Estimation of the age/maturity of beef and sheep using spatially resolved visible-near-infrared spectroscopy

Summary

​Due to the relationships between animal age and maturity, collagen structure and tenderness, "age" is a critical factor in determining the market suitability of carcases and eating quality within the Meat Standards Australia Australia (MSA) grading system. Consequently, demarcation occurs after 30 months with older carcases attracting discounts, yet considerable debate still remains on the best way to determine "age". Despite cattle processors assessing the changes of cartilage along the spine by 'ossification', as an additional tool to determine physiological age, estimates can vary with nutrition and health stress. Consequently, an inexpensive, accurate and standardised measure of age could be utilised by industry to facilitate better identification of the true chronological and physiological age of animals.

The main aim of this project was to examine whether visible-near-infrared (Vis-NIR) spectra of hide samples could be used to accurately estimate the age/maturity of slaughter animals, with a particular focus on beef. Another aim was to investigate the possibility of predicting meat quality with direct relevance to the estimation of shear force and soluble collagen. Due to the encouraging results obtained from the muscle spectra data, the potential of the instrument to predict marbling scores using Vis-NIR scans of muscle tissue was also undertaken.

This work used a device which provided spatially resolved spectroscopy (SRS) measurements instead of standard reflectance spectroscopy measurements. Spatially resolved spectroscopy consists of using a fibre-optic probe which delivers light through a fibre and collects reflected light. The reflected light is from different distances from the source fibre using detection fibres placed at different distances from the source fibre. Thus, it provides reflected signals which have traversed different paths and distances within the tissue sample. This set up was used since it provides the means to find the optimal source-detector distance in terms of calibration model performance. This information can then be used to simplify the design of the fibre probe and make the device cheaper for commercial use. The probe used in this study consisted of a source fibre which delivers the light to the sample surround by 5 concentric rings of detection fibres. These fibres collect light reflected from the sample at different radial distances from the source fibre.

Eighty Angus cattle, with accurate date of birth, were sampled from 3 producers at 2 abattoirs with animal background and slaughter data (including dentition and ossification score) were recorded. After the hides had been removed from the carcase, 4 different locations were scanned (neck, armpit, rib region and under hind leg) using the SRS Vis-NIR system with a wavelength range of 390– 1000 nm. At 24 hours post mortem (pm), the striploin and eye round muscles were collected for scanning and meat quality analysis.

Partial least squares regression was then conducted to determine the potential to predict age and meat quality traits (shear force and soluble collagen) using spectra collected from the hide and from the striploin and eye round muscle scanned at 2, 14 and 28 days pm. Internal cross validation was done using the leave-one-out method and spectra were pre-processed using an Automatic Whittaker Filter for baseline correction and mean centred. The Automatic Whittaker was chosen after comparison of model performance using a number of standard spectra pre-processing techniques using the initial set of data collected for building calibration models. Models were also built using Genetic Algorithm for wavelength selection and the performances compared with those built without wavelength selection.

Overall, data collected and analysed to date indicates that the prediction of chronological age is possible using spectra collected on the hide at the neck by ring 5 (R2CV = 0.75, RMSECV = 1.6 years). Although, not as accurate, the prediction using spectra collected at location C (rib region) ring 5 (R2CV = 0.69, RMSECV = 1.8 years) and location A (neck) ring 1 also yielded some predictive ability (R2CV = 0.70, RMSECV = 1.7 years). There was also some potential to predict age using the eye round (R2CV = 0.69, RMSECV = 1.78 years) by ring 5.

Partial least squares (PLS) regression analysis was conducted to determine the potential to predict tenderness of the eye round muscle at 2, 14 and 28 days post mortem using SRS Vis-NIR spectra. collected at 2, 14 and 28 days post mortem, respectively. Regression models were built to predict shear force values. The results suggest that there is potential for predicting shear force when spectra from the eye round are used (R2CV = 0.69, RMSECV = 5.7 N).  When spectra from the striploin muscle were used, the predictability was generally poor. Models built to predict shear force from hide scans indicated some predictive ability particularly when scans from location C (rib region) collected by ring 5 is used (R2CV = 0.52, RMSECV = 6.2 N).

As in the case of shear force prediction, scans of eye round muscles indicated the ability to predict soluble collagen content (R2CV = 0.73, RMSECV = 0.44) while scans of striploin muscles did not indicate any ability to predict soluble collagen content. The prediction of the shear force values appears to be an indirect prediction of soluble collagen, which may explain the differences in prediction of shear force between the striploin and eye round as connective tissue plays a major role in determining meat toughness in the eye round.

Analysis also indicated that scans of striploin muscles have the potential to predict MSA marble scores when scans of samples at 2 days pm were used (R2CV = 0.49, RMSECV = 85).

The prediction of physiological age through calibration models built using scans of hides to predict ossification scores was considered. Analysis suggests that the predictive ability of the scans is limited with ring 5 measurements at location A providing the best results (R2CV = 0.63 and RMSECV = 101).  This limitation is probably due to the lack of sensitivity of ossification scores for cattle over the age of 5-6 years. Similar results were obtained when muscle scans were used with eye round muscle scans performing slightly better (R2CV = 0.62 and RMSECV = 103) than the strip loin muscle scans (R2CV = 0.55 and RMSECV = 112).

While the results discussed in this report were based on models which incorporated wavelength selection by genetic algorithm, analyses were also carried out by building models without wavelength selection. In all cases, the models arising from application of wavelength selection led to lower root mean square error of cross validation. Given that the age distribution is skewed towards the age of 2-3 years with relatively lower numbers of samples across other ages, there is potential for model overfitting to occur when applying the genetic algorithm. This could lead to more optimistic estimates of predictive ability than warranted. Therefore, it is essential to collect additional data across the age span of 0 – 13 years with an approximate uniform distribution of ages. Such a dataset will allow a more robust calibration model to be built. Additionally, a sufficient number of samples which span the desired age range can be set aside to be used as an external validation (i.e. an independent test) set so that a reliable estimate of predictive ability of the models can be obtained.

Given that the commercial age for slaughter of beef in Australia is around 18-30 months old there was a delay in sampling the other age categories, mainly younger than one year old and older than 5-6 years old. Thus, data collection of vealers and cull cows is required to provide a wider distribution in age to create a more accurate and robust calibration model. It is crucial to have enough data for animals older than 5-6 years old as it has been established that eating quality differs within this age category while ossification is not entirely distinctive for animals in this age category (reaching maximum score of 590).

Analysis indicates that, in most cases, spectra from ring 5 lead to models that were better compared to measurements from other rings. This raises the possibility that, in the future it may be possible to simplify the probe design and thus reduce the cost of a commercial device for estimating age and meat quality traits in addition to possibly reducing measurement time which currently takes approximately 1 minute. Also, based on the analyses carried out, the number of locations from which data will be collected in the future can be restricted to locations A (neck) and C (rib region).