui, X.-M. Yu, W.-S. Cai, and X.-G. Shao, “Understanding the thermal stability of human serum pro- teins with the related near-infrared spectral variables selected by Monte Carlo-uninformative variable elimination,” Chinese Chemical Letters, vol. 28, no. 7, pp. 1447–1452, 2017. [21] X. Cui, X. Yu, W. Cai, and X. Shao, “Water as a probe for serum-based diagnosis by temperature-dependent near-in- frared spectroscopy,” Talanta, vol. 204, pp. 359–366, 2019. [22] Z. Xiaobo, Z. Jiewen, M. J. W. Povey, M. Holmes, and M. Hanpin, “Variables selection methods in near-infrared spectroscopy,” Analytica Chimica Acta, vol. 667, no. 1-2, pp. 14–32, 2010. [23] S. Wold, K. Esbensen, and P. Geladi, “Principal component analysis,” Chemometrics and Intelligent Laboratory Systems, vol. 2, no. 1-3, pp. 37–52, 1987. [24] D. Ballabio and V. Consonni, “Classification tools in chem- istry. Part 1: linear models. PLS-DA,” Analytical Methods, vol. 5, no. 16, pp. 3790–3798, 2013. [25] M. Farres, ´ S. Platikanov, S. Tsakovski, and R. Tauler, “Comparison of the variable importance in projection (VIP) and of the selectivity ratio (SR) methods for variable selection and interpretation,” Journal of Chemometrics, vol. 29, no. 10, pp. 528–536, 2015. [26] L. R. Watkins, “Review of fringe pattern phase recovery using the 1-D and 2-D continuous wavelet transforms,” Optics and Lasers in Engineering, vol. 50, no. 8, pp. 1015–1022, 2012. [27] P. Ye, G. Ji, L.-M. Yuan et al., “A sparse classification based on a linear regression method for spectral recognition,” Applied Sciences, vol. 9, no. 10, pp. 2053–2066, 2019. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of Analytical Methods in Chemistry Hindawi Publishing Corporation

Rapid Assessment of Exercise State through Athlete’s Urine Using Temperature-Dependent NIRS Technology

Download PDF
7 pages

Loading next page...
 
/lp/hindawi-publishing-corporation/rapid-assessment-of-exercise-state-through-athlete-s-urine-using-PpGa88B65u

References (35)

Publisher
Hindawi Publishing Corporation
Copyright
Copyright © 2020 Lihe Ding et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
ISSN
2090-8865
eISSN
2090-8873
DOI
10.1155/2020/8828213
Publisher site
See Article on Publisher Site

Abstract

Hindawi Journal of Analytical Methods in Chemistry Volume 2020, Article ID 8828213, 7 pages https://doi.org/10.1155/2020/8828213 Research Article Rapid Assessment of Exercise State through Athlete’s Urine Using Temperature-Dependent NIRS Technology 1 2 2 2 1 1 Lihe Ding, Lei-ming Yuan , Yiye Sun, Xia Zhang, Jianpeng Li, and Zou Yan School of Physical Education & Sport Science, Wenzhou Medical University, Wenzhou 325035, China College of Electric & Electronic Engineering, Wenzhou University, Wenzhou 325035, China Correspondence should be addressed to Lei-ming Yuan; yuan@wzu.edu.cn Received 9 April 2020; Accepted 18 August 2020; Published 29 August 2020 Academic Editor: Alessandro Buccolieri Copyright © 2020 Lihe Ding et al. )is is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Athletes usually take nutritional supplements and perform the specialized training to improve the performance of sport. A quick assessment of their athletic status will help to understand the current physical function of athletes’ status and the effect of nutritional supplementation. Human urine, as one of the most important body indicators, is composed of many metabolites, which can provide effective monitoring information for physical conditions. In this study, temperature-dependent near-infrared spectroscopy (NIRS) technology was used to collect the spectra of athlete’s urine for evaluating the feasibility of rapidly detecting the exercise state of the basketball player. To obtain the detection results accurately, several chemometrics methods including principal component analysis (PCA), variables selection method of variable importance in projection (VIP), continuous 1D wavelet transform (CWT), and partial least square-discriminant analysis (PLS-DA) were employed to develop a classifier to distinguish the physical status of athletes. )e optimal classifying results were obtained by wavelet-PLS-DA classifier, whose average precision, sensitivity, and specificity are all above 0.95, and the overall accuracy of all samples is 0.97. )ese results demonstrate that temperature-dependent NIRS can be used to rapidly assess the physical function of athlete’s status and the effect of nutritional supplementation is feasible. It can be believed that temperature- dependent NIR spectroscopy will obtain applications more widely in the future. human body is one of the most important outputs composed 1. Introduction of many metabolites. It has been widely used for assessment To achieve a high level of athletic performance, hard of human health and disease diagnosis, and it will help to trainings and reasonable dietary nutritional supplements quickly understand the body’s physical function and nu- are necessary. However, up to now, there is no rapid tritional diet [3, 4]. Creatine, as an auxiliary nutritional monitoring method to assess the exercise state of athlete supplement, will be metabolized and decomposed after and the effect of nutritional supplementation. Conse- consumption and eventually be eliminated from the body by quently, it usually leads to the over or under training for urination [5]. )erefore, urine can be used as an analytical athlete and affects the performance and physical health of test to assess whether the athlete takes the sport’s supple- athlete [1]. )erefore, it is necessary to develop a new ment creatine. Traditional methods of detecting the sup- technology to accurately assess the exercise state and plement creatine usually have the disadvantages of complex physical function of athletes. )is will provide scientific sample preparation, being time-consuming, and the need to training guidance and scientific information of nutritional use toxic and harmful reagents [6–8], which cannot meet the supplements for athletes. needs of rapid and accurate detection. )erefore, it is Body metabolites contain a variety of primary and necessary to develop a new technology to rapid detect the secondary metabolites, which can reflect athlete’s physical status of basketball player before and after taking creatine. function and state and provide the most intuitive infor- Near infrared spectroscopy (NIRS) is known as a fast mation for human health and exercise states [2]. Urine of and nondestructive analysis technology in the wavelength 2 Journal of Analytical Methods in Chemistry range of 780–2500 nm. NIRS has been widely applied in 2. Materials and Methods food, agriculture, biology, and chemistry fields [9–14]. At 2.1. Sample Collection and Preparation. 15 male basketball present, many researchers have used NIRS to qualita- players at the age range of 18 to 23 (weight range of 75± 5 kg) tively and quantitatively analyze the metabolites of ath- from Wenzhou Medical University were convened for this letes in sports medicine fields, such as quantitative experimental trials. Participants were interviewed to obtain determination of glucose content in urine [15, 16], rapid body information including drug intake, nutritional sup- detection, and analysis of trace protein in urine [17]. plements, past medical diseases, and anthropometric data. However, the lack of the spectral information greatly All participants agreed to participate in the study and signed limits the application broad of NIRS [18]. To solve this the informed consent. problem, temperature-dependent near-infrared spec- Before the urine collection, all athletes are required to troscopy technology was proposed [19], which added the write down the dietary information on the day of urine information of temperature into spectral dataset. )us, collection and to drink 100 ml of water 3 hours before more useful spectral information about temperature urine collection and underwent 48 hours without exercise. influence on analytes can be used to construct the In this experiment, all athletes will have their urine col- multivariate model. As a new type of spectroscopy lected twice before and after the specialized exercise. 15 ml technology, a few applications have been applied and urine sample was first collected before training. All ath- good analytical results were also obtained [20, 21]. letes will take 1.5 g creatine before training and then train Compared with traditional NIRS technology, tempera- for 120 minutes, and the other 15 ml urine sample will be ture-dependent NIRS technology can effectively increase collected after 5 minutes of rest. When the collection of the amount of spectral information and help improve the urine was finished, all urine samples were bottled into accuracy and stability of calibration model. 15 ml centrifuge tubes and immediately refrigerated at Spectral data, especially temperature-dependent near- −20 C for future use. infrared (NIR) spectra data, is a kind of high-dimensional data, containing mass spectral information. )erefore, development of multivariate calibration model is usually 2.2. Collection Temperature-Dependent NIRS Data. )e required for dimensional reduction, denoising operation, collection of temperature-dependent NIRS data from 4000 variables selection, and vector projection [22]. )e −1 to 12000 cm was performed on a Vertex 70 spectrometer commonly used method for dimensional reduction is (Bruker Optics Inc., Ettlingen, Germany). )e temperature principal component analysis (PCA) method [23], which control equipment used in this study is the 2216e temper- can be effective in extracting several principal components ature controller (Bruker Optics Inc., Ettlingen, Germany), from high-dimensional spectra data and reducing the which can provide a precision temperature (±0.1 C). In this dimension of spectra data. In this paper, temperature- study, the temperature range of this experiment is from 20 C dependent NIRS with a temperature dimension that has ° ° to 50 C with a step of 5 C, and the urine will be kept in the one more dimension than that of traditional spectral condition of 7 temperature points orderly from low to high. dataset. Hence, suitable chemometrics method is neces- To increase the ratio of signal to noise and reduce the sary to process these high-dimensional spectral data. In random errors, three spectra of each urine samples with scan addition to the PCA, the classic partial least squares number 64 were collected at each temperature. Finally, the discriminant analysis (PLS-DA) model was also consid- NIR matrix 210 × with 2074 columns and 210 rows were ered to discriminate the pre- and postexcise group [24]. obtained for analysis. Additionally, another two analysis methods of variable importance in projection (VIP) [25] and continuous 2.3. Multivariate Analysis Methods and Model Evaluation wavelet transform (CWT) [26, 27] were also introduced to further improve the classifying performance of the cali- 2.3.1. Principal Component Analysis (PCA). Principal bration model. component analysis (PCA) is one of the most used methods )e main research objective of this study is to verify in chemometrics. It is usually implemented to reduce the the feasibility of applying temperature-controlled NIRS dimensionality of dataset and provide the score plot for technology to quickly discriminant analysis of pre- and visualizing the distribution of samples [28]. As a commonly postexercise states of basketball players after eating cre- used exploratory method, PCA can make the primary atine. )e specific goals are as follows: (1) collecting NIR evaluation of similarity between samples’ classes. )e main data of urine samples at a series of temperature condi- principal of PCA is to convert a set of correlated variables tions; (2) making the preliminary spectral exploration into several linearly independent principal components with varying temperatures and sample’s visualization of (PCs) using orthogonal transformation [29]. As a result, the spectral data using PCA algorithm; (3) constructing the dimensionality of dataset is greatly reduced through PCA multivariate calibration classifiers between NIRS dataset analysis. and exercise state using PLS-DA algorithm; (4) identifying the optimal variables and enhancing the resolution of spectra using VIP and CWT algorithms; (5) comparing 2.3.2. Classification Algorithm. )e classical linear classi- performances of all classifiers and identifying the best fication method of PLS-DA algorithm was applied in this detection classifier. study. )e main principle of PLS-DA is to extract several Journal of Analytical Methods in Chemistry 3 latent variables (LVs), which are the linear combination intensity, it can be found that the spectral intensity of the original variables from the independent variables in generally shows a downward trend with increasing the format of matrix X. )en the relationship between temperature except for the spectra of 30 C. Besides, with independent variables X and dependent variables Y is increasing temperature of urine sample, the spectral in- −1 established and the prediction value of each sample is tensity at 11250 cm is bizarre and decreases with big ° ° ° obtained by this developed relationship. To achieve jump at low temperature (20 C, 25 C, and 30 C) but small sample classification, class of each sample is determined slope at high. )e possible reason is that the substances in ° ° according to the threshold value of class which is cal- urine from 25 C to 30 C have been changed. On the other culated and identified by Bayesian statistics [24]. Finally, hand, although the spectral intensity decreases with the the class of all samples is identified using PLS-DA temperature, it is difficult to fit using linear functions. classifier. Based on the above analysis, it can be found that it is hard to directly distinguish the samples before and after training only by observing the intensity of spectral curve. 2.3.3. CWT and VIP Methods. To enhance the resolution of )erefore, multivariate calibration model will be applied spectral, remove the noise/uninformative variables, and in the subsequent analysis to further analyze the spectral improve the performance of calibration model, continuous dataset. 1D wavelet transform (CWT) was applied in this paper [30]. During the implement, “Sym2” wavelet filter and the scale parameter 20 were used to process the NIRS data. As known 3.2. PCA of Temperature-Dependent NIRS Data. Prior to the that there are many redundant and uninformative variables calibrating analysis, it is recommended to explore the involved in the spectral matrix that will lead to the poor structure of spectral dataset. In this study, the effective performance of the calibration model. To solve this issue, statistical method of PCA was used to explore and vi- variable selection was commonly used to identify the op- sualize the space distribution of samples by extracting timal variables prior to establishing the multivariate model. several new principal components from high-dimen- In this study, variable importance in projection (VIP) was sional dataset. PCA was firstly performed on the tem- considered to select the optimal variables. During the perature-dependent NIRS dataset to plot the score scatter processing stage of VIP, V value that reflects the contri- of samples and observe the sample’s distribution. bution of each variable was calculated based on the PLS Figure 2(a) shows the PCA score plots of 210 urine regression to identify the optimal variable [31]. Finally, those spectral samples collected from before and after training. optimal variables with V value larger than 1 were eventually It is clear that PC1 and PC2 account for 92.94% and 6.48% selected. contributions to the original spectra, respectively, and the total cumulative variance contributes close to 99.42%. However, the samples from before and after training were 2.3.4. Model’s Evaluation. In this study, four evaluation still overlapped in these two-dimensional space, which parameters, namely, sensitivity, specificity, precision, and indicated that the urine samples from before and after accuracy, are used to evaluate the performance of the training were difficult to separate based on first two PCs. classification model. )ey can accurately and objectively To further explore the distribution of samples, 210 evaluate the performance of PLS-DA classification model. A ° ° samples coming from different temperature of 20 C–50 C classification model with good performance should have the were considered. high value of sensitivity, specificity, precision, and accuracy [24, 32]. All algorithms and calculations are carried out in Taking a close observation on Figure 2(b), it shows the the MATLAB 2015b environment ()e Math Works, Natick, distribution of samples with some linear relationship at USA). the same temperature. In addition, with rising of the temperature of urine sample, samples are distributed 3. Results and Discussion from the lower right to the upper left except for 30 C. Moreover, there is no sample overlapped among different 3.1. Spectral Profile Analysis. In this study, temperature- temperatures. )ese indicate that the temperature will dependent NIRS data of basketball players’ urine were significantly affect the NIRS response signals of urine collected in temperature ranges from 20 to 50 degrees samples. But PCA is an unsupervised method and cannot Celsius in steps of 5. )e corresponding spectral profiles accurately distinguish the categories of urine samples. are shown in Figure 1, where the average spectral curves )e possible reason is that the change of substances of the seven temperatures are collected before and after composition in urine before and after exercise is small, the basketball players’ exercise. It can be found that there −1 and the unsupervised principal component analysis are several obvious absorption peaks at 7800 cm , −1 −1 method cannot fully exhibit this change only using 9300 cm , and 10800 cm , which may be caused by several PCs. As a result, PCA failed to accurately dis- function groups of C-H, H-O, N-H, and C-O band in tinguish these samples that belong to two classes. To solve water, glucose, and protein from the urine [2]. Obviously, this issue, more information needed to be mined from there is no significant difference in spectral profiles be- NIRS data, and the classic PLS-DA classification model tween before and after exercise of basketball player by will be performed to construct the classification model. eye-naked judgement. On the consideration of spectral 4 Journal of Analytical Methods in Chemistry 1.0 1.2 1.0 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0.0 4500 6000 7500 9000 10500 12000 4500 6000 7500 9000 10500 12000 –1 –1 Wavenumber (cm ) Wavenumber (cm ) 20°C 40°C 20°C 40°C 25°C 45°C 25°C 45°C 30°C 50°C 30°C 50°C 35°C 35°C (a) (b) Figure 1: Average spectrum of basketball player’s urine with different temperature measurements. (a) Before exercise; (b) after exercise. 2 2 1.5 1.5 1 1 0.5 0.5 0 0 –0.5 –0.5 –1 –1 –1.5 –1.5 –2 –2 –2.5 –2.5 –6 –4 –2 0 26 4 –6 –4 –2 0 24 6 PC1 (92.94%) PC1 (92.94%) Before 20°C 40°C After 25°C 45°C 30°C 50°C 35°C (a) (b) Figure 2: Two-dimensional PCA and 2D-PCA analysis maps for the two classes of urine samples. (a) PCA plots of 210 urine spectral samples at 20 C. (b) PCA plots of urine samples at different temperature. 3.3. PLS-DA Model Based on Full and CWT Variables. First of all, PLS-DA classification model was established Based on the above analysis, the unsupervised PCA method based on the full variables of raw spectra to distinguish the urine cannot directly distinguish urine samples from the class of sample. In Table 1, it can be seen that the values of sensitivity, before and after training. )erefore the multivariate mod- specificity, and precision in calibration set are all 100%. eling method of PLS-DA was used to create the classification However, the result of validation and prediction set is lower than that in the calibration set. For urine samples from the model. Prior to establishing the PLS-DA model, 210 samples were randomly divided into the calibration set and pre- before-training group, the precision, sensitivity, and specificity diction set with the ratio of 2 :1. )en the classification in prediction set are 0.78, 0.82, and 0.71, respectively; mean- model was established based on the calibration set, and the while, the after-training group is 0.76, 0.71, and 0.82, and the number of latent variables (LVs) involved in PLS-DA model overall classification accuracy of all samples is 0.77. )is result was optimized using 10-fold cross-validation and was de- indicates that it is possible to rapidly detect the athletic state of termined at the lowest root mean square error of cross- athlete using NIRS technology. To further improve the per- validation (RMSECV). )e specifically calculated results are formance of classification model, continuous 1D wavelet shown in Table 1. transform was proposed to transform the temperature- PC2 (6.48%) Intensity PC2 (6.48%) Intensity Journal of Analytical Methods in Chemistry 5 Table 1: Classification results of PLS-DA model based on raw variables, optimal variables selected from raw spectra, and optimal variables from CWT spectra. Calibration Validation Prediction Model Methods LVs Class 2 3 4 Pre Sen Spe Pre Sen Spe Pre Sen Spe 1 1.00 1.00 1.00 0.71 0.83 0.69 0.78 0.82 0.71 — 7 2 1.00 1.00 1.00 0.82 0.69 0.83 0.76 0.71 0.82 1 1.00 1.00 1.00 0.88 0.91 0.89 0.88 0.90 0.83 VIP 5 2 1.00 1.00 1.00 0.92 0.89 0.91 0.86 0.83 0.90 PLS-DA 1 1.00 1.00 1.00 0.99 1.00 0.99 0.95 1.00 0.94 Wavelet 8 2 1.00 1.00 1.00 1.00 0.99 1.00 1.00 0.94 1.00 1 1.00 1.00 1.00 0.86 0.86 0.82 0.66 0.81 0.75 Wavelet-VIP 9 2 1.00 1.00 1.00 0.82 0.82 0.86 0.87 0.75 0.81 Notes: 1: latent variables; 2: precision; 3: sensitivity; 4: specificity. dependent NIRS dataset, and this would help improve the related to urine of basketball player were used to construct the PLS-DA classification model and the corresponding resolution of spectra. )en the PLS-DA classification model was built as in the previous step, and its performance was compared results were shown in Table 1. It can be seen that the with the full-variables-based PLS-DA model. In Table 1, it can performance of VIP-PLS-DA model in terms of the model be found that the precision, sensitivity, and specificity of class 1 precision (0.78 versus 0.88 and 0.76 versus 0.86), sensi- (labelled “before training”) and class 2 (labelled “after training”) tivity (0.82 versus 0.90 and 0.71 versus 0.83), and speci- in prediction set are 0.95, 1.00, and 0.94 and 1.00, 0.94, and 1.00, ficity (0.71 versus 0.83 and 0.82 versus 0.90) has been and the overall accuracy reached 0.97. Obviously, the perfor- improved when compared with FULL-PLS-DA model. )e possible reason for this improvement is that the mance of PLS-DA classification model has been significantly improved compared to that of full-variables-based PLS-DA unrelated variables may be deleted by VIP algorithm and the model becomes more stable and accurate. However, model. )e result shows that the continuous 1D wavelet transform (CWT) is an effective way to enhance the resolution the opposite result was obtained when the wavelet-VIP- PLS-DA model was considered compared with FULL- and further improve the accuracy of model. PLS-DA model. Specifically, there are no significant differences between 3.4. PLS-DA Model Based on Optimal Variables. Although a performances of these PLS-DA models in class 2, but the high classification accuracy has been obtained by the CWT- performance for class 1 is worse than FULL-PLS-DA model. PLS-DA model, the calculating process of these models is It demonstrates that although the continuous 1D wavelet can complex due to too many variables involved in the calcu- improve the performance of PLS-DA, the subsequent var- lation model. Moreover, it is known that there are many iable selection may not be suitable when the spectral data irrelevant, collinear, and redundant variables in spectral data was transformed by continuous 1D wavelet. )erefore, only which will lead to poor performance of classification model. CWTpretreatment is the better way to analyze NIRS data. In )erefore, it is necessary to hunt suitable variable selection addition, it can be found that there are many noise variables algorithms to identify a few important variables before −1 −1 in the range of 4000–5400 cm and 6500–7200 cm in establishing PLS-DA classification model [33]. Employ- Figure 3(a), which are contrary to the characteristic of ments of variable selections not only help reduce the di- continuous spectrum. To eliminate the effect of those noise mensionality of spectral data and the complexity of variables, we also performed the calculation only using the computation but also improve the accuracy and robustness −1 variables from the range of 11600–12000 cm and of the model. For this reason, classic variable selection al- −1 −1 5400 cm –6500 cm , and the overall accuracy is about gorithm VIP was applied on raw spectra data and the 0.85, which is consistent with the VIP-PLS-DA model. )is spectral data processed by continuous 1D wavelet transform means those unstable variables in the range of algorithm. −1 −1 4000–5400 cm and 6500–7200 cm do not work or even Results of variables selection through VIP are shown negatively affect the PLS-DA classification model. in Figure 3. According to the criterion of VIP algorithm, When all classification models, including FULL-PLS- those variables will be selected as the optimal variables DA, VIP-PLS-DA, wavelet-PLS-DA, and wavelet-VIP-PLS- whose VIP is larger than 1. )erefore, variables above the DA, are considered and compared, the best classification bold black line in Figure 3 are selected as optimal vari- model is wavelet-PLS-DA, whose the overall accuracy rea- ables. Specifically, 604 and 483 feature variables were ches 0.97, and the better one is VIP-PLS-DA model. )e identified from raw spectrum and wavelet-transformed worst one is the wavelet-VIP-PLS-DA model with accuracy spectrum. Compared with 2074 full variables, the number of 0.77. )ese results demonstrate that it is feasible to use of variables was decreased by 70.87% and 76.71%, and the multivariate calibration model and NIRS data to determine dimension of the spectra had been greatly decreased. the exercise status of athlete. In this primary study, there is When the variables selection was completed, those feature still a lot of work that needs to be further improved and variables that contained the most useful information 6 Journal of Analytical Methods in Chemistry 3.5 3.5 2.5 2.5 1.5 1.5 0.5 0 0.5 4000 6000 8000 10000 12000 4000 6000 8000 10000 12000 –1 –1 Wavenumber (cm ) Wavenumber (cm ) (a) (b) Figure 3: Optimal variables selected by VIP from raw spectrum (a) and wavelet transformed spectrum (b). Prevention, diagnosis, and treatment of the overtraining supplemented, such as more exercise types, more nutritional syndrome: joint consensus statement of the European college supplements, more reasonable experimental designs, and of sport science and the American college of sports medicine,” more effective analysis methods. Medicine & Science in Sports & Exercise, vol. 45, no. 1, pp. 186–205, 2013. 4. Conclusion [2] L. P. Moreira, L. Silveira Jr., M. T. T. Pacheco, and D. D. F. M. Rocco, “Detecting urine metabolites related to In this study, the urine samples of basketball players were training performance in swimming athletes by means of collected from before- and after-training groups and were Raman spectroscopy and principal component analysis,” measured using NIRS technology, coupled with the newly Journal of Photochemistry and Photobiology B: Biology, proposed temperature-dependent approach in the temperature vol. 185, pp. 223–234, 2018. ° ° ° range of 20 C to 50 C with step of 5 C to collect the NIRS data. [3] W. K. Han, S. S. Waikar, A. Johnson et al., “Urinary bio- To distinguish the exercise state of athletes, the classic linear markers in the early diagnosis of acute kidney injury,” Kidney classification method PLS-DA was established based on the International, vol. 73, no. 7, pp. 863–869, 2008. [4] S. D. Smith, M. A. Wheeler, J. Plescia, J. W. Colberg, processed variables that were preprocessed by CWT, VIP, and R. M. Weiss, and D. C. Altieri, “Urine detection of survivin their combinations. Comprehensively, comparing perfor- and diagnosis of bladder cancer,” JAMA, vol. 285, no. 3, mances of all PLS-DA models, CWT-PLS-DA has the best pp. 324–328, 2001. performance whose average precision, sensitivity, and specificity [5] T. Guinovart, D. Hernandez-Alonso, ´ L. Adriaenssens et al., in prediction set are 0.98, 0.97, and 0.97, respectively. )e result “Characterization of a new ionophore-based ion-selective indicates that temperature-dependent NIRS is a potential electrode for the potentiometric determination of creatinine technique to accurately assess the exercise status of athletes and in urine,” Biosensors and Bioelectronics, vol. 87, pp. 587–592, will help optimize the amount of training and nutritional supplements for athletes in the further. [6] S. Sununta, P. Rattanarat, O. Chailapakul, and N. Praphairaksit, “Microfluidic paper-based analytical devices for determination of creatinine in urine samples,” Analytical Data Availability Sciences, vol. 34, no. 1, pp. 109–113, 2018. [7] M. M. Erenas, I. Ortiz-Gomez, ´ I. de Orbe-Payá et al., “Ion- All the data supporting the current findings reported in this ophore-based optical sensor for urine creatinine determina- manuscript are available from the corresponding author tion,” ACS Sensors, vol. 4, no. 2, pp. 421–426, 2019. upon request. [8] H. Jurdakov ´ a, ´ R. Gorov ´ a, ´ G. Addova, ´ A. Salingova, ´ and I. Ostrovsky, ´ “FIA-MS/MS determination of creatinine in Conflicts of Interest urine samples undergoing butylation,” Analytical Biochem- istry, vol. 549, pp. 113–118, 2018. )e authors declare that they have no conflicts of interest. [9] C. Pasquini, “Near infrared spectroscopy: a mature analytical technique with new perspectives—a review,” Analytica Chi- Acknowledgments mica Acta, vol. 1026, pp. 8–36, 2018. [10] L. Wang, D.-W. Sun, H. Pu, and J.-H. Cheng, “Quality )is work was supported by the National Natural Science analysis, classification, and authentication of liquid foods by Foundation of China (61705168) and Wenzhou Municipal near-infrared spectroscopy: a review of recent research de- Science and Technology Bureau (G20190024). velopments,” Critical Reviews in Food Science and Nutrition, vol. 57, no. 7, pp. 1524–1538, 2017. [11] B. Grassi and V. Quaresima, “Near-infrared spectroscopy and References skeletal muscle oxidative functionin vivoin health and disease: [1] R. Meeusen, M. Duclos, C. Foster et al., “European college of a review from an exercise physiology perspective,” Journal of sport science; American college of sports medicine. Biomedical Optics, vol. 21, no. 9, Article ID 091313, 2016. VIP VIP Journal of Analytical Methods in Chemistry 7 [12] L.-M. Yuan, F. Mao, X. Chen, L. Li, and G. Huang, “Non- [28] X. Chen, H. Ding, L.-M. Yuan, J.-R. Cai, X. Chen, and Y. Lin, invasive measurements of “Yunhe” pears by vis-NIRS tech- “New approach of simultaneous, multi-perspective imaging for quantitative assessment of the compactness of grape nology coupled with deviation fusion modeling approach,” bunches,” Australian Journal of Grape and Wine Research, Postharvest Biology and Technology, vol. 160, p. 111067, 2020. vol. 24, no. 4, pp. 413–420, 2018. [13] M. Zareef, F. Y. H. Kutsanedzie, A. A. Agyekum et al., “An [29] M. Ringner, ´ “What is principal component analysis?” Nature overview on the applications of typical non-linear algorithms Biotechnology, vol. 26, no. 3, pp. 303-304, 2008. coupled with nir spectroscopy in food analysis,” Food Engi- [30] X. Cui, X. Liu, X. Yu, W. Cai, and X. Shao, “Water can be a neering Reviews, vol. 12, no. 2, pp. 173–190, 2020. probe for sensing glucose in aqueous solutions by temperature [14] R. Sheng, W. Cheng, H. Li, S. Ali, A. Akomeah Agyekum, and dependent near infrared spectra,” Analytica Chimica Acta, Q. Chen, “Model development for soluble solids and lycopene vol. 957, pp. 47–54, 2017. contents of cherry tomato at different temperatures using [31] M. Nie, L. Meng, X. Chen et al., “Tuning parameter identi- near-infrared spectroscopy,” Postharvest Biology and Tech- fication for variable selection algorithm using the sum of nology, vol. 156, p. 110952, 2019. ranking differences algorithm,” Journal of Chemometrics, [15] H. Kim and D. G. Allen, “Using digital filters to obtain ac- vol. 33, no. 4, p. e3113, 2019. curate trended urine glucose levels from toilet-deployable [32] X. Chen, Y. Xu, L. Meng et al., “Non-parametric partial least near-infrared spectrometers,” Journal of Analytical & Bio- squares-discriminant analysis model based on sum of ranking analytical Techniques, vol. 7, no. 5, pp. 5–8, 2016. difference algorithm for tea grade identification using elec- [16] N. Yamamoto, N. Kawashima, T. Kitazaki et al., “Ultrasonic tronic tongue data,” Sensors and Actuators B: Chemical, standing wave preparation of a liquid cell for glucose mea- vol. 311, Article ID 127924, 2020. surements in urine by midinfrared spectroscopy and potential [33] L.-M. Yuan, J.-R. Cai, L. Sun, E. Han, and T. Ernest, application to smart toilets,” Journal of Biomedical Optics, “Nondestructive measurement of soluble solids content in vol. 23, no. 5, Article ID 050503, 2018. apples by a portable fruit analyzer,” Food Analytical Methods, [17] R. Raghavachari, “Biomedical applications of near-infrared vol. 9, no. 3, pp. 785–794, 2016. spectroscopy,” in Near-Infrared Applications in Biotechnology, pp. 327–351, CRC Press, Boca Raton, FL, USA, 2019. [18] K. Liu, X. Chen, L. Li, H. Chen, X. Ruan, and W. Liu, “A consensus successive projections algorithm—multiple linear regression method for analyzing near infrared spectra,” Analytica Chimica Acta, vol. 858, pp. 16–23, 2015. [19] Y. Sun, X. Cui, W. Cai, and X. Shao, “Understanding the complexity of the structures in alcohol solutions by tem- perature-dependent near-infrared spectroscopy,” Spectrochi- mica Acta Part A: Molecular and Biomolecular Spectroscopy, vol. 229, Article ID 117864, 2020. [20] X.-W. Liu, X.-Y. Cui, X.-M. Yu, W.-S. Cai, and X.-G. Shao, “Understanding the thermal stability of human serum pro- teins with the related near-infrared spectral variables selected by Monte Carlo-uninformative variable elimination,” Chinese Chemical Letters, vol. 28, no. 7, pp. 1447–1452, 2017. [21] X. Cui, X. Yu, W. Cai, and X. Shao, “Water as a probe for serum-based diagnosis by temperature-dependent near-in- frared spectroscopy,” Talanta, vol. 204, pp. 359–366, 2019. [22] Z. Xiaobo, Z. Jiewen, M. J. W. Povey, M. Holmes, and M. Hanpin, “Variables selection methods in near-infrared spectroscopy,” Analytica Chimica Acta, vol. 667, no. 1-2, pp. 14–32, 2010. [23] S. Wold, K. Esbensen, and P. Geladi, “Principal component analysis,” Chemometrics and Intelligent Laboratory Systems, vol. 2, no. 1-3, pp. 37–52, 1987. [24] D. Ballabio and V. Consonni, “Classification tools in chem- istry. Part 1: linear models. PLS-DA,” Analytical Methods, vol. 5, no. 16, pp. 3790–3798, 2013. [25] M. Farres, ´ S. Platikanov, S. Tsakovski, and R. Tauler, “Comparison of the variable importance in projection (VIP) and of the selectivity ratio (SR) methods for variable selection and interpretation,” Journal of Chemometrics, vol. 29, no. 10, pp. 528–536, 2015. [26] L. R. Watkins, “Review of fringe pattern phase recovery using the 1-D and 2-D continuous wavelet transforms,” Optics and Lasers in Engineering, vol. 50, no. 8, pp. 1015–1022, 2012. [27] P. Ye, G. Ji, L.-M. Yuan et al., “A sparse classification based on a linear regression method for spectral recognition,” Applied Sciences, vol. 9, no. 10, pp. 2053–2066, 2019.

Journal

Journal of Analytical Methods in Chemistry – Hindawi Publishing Corporation

Published: Aug 29, 2020

There are no references for this article.