Purpose: The aim of this work was to: (a) develop advanced modelling methodologies based on either one or both multispectral imaging and FTIR spectroscopy, in order to assess minced meat spoilage in various packaging and storage conditions; and (b) validate the results using independent data.
Methods: For this study, fresh minced beef was purchased and divided in 75g-portions in four occasions. They were then placed on styrofoam trays, packaged aerobically and under modified atmosphere packaging (80% O2-20% CO2). In the first two occasions, samples were stored in isothermal conditions at 4oC and 10oC until spoilage was pronounced. Four samples were analysed beginning with multispectral image acquisition, followed by the microbiological analysis of total viable counts (TVC) and FTIR spectroscopy per temperature and packaging at appropriate time intervals. Prediction models were developed, such as Partial Least-Squares Regression, Artificial Neural Networks and Support Vector Machines. Models were validated using (a) a percentage of the original data and (b) external validation data consisting of two batches in isothermal (4°C) and dynamic storage conditions (at 4 and 10° iteratively). In total, approximately 340 TVC measurements were collected, along with multispectral images and FTIR spectra.
Results: Results varied depending on the model, instrument and validation scheme. Successful models yielded a Mean Square Error (MSE) close to 0.2 (log CFU/g)2, whereas external validation yielded a higher but acceptable MSE and in some cases failed.
Significance: In conclusion, while this study showed the applicability of rapid methods for the assessment of spoilage, it also proved the necessity of external validation for the determination of the best models, as it avoids overfitting data leading to overoptimistic results, and takes into account the variability found among different meat batches.