Meat freshness revealed by visible to near-infrared spectroscopy and principal component analysis

Increasing concerns about adulterated meat encouraged industry looking for new non-invasive methods for rapid accurate meat quality assessment. Main meat chromophores (myoglobin, oxy-myoglobin, fat, water, collagen) are characterized by close comparable absorption in visible to near-infrared (NIR) spectral region. Therefore, structural and compositional variations in meat may lead to relative differences in the absorption of light. Utilizing typical fiber-optic probes and integrating sphere, a degradation of pork samples freshness was observed at room temperature referring to the relative changes in absorbance of main meat chromophores. The application of principal component analysis (PCA) used for examination of measured absorbance spectra revealed more detailed sub-stages of freshness, which are not observed by the conventional analysis of the reflectance spectra. The results show a great potential of the combined application of optical-NIR spectroscopy with complementary use of PCA approach for assessing meat quality and monitoring relative absorbance alternation of oxymyoglobin and myoglobin in visible, and fat, water, collagen in NIR spectral ranges.

In current paper, we consider if the relative spectral changes of absorbance in visible and NIR parts of the spectrum, measured by typical fiber-optic probes or an integrating sphere, can be associated with the freshness stages of meat samples. As a complementary analysis, we applied principal component analysis (PCA) method  Figure 1. Optical density spectra of main chromophores of muscle tissues, including: (a) myoglobin, oxymyoglobin (400-700 nm); (b) water, myoglobin, oxymyoglobin, collagen, fat (700-1100 nm); (c) water, collagen, fat (1100-2400 nm). Adapted from [45][46][47].
[48] on the absorbance dataset to find sub-stages of freshness decay which might be not revealed in spectroscopic analysis. In brief, the novelty of this work includes the type of sample, which is unprocessed, the storage condition during the measurements (at room temperature) and the short time duration of the measurements (on average 6 h) to investigate early changes in absorbance spectra of different chromophores. In another configuration, presented in figure 2(b), a standard portable spectrophotometer operated within the 400-1100 nm spectral range, is utilized. The spectrophotometer is equipped with a fiberoptic probe (for illumination and detection of light) Since the distance between the centers of the fibers is 530 μm, the distance between the 1st (illuminating) and 11th (collecting) fiber would be 5.3 mm while the minimal source-detector fiber separation is 0.53 mm (see the closeup view in figure 2(b)). This experimental setup comprises a light source Illuminator EK-1 Fiber Optic Light Source LE.5210-110 (EUROMEX, The Netherlands) with a halogen lamp and a compact CCS200 spectrometer (Thorlabs, USA), both connected to the fiber-optic probe.
For each configuration, porcine muscle meat samples were purchased from the local supermarkets on the first day after butchery. Three samples for visible and two samples for NIR spectra with integrating sphere configuration and thirteen samples with optical fibers configuration for each measurement were placed in a plastic Petri dish (5 cm in diameter, 1 cm high) with a rectangular hole to provide direct access of light to the sample. For all of the measurements, the samples were covered by plastic foil in order to delay moisture evaporation from the surface and prevent drying. Humidity in the laboratory room was controlled (at 80% level) and remained constant during all the measurements. Controlling relative humidity of the air during meat aging process needed to be controlled since high humidity will ease the spoilage bacteria growth and cause an unpleasant sticky surface while low humidity restricts bacterial growth but increases water evaporation causing dryness and less juiciness of meat. However, since in this work, we are measuring the early stages of meat loss of freshness for a small area of the sample and evaporation of water from the sample surface was suppressed by covering the meat surface with plastic films, the small changes of humidity cannot affect the results strongly [49].
By adjustment of the integration time and calibration, reflectance spectra (R) were obtained and converted to absorbance spectra [20]. The reflectance spectra were recorded every half an hour during on average six hours at room temperature (23°C). The Savitzky-Golay fitting algorithm was applied to remove random variations in the measured spectra. This technique clearly improves the visual appearance of the spectra [50]. Finally, the area between isosbestic points within the absorption bands responsible for the associated meat chromophores has been integrated and termed as 'integrated absorbance' and then, plotted their values over time. Eventually, we introduced a new term called 'degradation kinetics' for each of the meat chromophores, defined by the dependence of the 'integrated absorbance' over time to track their changes during freshness decay.

Monte Carlo simulations
Monte Carlo (MC) simulations are a well-established and effective approach to model light propagation in turbid media such as biological tissues [51] which can keep the track of photon transportation [52]. MC simulation consists of a sequential generation of trajectories of so-called photon packets from the source (the entrance to the medium) to the detector (the area where the photon leaves the medium [53]. Here, we used a free online simulation platform [54] implementing the MC method to estimate a sampling volume [53,55] and a probing depth in each measurement configuration. The optical parameters used in the simulations corresponded to muscle tissue at 632.8 nm [52,56] (for the illustrative purpose) are shown in table 3 [57]. Although the specific light distribution pattern depends on the wavelength, qualitatively the discussed difference between the configurations (integrating sphere and fiber-optic) will remain.
The integrating sphere configuration comprised a collimated light beam (size: 10 mm) normally incident on a rectangular meat sample (20×20×5 mm 3 ). Light reflected from the sample (from the surface and deeper regions) was collected from all directions within a 20 mm size area coincident with the incident beam. The fiberoptic configuration (see figure 2(b), inset) with 300 μm source and detecting fibers for two separation distances was also simulated. In this configuration, the sample size was either 2.5×2×2 mm 3 (source-detector distance: 0.53 mm) or 6×2×2 mm 3 (source-detector distance: 5.3 mm).

Principal component analysis (PCA)
To identify the most important directions of variability in a multivariate data matrix and to present the results in a graphical plot, multivariate statistical methods such as principal component analysis (PCA) can be applied [48]. Principal Components Analysis (PCA) is a data analysis tool which is mostly used to reduce the dimensionality (number of variables) of many interrelated variables, while retaining as much of the information (variation) as possible [58]. The calculated factors or pc's that are an uncorrelated set of variables are ordered in a way that the first few keep most of the variation present in all of the original variables.
There are a wide variety of PCA applications in different fields to classify large scattered datasets. Specifically, it has been an effective promising method utilized in meat quality assessment [58][59][60][61][62][63][64] such as beef characterization [48], Classification of Beef and Pork Aroma [59], classification of hairtail fish and pork freshness [60] or freshness assessment of cooked beef during storage [61]. Here, we performed PCA on the whole processed and smoothed absorbance dataset for each measurement time points for both configurations to detect and discriminate sub-stages of freshness levels correlated to chromophores changes over time which might be not recognizable in spectroscopic analysis.

Spectroscopic measurements
The absorbance spectra showed main peaks associated with different meat chromophores (oxymyoglobin, water, fat, and protein) in the pork samples and furthermore, the height decrease of those curves related to later times of measurement was easily observable. In addition, there was a noticeable decrease of the magnitudes of absorbance in both visible and NIR spectral regions caused by changes in the chemical composition in pork during freshness decay. Figure 3 shows absorbance spectra for the integrating sphere configuration over 6 h. The curves refer to data obtained 0 (solid), 3 (dash), and 6 (dot) hours after keeping the sample at room temperature. The local absorbance peaks in the visible range (see figure 3(a)) at around 540 nm and 575 nm wavelengths are attributed to oxymyoglobin content in the sample responsible for the meat color [40]. In the NIR region (see figures 3(b) and (c)) the main peaks in the absorbance spectra appear between 1100 nm and 1600. The peak around 1200 nm in figure 3(b) arises from the second overtone of C-H stretching vibrations associated mainly with fat in the samples [10]. The absorbance peak around 1450 nm (see figure 3(c)) is related to the first O-H overtone that arises from water and water-bonded groups [10]. These results indicate that water is the major domain component, which affects the mean spectrum of the pork samples. The local peak around 1525 nm (see figure 3(c)) is attributed to the N-H bond that arises from protein content [29].
Then, we integrated the area between isosbestic points under the absorbance spectra (termed 'integrated absorbance') within absorption bands responsible for the associated meat chromophores: oxymyoglobin (515-600 nm for the integrating sphere setup), fat (1175-1290 nm), water (1414-1490 nm), proteins (1490-1567 nm) and plotted them over time. Eventually, we introduced a new term called 'degradation kinetics' for each of the meat chromophores, defined by the dependence of the 'integrated absorbance' over time.
Decrease of the absorbance caused by water loss and degradation of oxymyoglobin affects negatively sample freshness. Changes of absorbance over time for the integrating sphere configuration are shown in the visible (see figure 3(d)) and NIR (see figures 3(e) and(f)) spectral ranges for the indicated specific wavelengths attributed to the meat components. It was observed that for the both visible and NIR spectral regions, integrated absorbance for different wavelengths experienced a decreasing trend showing meat chromophores degradation, which could affect pork freshness [5,10]. Specifically, degradation of oxymyoglobin indicating color changes started from the beginning (see figure 3(d)), while in the NIR region (see figures 3(e) and (f)), integrated absorbance decreased slower starting from approximately 2.5 h, that could be interpreted as a beginning stage of freshness deterioration process. As we can see in figure 3(e), the integrated absorbance for fat did not show a sharp reduction in contrast to the water and proteins curves. This is caused by fat degradation occurring at a slower rate than these other considered components.
Similar experiments were performed in the fiber-optic configuration. Figure 4(a) shows significant peaks associated with oxymyoglobin absorbance measured at three different time points (0, 3, and 6 h after keeping the sample at room temperature) and the integrated absorbance selected region within 527-587 nm.
The decreasing trend for the integrated absorbance over time was detectable in this case as well (see figure 4(b)), although the data points deviated more from the fitting curve and the drop in the absorbance happened after about 4.5 h. Comparison of the curves in figures 3(d) and 4(b) (both referred to oxymyoglobin changes over time) reveals the difference between the decreasing trends. Since the pork samples were covered with a plastic film from all sides and were under stable and similar physiological conditions, we assume that in both configurations freshness decay started at about the same time. The explanation of the observed discrepancy is elucidated further on.

Monte Carlo simulations
MC simulations were capable to elucidate the reasons of different degradation kinetics of oxymyoglobin and other chromophores. The difference between the two setups from optical point of view is the sensing depth: in the case of the integrating sphere setup, it was shallower due to higher contribution of the surface and subsurface reflected photons. In the second (fiber-optic) configuration, the probing depth was managed through changing the source-detector separation, i.e. by choosing the proper detecting fiber (since the illuminating fiber was kept the same). Results of the MC simulations (figure 5) illustrate this aspect. These results clarify the difference observed in figures 3(d) and 4(b). Despite detecting the same substance (oxymyoglobin), the indicated plots showed completely different trends: in the case of integrating sphere (see figure 3(d)), the degradation happened immediately from the beginning, while in the case of fiber-optic setup (see figure 4(b)) the degradation was significantly (4.5 h) delayed. Therefore, the delay was caused by larger depths achieved by detected photons in the latter case.

Principal component analysis (PCA)
Multivariate statistical analysis is frequently applied to spectral data due to its potential to deal with large complex co-linear information, reducing this original data to a lower dimension without overlooking useful information. Thus, PCA was applied to the processed and smoothed absorbance dataset obtained from the samples to correlate scattering alignments of data respect to each component axis with different freshness stages of meat. Figure 6 displays the corresponding scores plot of the raw pork samples and their spectra measured with two configurations and different spectral ranges (Vis/NIR) for the first and the second principal components.  PCA method provides complementary information and distinguishes the level of freshness between totally fresh sample in the beginning of the measurement (0 h; green) and less fresh sample (0.5 h-2 h; yellow cluster) in addition to another classification between 2.5 h-5 h (orange cluster) and 5.5 h-6 h (red cluster) for non-fresh sample.
For of the NIR spectra measured with the integrating sphere (see figure 6(b)), the first three principal components were responsible for 98.7% of variability of the data; the first, second and third principal components variability were 72.6%, 23.1% and 3%, respectively. Similar to visible spectra, PC 1 is the separator reference axis with negative values for fresh sample and positive value for non-fresh sample. According to the plot, the transition to non-fresh stage happens after 3.5 h which is the same as spectral analysis for fat absorption (see figure 3(e)), while for water and protein, spectral changes are detected earlier and after around 2.5 h (see figure 3(f)). The stages of freshness have been divided as the following:   figure 4(b)) with 47.9% and 16.8% of the total variance in the dataset which clearly caused by more scattered data compared to two previous spectra and therefore, leads to difficulty in distinction between freshness sub-stages.  Here again, classification between stages of freshness is defined by PC 1 with negative values correlated to fresh sample (green cluster) and positive values for less fresh and non-fresh samples (orange and red clusters). But similar to spectral analysis in figure 4(b), the beginning of decay happens after 4.5 h although the presence of data referred to 3.5 h and 4 h in red cluster is questionable.
Briefly, for all the measurements in both visible and NIR ranges, discrimination between fresh and non-fresh pork samples is clearly observed according to the changes in absorbance of different chromophores through using PC method applied to the absorbance spectra. The PCA results presented here are just based on the experimental data but as the next stage of the work, simulation will be added [8,61,[65][66][67].

Conclusion
The study presents a methodology to detect earlier stages of pork freshness loss at room temperature with the aim of decreasing the costs of meat quality monitoring. According to the obtained results, it is possible to observe the decreasing trend in the light absorbance for different pork chromophores in both visible and NIR spectral ranges showing loss of freshness over time. The compact fiber-optic linear array allows for retrieval of freshness decay depth simply by changing the detecting fiber (keeping the illuminating fiber the same). We believe this configuration can serve as a future base for development of a portable low-cost meat freshness sensor. The PCA method was used as a complementary analysis tool to classify the different stages of freshness and it succeeded to reveal sub-stages, which were not detectable by conventional reflectance spectroscopy. Further research with other types of meat of different age could help to build a comprehensive model of meat and its composition depending on freshness.