Miniaturized near-infrared instruments in dairy products or dairy industry: First steps in a long-distance race?

NIR and dairy products

Only a few portable NIR instruments among those on the market have been used in dairy industry and dairy products analysis: SCiO (Consumer Physics), MicroPhazir (originally Polycromix Inc. now Thermo Fisher), MicroNIR (Viavi Solutions), NeoSpectra-Micro (Si-Ware Systems) and NIRscan Nano Evaluation Module (Texas Instruments). ⁶ Some of them are designed mainly for solid samples or for the incorporation in on-line measurements and processes. The analysis of liquid samples requires therefore a thorough study of the specific solution to successfully carry out the measurement. NIR measurements go hand-in-hand with chemometric techniques when extracting useful information from the data. Optimization of both the spectral acquisition strategy and the chemometric method to be applied is thus required. However, it is worthwhile to consider that miniaturized NIR instruments are expected to be used in a high variety of situations due to their portability. The optimization of the whole analytical protocol could be even more tricky than in the case of benchtop instrumentation, as we have experienced in our work (our research groups work with SCiO, NeoSpectra-Micro and NeoSpectra-Scanner devices), both from the experimental and from the modelling points of view.

It is interesting to note that there are even already miniaturized NIR spectrometers such as SCiO having Android/iOS apps with a set of pre-established calibration models (in the cloud) for measuring nutritional values of dairy products such as hard and soft cheese, yogurts and puddings. With these tools, SCiO modelists claim that final users without any knowledge about spectroscopy or data processing can have in a few seconds and in their smartphones the results of the nutritional values (typically with no information concerning the associated errors) of e.g. the cheese they are buying at their local supermarket.

A few reported papers in the last five years deal about classification involving adulteration or authentication of dairy products. Due to the multivariate structure of NIR data, different multivariate classification techniques such as SIMCA (soft independent modelling of class analogies) or PLS-DA (partial least squares - discriminant analysis) have been used for the classification of dairy products. For instance, dos Santos Pereira et al. ⁷ studied the authentication of goat milk adulterated with cow milk, classifying correctly 100% of the pure goat milk samples with only one wrong classified sample in the test set. Karunathilaka et al. ⁸ studied the adulteration of milk powder using 11 potential adulterants, comparing a portable NIR instrument with two benchtop Fourier Transform-InfraRed (FT-IR) instruments. It is important to mention that the spectral range of the two benchtop FT-IR (12.500–4.000 cm⁻¹) in comparison was significantly larger than the spectral range of the portable NIR (6.266–4.167 cm⁻¹), making therefore the comparison biased. Not surprisingly, the classification models produced better specificity values and better detection capabilities for the FT-IR instruments than for the portable NIR, due to the lower spectral resolution and narrower spectral range, according to the authors. Liu et al. ⁹ successfully studied the authentication of milk, being able to distinguish organic milk from conventional milk, obtaining similar classification abilities than benchtop FT-NIR instruments.

NIR miniaturized instrumentation has also been used in the quantification of several compounds in dairy products. In these cases, multivariate regression techniques (being PLS, the most widely reported) are used to build the regression model between the NIR spectroscopic data and the compound of interest. Parameters such as root-mean square error of cross-validation (RMSECV) or root-mean square error of prediction (RMSEP) are widely used in assessing the quality of the multivariate models. Not surprisingly, milk has been the most studied dairy product. Several compounds such as protein, fat and solids-non-fat, ¹⁰ lactose, protein, fat and solids-non-fat, ¹¹ fat, ¹² lactose ¹³ and fatty acids ¹⁴ in milk have been predicted using NIR portable instrumentation and multivariate regression models. In most of the cases, the authors claim that good and reliable results are obtained, comparable to those obtained from lab-based instruments. ¹⁰ There has been also described the implementation of a new mobile application at the milking stage allowing to know in real-time the quality control parameters for each individual cow milk sample. ¹¹

NIR miniaturized instruments have also been applied to the determination of several compounds in cheese. Wiedemair et al. ¹⁵ analysed the content of water and fat content. The results showed a high correlation with the reference data, with a comparable performance of the portable NIR instrument regarding to a benchtop device. Ma et al. ¹⁶ analysed intact casein and total protein in cheddar cheese finding reliable results in the determination of these two parameters. Finally, Eskildsen et al. ¹⁷ obtained good estimates of fat and dry matter in on-line scanning measurements in blocks of Swiss cheese.

Is it worth to jump on the bandwagon?

Looking at the conclusions in the reviewed published papers, most of the authors conclude that the results are good for the proposed goals and comparable (in those cases where a comparison is made) to benchtop instruments or to reference methods, proving that miniaturized NIR instruments show potential in the dairy industry or in dairy products.

So far, the range of target compounds analysed in dairy products with NIR portable instruments is somehow limited and mainly restricted to classical macronutrients such as fat or proteins. The analysis of more compounds would be desirable in order to expand the range of applicability of this miniaturized instrumentation. Related to this, a thorough evaluation of the detection performances at low concentrations (including limits of detection and working ranges) would be desirable and would help the user to understand the cases in which these instruments can be applied. So far, the lowest reported concentrations using NIR miniaturized instrumentation in dairy products involve ranges with lower values of the order of 0.02 g/100 g for fatty acids in milk ¹⁴ using MicroPhazir but without a rigorous evaluation of the limit of detection. A rigorous evaluation of the limit of detection has only been reported in one article, ¹² being the limit of detection of 0.245 g/100 mL for the detection of fat in commercial milks.

From our experience, gained with the use of some of these instruments, signals are frequently very broad as an effect of the very small spectral range if compared with the benchtop instrumentation, which has reached outstanding performances in the last years. In some cases, signals are also quite noisy, what often requires many scans making the final analytical protocol not as fast and easy as it would be desirable. Moreover, we noticed problems in the temperature stability of some of the systems, suggesting that an instrumental optimization is still required to obtain robust devices. Another important issue is the type of sample to be measured, and in the specific case of milk is not so obvious to find a sample holder easy-to-use and easy-to-clean on the field and adequate to the spectroscopic analysis. In the case of not homogeneous samples, sample rotation or other strategies may be required to catch all the variability in the sample. Thus, in our opinion, the use of a portable NIR instrumentation is not so easy as it could initially seem, especially if quantitative values are of interest. Apart from these issues, which we do not perceive as problems but rather as stimuli in search of new analytical solutions, the emergence of portable NIR instruments may open a totally new perspective in the application of NIR techniques to dairy products. Easier on-site analysis or the possibility of carrying on-line measurements even in the smallest farms or facilities are goals that could be reached with this technology.