Calibration and standardisation of Dual Energy X-ray Absorptiometry to predict lamb carcass composition in Australian abattoirs.

Stephen L Connaughton

he Australian sheep meat industry has relied on long standing subjective measurement practices to predict the lean meat yield (LMY%) of lamb carcasses in abattoirs as a way of sorting product, as well as a source of feedback to the producers of those sheep. Dual energy X-ray absorptiometry (DXA) is spearheading a push for objective measures of LMY%, with the installation of prototype devices in multiple Australian sheep abattoirs. This thesis explores the installation of these novel devices, and the requirement for standardisation and calibration of the devices against the gold-standard measure of carcass composition by computed tomography (CT), and calibration against one another. The first experiment involved scanning 454 lambs of a large phenotypic and genotypic range through the DXA device at the flagship abattoir in South Australia, and then through a CT scanner for the gold standard carcass measurement. Excellent overall precision was recorded when predicting CT fat % (R2=0.91, RMSE=1.19%). Small biases present for sire breed, sire type, dam breed, hot carcass weight and c-site eye muscle area could be explained by a regression paradox, however biases amongst kill group (-0.73 to 1.01% for CT fat %, -1.48 to 0.76% for CT lean %) and the Merino sire type (0.36% for CT fat %, -0.73% for CT lean %) could not be explained by this effect. The second experiment tested the robustness of the DXA device predictions of CT composition of 30 lambs through changing abattoir and animal effects, including spray-chilling, repeatability of scanning over a 10-minute period, and over a 72-hour period. There was no prediction bias between the 15 spray-chilled and 15 non-spray-chilled carcasses. When repeat DXA scans were undertaken across a 10-minute period, there was a high level of repeatability for the prediction of CT fat %. When repeat scans were conducted at 6 time points across a 72-hour period the precision of the DXA prediction of CT fat % of 30 carcasses remained high (R2 = 0.94, RMSEP = 1.20%), although small biases existed between time points (P<0.01). The third experiment explored possible explanations for the findings of imprecision and inaccuracy within the first two experiments, testing for inaccuracies over the full height of the detector panels, and over the course of a production day. Tissue phantoms of varying fat and lean content (74%, 50%, 36%, 24% and 4% fat), and varying thicknesses (12, 40, 80 and 200mm) were randomly assigned across 9 possible positions between the DXA x-ray source and detector on a vertical shelf, which was repeated in 15 different variations. There was a significant difference in composition predictions at the third lowest position, most likely due to an increased intensity of x-ray photons from the perpendicular beam at this point, causing errors in the high energy detector. Mean x-ray intensity increases in both the high and low energy detectors throughout each production day, and this increase is different between days. There is a significant increase in x-ray intensity in the first 2-3 minutes after the restarting of the x-ray tube. This finding from experiment 3 informs the introduction of algorithm β2 using the unattenuated space values data in experiment 4, which scanned 30 lambs once per day over three days. There was a significant difference in lean % predictions across the three days (up to 6.02%) using the original algorithm β1, which was reduced to 1.01% when algorithm β2 is used. The repeatability experiment from experiment 2 was repeated, using 40 lambs in six scanning groups over three days, with each scanning group of 10 scanned five times within a 15-minute period, using algorithms β1 and β2. There was no significant improvement in accuracy between the five scanning runs between algorithm β1 and β2, while the overall repeatability of the DXA system remains high, with a coefficient of variation below 0.65%. Finally, the fifth experiment had 60 carcasses selected at two sites with on-line DXA systems installed, with 10 carcasses selected each day at each site over three successive days. Each carcass was scanned at its selection site, transported to the alternate site within 24 hours to be scanned there, and finally transported for CT scanning. Both sites had excellent precision when predicting CT fat % within each day (R2=0.85 to 0.96, RMSE=1.09% to 1.54%), however were inaccurate across sites by up to 34.4 CT fat % units. A plastic phantom used to calibrate the two DXA systems using R-value and thickness value adjustments in algorithm β3 corrected this inaccuracy to a maximum difference between sites of 1.58 CT fat % units while maintaining the same precision within days.

Calibration and standardisation of Dual Energy X-ray Absorptiometry to predict lamb carcass composition in Australian abattoirs.

Files and links (1)

Abstract

Details

UN Sustainable Development Goals (SDGs)

Metrics