Agricultural University of Athens
Implementation of rapid methods of analysis and model development in quality assessment of raw and processed poultry meat
Abstract
dc:descriptionNon-invasive rapid methods have been introduced over the years in the assessment of food quality and they have been well established in the food industry in the context of technological evolution as consumers’ demands for high quality and safety foods constantly increases. In the present thesis, rapid spectroscopic and biomimetic sensors have been investigated for their potential to accurately assess quality in different poultry products (chicken breast and thigh fillets, chicken marinated souvlaki and chicken burger). Multispectral Imaging (MSI), Fourier Transform Infrared spectroscopy (FT-IR) and electronic nose (E-nose) were employed (individually and in combination) in tandem with multivariate data analysis for the assessment of the microbiological quality and the spoilage level in chicken samples, as well as in the determination of the “time from slaughter”. For this purpose, different batches of chicken samples were subjected to storage experiments including both isothermal and dynamic temperature conditions and analyzed microbiologically to determine the population dynamics of the indigenous microbiota. In parallel, spectroscopic data were acquired through MSI and FT-IR instrumental analysis, whereas the volatile fingerprint of samples during storage was recorded by means of an E-nose. Regression and classification (linear and nonlinear) models assessing poultry meat quality were developed and validated with data from independent experiments (different batch/season of slaughter, dynamic temperature conditions of storage or different analysts). Moreover, ensemble methods and data fusion were performed to the existing data in an attempt to enhance the predictive performance of the developed models. Furthermore, the safety of poultry meat with special focus on Campylobacter spp. presence and survival in stored marinated chicken at refrigeration temperatures was explored via predictive modeling and molecular analysis.In chapter 2, MSI analysis was implemented on an industrial scale in chicken products for the assessment of their quality. For this purpose, chicken breast fillets, thigh fillets, marinated souvlaki and burger were analyzed microbiologically for the enumeration of TVCs and Pseudomonas spp., while MSI spectral data were acquired at the same time points as for microbiological analysis. Partial Least Squares Regression (PLS-R) models were developed based on MSI data for the determination of the “time from slaughter” parameter for each product type. Results showed that PLS-R models could predict accurately the time from slaughter in all products with the chicken thigh model providing the lowest RMSE value (0.160), followed by the chicken burger model (RMSE= 0.285).In chapter 3, FT-IR and MSI spectroscopic methods were evaluated for their efficacy to assess spoilage on the surface of chicken breast fillets in tandem with multivariate data analysis. Briefly, stored samples of chicken breast fillets at isothermal conditions (0, 5, 10, 15 οC) were analyzed microbiologically for the enumeration of TVCs and Pseudomonas spp. and also by FT-IR and MSI sensors. Multivariate data analysis was performed via two software platforms (a commercial software and a publicly available developed platform) by applying several machine learning models for the estimation of TVCs and Pseudomonas spp. population of the surface of the samples. The performance of the obtained models was assessed by intra batch and independent batch testing. PLS-R models from the commercial software predicted TVCs with RMSE values of 1.359 and 1.029 log CFU/cm2 for MSI and FT-IR analysis, respectively. Moreover, RMSE values for Pseudomonas spp. model were 1.574 log CFU/cm2 for MSI data and 1.078 log CFU/cm2 for FT-IR data. From the implementation of the in-house sorfML platform, ANN models developed with MSI data provided the lowest RMSE values (0.717 log CFU/cm2) for intra-batch testing, while least-angle regression (lars) models developed with FT-IR data demonstrated RSME values of 0.904 and 0.851 log CFU/cm2 in intra-batch and independent batch testing, respectively.In chapter 4, FT-IR and MSI spectral data were employed in combination with machine learning classification models for the evaluation of spoilage in chicken breast fillets. In this context, chicken breast samples were subjected to storage experiments using eight isothermal (0, 5, 10, 15, 20, 25, 30, 35 oC) and two dynamic temperature profiles for up to 480 h. At pre-determined intervals, samples were analyzed microbiologically for the enumeration of TVCs, while in parallel MSI and FT-IR instrumental analysis was performed. In addition, sensory analysis was undertaken by 14- member untrained panel for the assessment of fresh and spoiled samples. Based on the outcome of sensory analysis (threshold of spoilage: TVCs = 6.2 log CFU/cm2), samples were divided in two quality classes, namely fresh and spoiled. Eight machine learning models (single-based and ensemble) were developed with MSI and FT-IR spectral data for the detection of spoilage, whereas their performance was validated by an independent data set from the two dynamic temperature profiles. MSI analysis and subspace ensemble provided the highest overall accuracy (64.8 %), while this combination demonstrated also acceptable values of specificity and sensitivity (69.7 %). On the contrary, FT-IR spectral data presented slightly better performance with Partial Least Squares-Discriminant Analysis (PLS-DA), as the samples were classified correctly with an overall accuracy of 67.6 %.In chapter 5, FT-IR and MSI rapid techniques were employed for the assessment of the microbiological quality in chicken thigh fillets via qualitative and quantitative machine learning models. For this purpose, chicken thigh fillets were stored at eight isothermal (0, 5, 10, 15, 20, 25, 30, 35 oC) and two dynamic temperature profiles and analyzed microbiologically for the determination of TVCs and Pseudomonas spp., whereas MSI and FT-IR spectral data were acquired at the same time points. Samples were also evaluated by a sensory panel which established a TVC spoilage threshold at 6.99 log CFU/cm2. PLS-R models were implemented for the estimation of TVCs and Pseudomonas spp. counts on chicken’s surface. Moreover, classification models (LDA, QDA, SVMs, QSVMs) were developed for the discrimination of samples in two quality classes (fresh vs. spoiled). PLS-R models coupled to MSI data predicted TVCs and Pseudomonas spp. counts satisfactorily, with RMSE values of 0.987 and 1.215 log CFU/cm2, respectively. SVM model developed with MSI data exhibited the highest performance with an overall accuracy of 94.4%, while in the case of FT-IR, acceptable classification was obtained with the QDA model (overall accuracy 71.4%).In chapter 6, FT-IR, MSI and E-nose have been explored individually and in combination via data fusion for their efficacy in the evaluation of quality in marinated chicken souvlaki. In brief, chicken marinated souvlaki samples were subjected to storage experiments at both isothermal and dynamic temperature conditions. During storage, microbiological analyses were performed for the determination of the population dynamics of TVCs and Pseudomonas spp. in parallel with FT-IR, MSI and E-nose analyses. PLS-R and SVM-R models were developed and validated for the estimation of TVCs on chicken marinated souvlaki. Furthermore, three classification models (LDA, LSVM and QSVM) were investigated for the classification of stored samples in 2 and 3 quality classes (fresh vs spoiled; fresh, semi-fresh and spoiled). The developed models were externally validated with data obtained by six different analysts and three different batches of marinated souvlaki. The PLS-R models developed on MSI and FT-IR/MSI spectral data provided the best predictions of TVCs, with RMSE values of 0.998 and 0.983 log CFU/g, respectively. Moreover, for SVM models developed on MSI and FT-IR/MSI data, the population of TVCs was efficiently predicted with RMSE being 0.973 and 0.999 log CFU/g, respectively. For the classification models with 3 quality classes, the overall accuracy was calculated below 60 % in all cases. On the contrary, for the 2-class models, FT-IR/MSI spectral data analyzed by CSVM model exhibited overall accuracy of 87.5 %, followed by MSI data analyzed by LSVM model providing overall accuracy of 80 %. Finally, middle level data fusion of FT-IR to MSI was proven as a promising alternative for the assessment of quality in this poultry product. In chapter 7, the survival of Campylobacter spp. was investigated after inoculation of six strains (four Campylobacter coli strains and two Campylobacter jejuni strains) in chicken marinated souvlaki. Moreover, the microbial growth of the indigenous microbiota of the inoculated and non-inoculated chicken marinated souvlaki was examined. Inoculated and non-inoculated chicken marinated souvlaki samples were stored at three different isothermal conditions (0, 5, and 10 oC) and a dynamic temperature profile. At predetermined intervals, inoculated and non-inoculated samples were microbiologically analyzed for the enumeration of TVCs, Pseudomonas spp., anaerobic bacteria and Campylobacter spp. A one-step modelling approach was employed for chicken marinated souvlaki (inoculated and non-inoculated) for the determination of the kinetic parameters of growth for TVCs and Pseudomonas spp. Model validation was performed with an independent dataset derived from a dynamic temperature profile storage experiment. Further on, survival models predicting Campylobacter spp. counts during low storage temperatures were developed and assessed. Molecular analysis via Random amplified polymorphic DNA PCR (RAPD-PCR) was conducted with isolates obtained from three time points during the experiments. The developed models for TVCs and Pseudomonas spp. in inoculated and non-inoculated samples exhibited RMSE values lower than 0.941 log CFU/g. Campylobacter spp. survived despite the barrier of the low storage temperature where a decline of 1.5 log CFU/g was observed. From the survival models, the highest accuracy was provided by the Weibull model at 5 oC with RSME values of 0.112 log CFU/g. Molecular results confirmed that both C. coli and C. jejuni strains could survive during low temperature storage experiments with the exception of 5 oC, where only C. coli could be retrieved.
Degree
thesis:*- Grantor dc:publisher
- Agricultural University of Athens
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Spyrelli, Evgenia
- Σπυρέλλη, Ευγενία
Subjects
dc:subject × 25- Προϊόντα πουλερικών
- Φασματοσκοπικές μέθοδοι
- Μικροβιολογική αλλοίωση
- Βιομιμητικοί αισθητήρες
- Πολυμεταβλητή ανάλυση δεδομένων
- Ασφάλεια πουλερικών
- Poultry products
- Spectroscopic methods
- Microbiological quality
- Biomimetic sensors
- Multivariate data analysis
- Data fusion
- Food safety
- Βιολογία
- Φυσικές Επιστήμες
- Μικροβιολογία
- Άλλες Επιστήμες Μηχανικού και Τεχνολογίες
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη και Τεχνολογία τροφίμων
- Biological Sciences
- Natural Sciences
- Microbiology
- Other Engineering and Technologies
- Engineering and Technology
- Food Science and Technology
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/51650