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. The 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. These 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.
To achieve a high level of athletic performance, hard trainings and reasonable dietary nutritional supplements are necessary. However, up to now, there is no rapid monitoring method to assess the exercise state of athlete and the effect of nutritional supplementation. Consequently, it usually leads to the over or under training for athlete and affects the performance and physical health of athlete [
Body metabolites contain a variety of primary and secondary metabolites, which can reflect athlete’s physical function and state and provide the most intuitive information for human health and exercise states [
Near infrared spectroscopy (NIRS) is known as a fast and nondestructive analysis technology in the wavelength range of 780–2500 nm. NIRS has been widely applied in food, agriculture, biology, and chemistry fields [
Spectral data, especially temperature-dependent near-infrared (NIR) spectra data, is a kind of high-dimensional data, containing mass spectral information. Therefore, development of multivariate calibration model is usually required for dimensional reduction, denoising operation, variables selection, and vector projection [
The main research objective of this study is to verify the feasibility of applying temperature-controlled NIRS technology to quickly discriminant analysis of pre- and postexercise states of basketball players after eating creatine. The specific goals are as follows: (1) collecting NIR data of urine samples at a series of temperature conditions; (2) making the preliminary spectral exploration with varying temperatures and sample’s visualization of spectral data using PCA algorithm; (3) constructing the multivariate calibration classifiers between NIRS dataset 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 performances of all classifiers and identifying the best detection classifier.
15 male basketball players at the age range of 18 to 23 (weight range of 75 ± 5 kg) from Wenzhou Medical University were convened for this experimental trials. Participants were interviewed to obtain body information including drug intake, nutritional supplements, past medical diseases, and anthropometric data. All participants agreed to participate in the study and signed the informed consent.
Before the urine collection, all athletes are required to write down the dietary information on the day of urine collection and to drink 100 ml of water 3 hours before urine collection and underwent 48 hours without exercise. In this experiment, all athletes will have their urine collected twice before and after the specialized exercise. 15 ml urine sample was first collected before training. All athletes will take 1.5 g creatine before training and then train for 120 minutes, and the other 15 ml urine sample will be collected after 5 minutes of rest. When the collection of urine was finished, all urine samples were bottled into 15 ml centrifuge tubes and immediately refrigerated at −20°C for future use.
The collection of temperature-dependent NIRS data from 4000 to 12000 cm−1 was performed on a Vertex 70 spectrometer (Bruker Optics Inc., Ettlingen, Germany). The temperature control equipment used in this study is the 2216e temperature controller (Bruker Optics Inc., Ettlingen, Germany), which can provide a precision temperature (±0.1°C). In this study, the temperature range of this experiment is from 20°C to 50°C with a step of 5°C, and the urine will be kept in the condition of 7 temperature points orderly from low to high. To increase the ratio of signal to noise and reduce the random errors, three spectra of each urine samples with scan number 64 were collected at each temperature. Finally, the NIR matrix 210 × with 2074 columns and 210 rows were obtained for analysis.
Principal component analysis (PCA) is one of the most used methods in chemometrics. It is usually implemented to reduce the dimensionality of dataset and provide the score plot for visualizing the distribution of samples [
The classical linear classification method of PLS-DA algorithm was applied in this study. The main principle of PLS-DA is to extract several latent variables (LVs), which are the linear combination of the original variables from the independent variables in the format of matrix
To enhance the resolution of spectral, remove the noise/uninformative variables, and improve the performance of calibration model, continuous 1D wavelet transform (CWT) was applied in this paper [
In this study, four evaluation parameters, namely, sensitivity, specificity, precision, and accuracy, are used to evaluate the performance of the classification model. They can accurately and objectively evaluate the performance of PLS-DA classification model. A classification model with good performance should have the high value of sensitivity, specificity, precision, and accuracy [
In this study, temperature-dependent NIRS data of basketball players’ urine were collected in temperature ranges from 20 to 50 degrees Celsius in steps of 5. The corresponding spectral profiles are shown in Figure
Average spectrum of basketball player’s urine with different temperature measurements. (a) Before exercise; (b) after exercise.
Prior to the calibrating analysis, it is recommended to explore the structure of spectral dataset. In this study, the effective statistical method of PCA was used to explore and visualize the space distribution of samples by extracting several new principal components from high-dimensional dataset. PCA was firstly performed on the temperature-dependent NIRS dataset to plot the score scatter of samples and observe the sample’s distribution. Figure
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.
Taking a close observation on Figure
Based on the above analysis, the unsupervised PCA method cannot directly distinguish urine samples from the class of before and after training. Therefore the multivariate modeling method of PLS-DA was used to create the classification model. Prior to establishing the PLS-DA model, 210 samples were randomly divided into the calibration set and prediction set with the ratio of 2 : 1. Then the classification model was established based on the calibration set, and the number of latent variables (LVs) involved in PLS-DA model was optimized using 10-fold cross-validation and was determined at the lowest root mean square error of cross-validation (RMSECV). The specifically calculated results are shown in Table
Classification results of PLS-DA model based on raw variables, optimal variables selected from raw spectra, and optimal variables from CWT spectra.
Model | Methods | LVs1 | Class | Calibration | Validation | Prediction | ||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
Pre2 | Sen3 | Spe4 | Pre | Sen | Spe | Pre | Sen | Spe | ||||
PLS-DA | — | 7 | 1 | 1.00 | 1.00 | 1.00 | 0.71 | 0.83 | 0.69 | 0.78 | 0.82 | 0.71 |
2 | 1.00 | 1.00 | 1.00 | 0.82 | 0.69 | 0.83 | 0.76 | 0.71 | 0.82 | |||
VIP | 5 | 1 | 1.00 | 1.00 | 1.00 | 0.88 | 0.91 | 0.89 | 0.88 | 0.90 | 0.83 | |
2 | 1.00 | 1.00 | 1.00 | 0.92 | 0.89 | 0.91 | 0.86 | 0.83 | 0.90 | |||
Wavelet | 8 | 1 | 1.00 | 1.00 | 1.00 | 0.99 | 1.00 | 0.99 | 0.95 | 1.00 | 0.94 | |
2 | 1.00 | 1.00 | 1.00 | 1.00 | 0.99 | 1.00 | 1.00 | 0.94 | 1.00 | |||
Wavelet-VIP | 9 | 1 | 1.00 | 1.00 | 1.00 | 0.86 | 0.86 | 0.82 | 0.66 | 0.81 | 0.75 | |
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.
First of all, PLS-DA classification model was established based on the full variables of raw spectra to distinguish the urine sample. In Table
Although a high classification accuracy has been obtained by the CWT-PLS-DA model, the calculating process of these models is complex due to too many variables involved in the calculation model. Moreover, it is known that there are many irrelevant, collinear, and redundant variables in spectral data which will lead to poor performance of classification model. Therefore, it is necessary to hunt suitable variable selection algorithms to identify a few important variables before establishing PLS-DA classification model [
Results of variables selection through VIP are shown in Figure
Optimal variables selected by VIP from raw spectrum (a) and wavelet transformed spectrum (b).
Specifically, there are no significant differences between performances of these PLS-DA models in class 2, but the performance for class 1 is worse than FULL-PLS-DA model. It demonstrates that although the continuous 1D wavelet can improve the performance of PLS-DA, the subsequent variable selection may not be suitable when the spectral data was transformed by continuous 1D wavelet. Therefore, only CWT pretreatment is the better way to analyze NIRS data. In addition, it can be found that there are many noise variables in the range of 4000–5400 cm−1 and 6500–7200 cm−1 in Figure
When all classification models, including FULL-PLS-DA, VIP-PLS-DA, wavelet-PLS-DA, and wavelet-VIP-PLS-DA, are considered and compared, the best classification model is wavelet-PLS-DA, whose the overall accuracy reaches 0.97, and the better one is VIP-PLS-DA model. The worst one is the wavelet-VIP-PLS-DA model with accuracy of 0.77. These results demonstrate that it is feasible to use multivariate calibration model and NIRS data to determine the exercise status of athlete. In this primary study, there is still a lot of work that needs to be further improved and supplemented, such as more exercise types, more nutritional supplements, more reasonable experimental designs, and more effective analysis methods.
In this study, the urine samples of basketball players were collected from before- and after-training groups and were measured using NIRS technology, coupled with the newly proposed temperature-dependent approach in the temperature range of 20°C to 50°C with step of 5°C to collect the NIRS data. To distinguish the exercise state of athletes, the classic linear classification method PLS-DA was established based on the processed variables that were preprocessed by CWT, VIP, and their combinations. Comprehensively, comparing performances of all PLS-DA models, CWT-PLS-DA has the best performance whose average precision, sensitivity, and specificity in prediction set are 0.98, 0.97, and 0.97, respectively. The result indicates that temperature-dependent NIRS is a potential technique to accurately assess the exercise status of athletes and will help optimize the amount of training and nutritional supplements for athletes in the further.
All the data supporting the current findings reported in this manuscript are available from the corresponding author upon request.
The authors declare that they have no conflicts of interest.
This work was supported by the National Natural Science Foundation of China (61705168) and Wenzhou Municipal Science and Technology Bureau (G20190024).