Background: Meat fraud generated a huge outrage amongst customers in 2013 in Europe
due to the horsemeat scandal. Portable and hand-held optical near-infrared (NIR, 4,00012,500
cm-1/800-2,500 nm) spectroscopy sensors are traded as promising fast, non-invasive and easy analytical
tools that might be applicable at any independent place of inspection. In order to embrace the on-going
trend towards instrumental miniaturization, it was the aim of the present feasibility study to evaluate the
application of Design of Experiment for frequently applied portable micro-electro-mechanical system
(MEMS) based spectrometer by comparing its performance to a bench-top Fourier-Transform polarization
near-infrared (FT-NIR) instrument.
Methods: 63 samples of different meat types (beef: 9, chicken: 10, mutton: 10, turkey: 10, pork: 10,
horse meat: 14) were measured in order to classify the meat-type using a portable micro-electromechanical
system (MEMS) based spectrometer and a bench-top Fourier-Transform polarization nearinfrared
(FT-NIR) instrument, in order to compare the performance of both systems. In a second step
different meat types were minced together in order to investigate the level of adulteration which can be
detected using MEMS and FT-NIR. Design of Experiment (DoE) was applied to enhance results.
Results: The accuracy of MEMS versus FT-NIR for identifying whole / minced pieces of chicken, pork,
turkey, beef and mutton meat (63 samples) against horse meat appeared to be 75.0-100.0% (MEMS)
vs. 62.5%-100.0% (FT-NIR) for whole pieces and 75.0-100.0% (MEMS and FT-NIR) for minced
meat. When mincing different types of meat together, a maximum of 4 and 1 factors were required for
establishing a PLS-R model using again the spectra recorded with MEMS and FT-NIR, respectively.
The resulting quality parameters for the MEMS device were: R2=0.06-0.62, Standard Error of Cross
Valdiation (SECV)= 17.33-32.91, Ratio of Performance to Deviation (RPD) =0,54-1,70 and for the FTNIR
system: R2=0.85-0.94, SECV=7.52-13.83%, RPD=2.2-5.7 (FT-NIR). The limit of detection was
found at 10% for the MEMS and at 1% for the FT-NIR device.
Conclusion: Meat classification can be performed using the bench-top FT-NIR as well as the hand-held
MEMS-NIR. Mincing the meat samples does not necessarily improve classification accuracy as information
about the surface structure is lost. NIRS prediction models for adulterations were established for
the bench-top system. Prediction models for the hand-held device are inconclusive and have to be improved
by a larger sample set and/or further progress in miniaturization technique. Low level adulteration
(<10%) may also be predictable with NIRS, but continuative research is necessary.