The accuracy of dual energy X-ray absorptiometry (DXA), weight, and P2 back fat to predict half-carcass and primal-cut composition in pigs within and across research experiments

CSIRO PUBLISHING

www.publish.csiro.au/journals/ajar Australian Journal of Agricultural Research, 2004, 55, 973–982

The accuracy of dual energy X-ray absorptiometry (DXA), weight,and P2 back fat to predict half-carcass and primal-cut composition

in pigs within and across research experiments

D. SusterA,B, B. J. LeuryB, C. D. HofmeyrA, D. N. D’SouzaC, and F. R. DunsheaA,B,D

ADepartment of Primary Industries, 600 Sneydes Rd, Werribee, Vic. 3030, Australia.BThe University of Melbourne, Parkville, Vic. 3010, Australia.

CAnimal Research and Development Unit, Agriculture Western Australia, 3 Baron Hay Crt, South Perth,WA 6151, Australia.

DCorresponding author; email: [email protected]

Abstract. A Hologic QDR4500A dual energy X-ray absorptiometer (DXA) was used to measure body compositionin 199 half-carcasses ranging from 15 to 48 kg. Half-carcasses were from animals of mixed sex and of eitherLarge White × Landrace or Large White × Landrace × Duroc descent. Half-carcasses were selected from 5 differentexperiments to evaluate DXA accuracy within and across experiments. Values determined by DXA including totaltissue mass, fat tissue mass, lean tissue mass, and bone mineral content, for the half-carcass and the shoulder,loin, belly, and ham primal cuts were evaluated by comparison with manually dissected composition. Relationshipsbetween manually dissected values and measurements of weight and backfat at the P2 site were also evaluated.Manually dissected values were strongly related to DXA-derived values, more so than with weight and P2 or acombination of both, particularly in the measurement of fat composition. In contrast to estimates derived fromweight and P2, DXA-derived estimates remained accurate even when between-experiment variation was included.However, because DXA estimates were different from manually dissected values, they would need to be adjustedwith the use of appropriate regression equations to correct the in-built algorithms. These results demonstrate theefficacy of DXA as a non-destructive method for determining the composition of the half-carcass and primal cuts,and its greater precision than current routinely used methods.

Additional keywords: dissection, lean tissue, fat, belly, REML.

IntroductionThe Australian pig industry currently relies on a numberof carcass traits, such as carcass weight and P2 fat depth,as the basis for specifications to supply pork to meetcustomer requirements (Gardner 1990). However, a recentstudy by Suster et al. (2003) showed that P2 fat depth wasa poor predictor of total carcass fat and lean tissue whenmeasurements were compared across different experiments.The reason for this may be differences in body fat distributionamong different animals. For example, D’Souza et al. (2004)demonstrated a dramatic increase in fat content and the fat tolean tissue ratio in the belly, relative to other primal cuts, in thefinisher stages of growth. Fat distribution throughout the bodymay also be altered by treatment with growth modifiers suchas exogenous administration of porcine somatotropin (Suster2004). In addition, heavy selection against P2 fat depth inthe past may have re-distributed carcass fat to other regionsof the body such as the belly in contemporary genotypes(Channon et al. 1999). This may have ramifications in markets

where pork buyers are willing to pay premiums for lean porkbellies. Therefore, there is a need for a practical, accurate, andnon-invasive technique that can measure whole and regionalcarcass composition.

One reliable and convenient method for determiningboth total and regional fat, lean tissue, and bone mineralcomposition is dual energy X-ray absorptiometry (DXA), theprinciples of which have been discussed by Kelly et al. (1998)and Laskey and Phil (1996).

Previous studies evaluating DXA measurements in pigshave demonstrated high degrees of precision when comparedwith chemically determined and manually dissected values(Mitchell et al. 1998; Lukaski et al. 1999; Marcoux et al.2003; Suster et al. 2003). However, it is clear that the systemmanufacturer, instrument generation, and software versioncan all affect measurements (Tothill et al. 1994; Kistorp andSvendsen 1998). Therefore, it may be difficult to compareDXA values among instruments and this may be exacerbatedby other factors such as varying animal genotypes, sexes,

© CSIRO 2004 10.1071/AR04052 0004-9409/04/090973

974 Australian Journal of Agricultural Research D. Suster et al.

and tissue hydration. In addition, Mitchell et al. (1998)observed that using correction equations developed for thehalf-carcass, for the prediction of primal cut composition,was not appropriate. Therefore, it is necessary to constructseparate prediction equations for the various primal cuts. Theaim of this study was to determine whether DXA could beused to predict the dissectible lean meat, fat tissue, and bonein the half-carcass and primal-cuts of pigs and evaluate howthis technique compares with P2 back fat, the commercialmeasure of carcass quality used in many countries. Accuracyof DXA for comparing measurements within and acrossexperiments was also evaluated.

Materials and methodsAnimals and scanning

Scanning with DXA was performed using a Hologic QDR4500A FanBeam X-Ray Bone Densitometer (Hologic, Inc., Waltham, MA, USA).The whole body scan mode was used on all half-carcasses (softwareV8.26a:3). The DXA was used to determine body composition in199 half-carcasses from mixed sex animals of either Large White ×Landrace (117) or Large White × Landrace × Duroc (82) descent.Half-carcasses ranged from 15 to 48 kg. Half-carcasses were selectedfrom 5 different ongoing experiments at either the Department of

(a) (b)

Fig. 1. (a) Half-carcass scan image generated with dual energy X-ray absorptiometry analysedwith regional grid divisions for the shoulder, loin, belly, and ham sections. (b) Half-carcass brokeninto the commercial primal cuts.

Primary Industries in Werribee or from the Animal Research &Development Unit, Agriculture Western Australia. These experimentsinvolved a range of treatments that related to the growth of pigs,including altered dietary protein level, sex, and treatment with porcinesomatotropin.

Measurements made by DXA included total tissue mass (TTM), leantissue mass (LTM), fat tissue mass (FTM), and bone mineral content(BMC) (TTM = LTM + FTM + BMC). The Hologic QDR4500ADXA was equipped with a step phantom that is constructed of acrylicand aluminium. Acrylic attenuates X-ray beams in a manner similarto fatty tissue and the addition of aluminium produces materials thatappear leaner (Kelly et al. 1998). This principle is exploited by the stepphantom for unit calibration, ensuring maintenance of accurate lean/fatcomposition results. The step phantom was scanned on a weekly basis.A spine phantom was also supplied by Hologic Inc. and consisted ofa simulated lumbar spine, of a known bone mineral content, encasedin an acrylic block. The spine phantom was scanned daily to ensure anaccurate and repeatable BMC measurement. Half-carcasses (right side)were positioned flat on the DXA table with the cut surface down, toensure that carcass orientation was consistent between scans.

Carcass dissection

Post-scanning, half-carcasses were refrigerated for 24 h and then brokeninto the commercial primal cuts of shoulder, middle (loin + belly) andham (Fig. 1). The head was cut off at the atlas joint, weighed, anddiscarded. Hocks were removed at the knee joint. The shoulder was

Composition of pig carcasses measured by DXA Australian Journal of Agricultural Research 975

separated from the middle by cutting between the 4th and 5th rib, in astraight line through the junction of the 4th and 5th thoracic vertebra. Theham was severed from the middle by a straight cut at a right angle to thelongitudinal axis between the last and second-last lumbar vertebra, justcranial to the hip bone. The middle was split by removal of the belly witha cut parallel to the dorsal edge, measured 5 cm from the ventral edge ofthe eye muscle. Primal cuts were weighed at this stage prior to manualdissection. Dissection was carried out to a retail level into bone, fattissue, lean meat, and rind and each of these components was weighed.Hocks were included in the bone weight of their respective shoulder orham component. Flare fat was included in the belly fat weight. Rind wasadded to the dissected lean meat since it is a proteinaceous tissue. Fatdepth at the P2 site (6.5 cm from the mid-line over the last rib) was alsomeasured using a ruler.

DXA regional analysis

The DXA regional analysis software was used in determining bodycomposition results. For soft tissue measurements, the QDR4500Asoftware allows the scanned image to be divided into head, arms, legs,and trunk. In this experiment, the entire half-carcass was analysed byplacing the scanned image in the left arm region. The QDR4500Aregional analysis software was then used to measure the compositionof the ham, loin, belly, and shoulder regions. A scanned imageincluding these sections is shown in Fig. 1. Regional grid placement wasperformed as closely as possible to the broken carcass, also displayedin Fig. 1. The ham was placed in the leg region of the regional analysisgrid. The arm region was manipulated to measure the composition ofall other primal cuts. The arm region was utilised because it providedresults that were more repeatable than when other regions of the regionalanalysis grid were used (Suster 2004).

Statistical analyses

Restricted maximum likelihood (REML) analysis (Genstat 2000) wasused to develop models relating dissected values for lean meat, fat tissue,and whole bone to those derived by DXA (LTM, FTM, and BMC) inthe half-carcass and primal cuts. Models for dissected values were alsoconstructed for weight and P2 for comparison with the models includingDXA measurements. In each model, a random effect of experimentwas included, so that the effect of experiment-to-experiment variationcould be evaluated. To increase the homogeneity of residuals and toimprove interpretation, most of the response and independent variableswere logarithmically transformed (to base 10) before conducting theREML analysis. For dissected values, the best practical model includingavailable DXA measurements was chosen with the use of Wald tests(Genstat 2000, pp. 441–445). Models including weight and P2 werealso constructed for comparison with the models including DXAmeasurements. Sex effects were examined but were negligible and sohave not been reported.

For each model, an error standard deviation (ESD) was calculatedwithin experiments and when between-experiment variation (BEV)was included. The within-experiment ESD was calculated as the squareroot of the residual variance and is appropriate for making comparisonsinvolving animals in the same experiment. The ESD includingbetween-experiment variation was calculated as the square root ofthe sum of the residual variance and the between-experiment variancecomponent. It is appropriate for making comparisons involving animalsthat might not come from the same experiment. Significance of theBEV effect was examined using a change of deviance test (Genstat2000, pp. 448–451) between the model with or without the BEVeffect.

For each of these two cases (including BEV or not), estimated95% proportionate probability intervals (95% PPI) of the deviationsof observations about the model predicted values were calculated bymultiplying the ESD by ±1.96, then back-transforming and finally

expressing the result as a percentage of the predicted value. An intervalof (90%, 120%) can be interpreted as meaning that it is estimated that95% of the chemically determined values will be between 90% and120% of the predicted value of the model. An analogue of % varianceaccounted for by the model (denoted by R2) was also calculated for boththe within- and between-experiment case. For the within-experimentcase:

R2 = 1 − residual variance of the variate

unmodelled variance of the variate

For the case including between-experiment variation:

R2 = 1 − (residual variance of the variate +between-experiment variance)/

(unmodelled variance of the variate +unmodelled between-experiment variance)

In the formulae above, the unmodelled variances refer to theresidual variance estimated for a model that does not include anypredictive factors. Therefore the R2 values are calculated relative to amodel that involves measuring nothing at all. When there is evidenceof BEV, the ESD and the range of the 95% PPI will be larger for thegeneral situation including BEV than the within-experiment situation.This is not always the case with R2, which is not a complete measureof the relationship even when those relationships are linear (Moore1995, p. 115). However, the R2 values are included for comparison withother published literature on predicting the body composition of farmanimals.

Results

Half-carcass weight and composition

The DXA-derived TTM in the half-carcass was stronglyrelated to weight, and was not different from weight at anyhalf-carcass weight (Fig. 2). The DXA-derived LTM, FTM,and BMC were also strongly related to their correspondingdissected vales but were different from the line ofidentity (Fig. 2). The DXA-derived LTM was systematicallyoverestimated by the DXA algorithms, whereas FTM wassystematically overestimated in carcasses with low amountsof fat but underestimated in carcasses with higher amounts offat. DXA-derived BMC was grossly underestimated relativeto dissected bone.

Models and their precision for weight and dissected leanmeat, fat, and bone in the half-carcass, developed from DXAvalues, are shown in Table 1. Inclusion of BEV affectedthe precision (P < 0.01) of all these models, although theeffect was most pronounced on the models for fat and wasvery small in models for weight and lean meat. Dual energyX-ray absorptiometry was able to predict weight to within 5%error and lean meat content to within 12% error, irrespectiveof BEV, 95% of the time. Within experiments, DXA couldpredict FTM to within 22% error, but when BEV was includedthis increased to about 40% error. DXA could predict BMCto within 13–17% error, depending on the criteria (within orbetween experiment).

Models, and their precision, for predicting dissectedlean meat, fat, and bone, as developed from a combinationof half-carcass weight and P2 backfat, are shown in


(a) Weight (b) Lean

(c) Fat (d ) Bone

Log

diss

ecte

d fa

t (kg

)Lo

g sc

ale

wei

ght (

kg)

Log

diss

ecte

d bo

ne (

kg)

Log

diss

ecte

d le

an (

kg)

1.6

1.4

1.2

1.0

0.8

1.8

1.6

1.4

1.2

1.0

1.2

1.0

0.8

0.6

0.4

0.2

0.0

1.2

1.0

0.8

0.6

0.4

0.2

0.0

1.0 1.2 1.4 1.6 1.8 0.8 1.0 1.2 1.4 1.6Log DXA TTM (kg) Log DXA LTM (kg)

0.0 0.2 0.4 0.6 0.8 1.0 1.2 –0.4 –0.2 0.0 0.2 0.4Log DXA FTM (kg) Log DXA BMC (kg)

Fig. 2. Relationships between the log (base 10) of (a) dual energy X-ray absorptiometry (DXA)-derived total tissue mass(TTM) and scale weight, and DXA-derived and dissected values for (b) lean tissue mass (LTM), (c) fat tissue mass (FTM),and (d) bone mineral content (BMC). A least squares linear line (solid) and a line of identity (dashed) have also been included.

Table 2. Inclusion of BEV reduced the precision ofmodels in all cases (P < 0.001). The precision of themodel including weight and P2 for the prediction oflean meat (ESD = 0.020 kg) was at a similar level tothe model including DXA (ESD = 0.021 kg) withinan experiment but not when BEV was included(0.031 v. 0.024 kg for weight plus P2 or DXA alone,respectively). For prediction of dissected fat, the modeldeveloped from a combination of weight and P2 wasmoderately precise within experiments but was extremelypoor when BEV was included (Table 2). The 95% PPIindicates that predictions from weight and P2 can befrom one-half to twice the true dissected fat content.The DXA model for the prediction of dissected fat(ESD = 0.045 kg) was better than those developed fromweight and P2 (ESD = 0.075 kg) within experiments, andespecially when BEV was included (0.148 v. 0.076 kg forweight plus P2 or DXA alone, respectively).

Primal cuts

Models, and their precision, developed from DXA and acombination of weight and P2 backfat for prediction ofmeasured attributes in the shoulder, middle, loin, belly, andham, are shown in Table 1 and Table 2, respectively. Ingeneral, models for the prediction of primal-cut composition,

although precise, were not as precise as those developed forthe half-carcass. Differences in precision between modelsdeveloped from DXA and those developed from weightand P2 were similar in all primal cuts to those previouslydescribed for the half-carcass. Inclusion of BEV reduced theprecision of all models across all the primal cuts (P < 0.001).An exception was the model for dissected bone in theshoulder, developed from a combination of weight and P2,which was not affected by the inclusion of BEV (P > 0.10).

Primal-cut weight

Within experiments, DXA was able to predict weight ofthe ham to within 4% error, which was similar to theprediction error in the half-carcass (Table 1). Inclusion ofBEV increased error in the DXA model for ham weightby <1%. Within experiments, predictive error for weightmodels developed from DXA was about 5% in the shoulderand middle and about 20% in the loin and belly. IncludingBEV increased the error around DXA models for shoulderand middle weight by 2% and for loin and belly weight byabout 5%.

Primal-cut lean tissue

The DXA was able to predict dissected lean meat in the hamto within 7% error within experiments and this error only


Table 1. Precision of models for log dissected composition (kg) from log dual energy X-ray absorptiometry (DXA)-derived total tissuemass (TTM), lean tissue mass (LTM), fat tissue mass (FTM), and bone mineral content (BMC) (kg)

Best prediction model Comparisons within experiments Comparisons generally(values in parentheses are standard errors) ESDA 95% PPIB (R2)C ESDA 95% PPIB (R2)C

(%) (%)

WeightHalf-carcass −0.054(±0.011) + 1.031(±0.007)∗logTTM 0.007 (96.9,103.2) 0.989 0.008 (96.6,103.6) 0.992Shoulder −0.008(±0.016) + 0.995(±0.014)∗logTTM 0.013 (94.5,105.9) 0.963 0.017 (92.7,107.8) 0.963Middle −0.042(±0.010) + 1.034(±0.008)∗logTTM 0.009 (95.8,104.4) 0.987 0.015 (93.5,106.9) 0.981Loin 0.004(±0.026) + 0.983(±0.028)∗logTTM 0.033 (86.2,116.0) 0.873 0.044 (81.9,122.1) 0.877Belly 0.011(±0.031) + 0.966(±0.037)∗logTTM 0.043 (82.2,121.6) 0.739 0.065 (74.5,134.3) 0.714Ham −0.037(±0.010) + 1.031(±0.010)∗logTTM 0.008 (96.3,103.8) 0.983 0.009 (96.0,104.1) 0.986

LeanHalf-carcass −0.250(±0.062) + 0.995(±0.024)∗logLTM 0.021 (90.9,110.0) 0.894 0.024 (89.5,111.7) 0.934

+ 0.0021(±0.0006)∗LTM%Shoulder −0.055(±0.023) + 0.958(±0.022)∗logLTM 0.019 (91.6,109.2) 0.913 0.030 (87.4,114.4) 0.910Middle −0.030(±0.027) + 1.671(±0.113)∗logLTM 0.025 (89.1,112.2) 0.938 0.035 (85.4,117.1) 0.936

− 0.684(±0.106)∗logTTMLoin −0.023(±0.033) + 1.559(±0.203)∗logLTM 0.039 (83.9,119.0) 0.800 0.055 (77.8,128.5) 0.839

− 0.622(±0.195)∗logTTMBelly 0.014(±0.037) + 1.356(±0.177)∗logLTM 0.049 (80.0,125.0) 0.661 0.062 (75.7,132.1) 0.744

− 0.448(±0.163)∗logTTMHam −0.140(±0.018) + 1.067(±0.017)∗logLTM 0.014 (93.7,106.8) 0.951 0.018 (92.4,108.3) 0.956

FatHalf-carcass 0.208(±0.102) + 1.661(±0.055)∗logFTM 0.045 (81.6,122.5) 0.903 0.076 (70.8,141.2) 0.852

− 0.493(±0.084)∗logTTMShoulder −0.217(±0.043) + 1.417(±0.067)∗logFTM 0.077 (70.6,141.7) 0.694 0.121 (57.9,172.7) 0.627Middle 0.337(±0.100) + 1.498(±0.070)∗logFTM 0.065 (74.4,134.4) 0.853 0.085 (68.2,146.7) 0.856

− 0.436(±0.108)∗logTTMLoin 0.066(±0.037) + 1.149(±0.053)∗logFTM 0.095 (65.1,153.7) 0.716 0.124 (57.1,175.1) 0.716Belly 0.139(±0.119) + 1.390(±0.106)∗logFTM 0.095 (65.2,153.3) 0.752 0.117 (59.1,169.3) 0.801

− 0.239(±0.164)∗logTTMHam −0.173(0.030) + 1.343(0.044)∗logFTM 0.056 (77.5,129.0) 0.830 0.089 (67.0,149.2) 0.759

BoneHalf carcass 1.038(±0.029) + 0.856(±0.029)∗logBMC 0.026 (88.7,112.7) 0.838 0.034 (85.8,116.5) 0.806

− 0.077(±0.011)∗BMC%Shoulder −0.192(±0.094) + 0.260(±0.053)∗logBMC 0.035 (85.2,117.4) 0.729 0.038 (84.1,118.9) 0.829

+ 0.629(±0.065)∗logTTMMiddle 0.050(±0.079) + 0.335(±0.030)∗logBMC 0.038 (84.3,118.7) 0.793 0.051 (79.3,126.1) 0.668

+ 0.441(±0.050)∗logTTMLoin 0.116(±0.081) + 0.426(±0.041)∗logBMC 0.047 (80.9,123.7) 0.755 0.054 (78.5,127.4) 0.776

+ 0.398(±0.057)∗logTTMBelly −0.809(±0.111) + 0.004(±0.019)∗logBMC 0.081 (69.2,144.4) 0.465 0.118 (58.6,170.6) 0.268

+ 0.710(±0.095)∗logTTMHam −0.125(±0.106) + 0.372(±0.056)∗logBMC 0.034 (84.6,116.8) 0.771 0.055 (78.0,128.2) 0.527

+ 0.572(±0.072)∗logTTM

AError standard deviation.BEstimated 95% proportionate probability interval (%).CAnalogue of % variance accounted for.

increased by 1% when BEV was included. Surprisingly, inboth scenarios (BEV or not), this precision was better thanthat observed in the half-carcass. In models for shoulder andmiddle, DXA could predict dissected lean meat to withinabout 10% error within experiments and about 15% errorwhen BEV was included. Lean meat in the loin and bellywas predicted to within 20–25% error within experimentsand about 30% error when BEV was included. Precisionof models developed from a combination of weight and

P2 (Table 2), for the prediction of LTM, was similar toDXA models both within experiments and when BEV wasincluded. Surprisingly, the DXA model for dissected leanmeat in the loin was not as strong as the model includingweight and P2.

Primal-cut fat

The DXA could predict dissected fat in the ham and middleto 30% error within experiments and about 50% error when


Table 2. Precision of models for log dissected values (kg) from log weight (WT) (kg) and P2 backfat (mm)

Best prediction model Comparisons within experiments Comparisons generally(values in parentheses are standard errors) ESDA 95% PPIB (%) (R2)C ESDA 95% PPIB (%) (R2)C

LeanHalf-carcass −0.080(±0.037) + 0.991(±0.025)∗logWT 0.020 (91.5,109.3) 0.908 0.031 (87.0,114.9) 0.896

− 0.0056(±0.0006)∗P2Shoulder −0.125(±0.026) + 1.011(±0.025)∗logWT 0.021 (91.1,109.8) 0.902 0.028 (88.2,113.3) 0.922

− 0.0036(±0.0006)∗P2Middle −0.050(±0.033) + 0.977(±0.029)∗logWT 0.027 (88.4,113.1) 0.877 0.052 (79.1,126.5) 0.799

− 0.0089(±0.0008)∗P2Loin −0.074(±0.024) + 0.980(±0.024)∗logWT 0.026 (88.9,112.7) 0.906 0.045 (81.6,122.5) 0.894

− 0.0076(±0.0007)∗P2Belly 0.001(±0.034) + 0.887(±0.040)∗logWT 0.043 (82.2,121.6) 0.739 0.065 (74.5,134.3) 0.714

− 0.0085(±0.0012)∗P2Ham −0.153(±0.022) + 1.055(±0.022)∗logWT 0.017 (92.6,108.0) 0.932 0.024 (89.8,111.4) 0.920

− 0.0041(±0.0005)∗P2

FatHalf-carcass −1.193(±0.163) + 0.863(±0.094)∗logWT 0.075 (71.4,140.1) 0.734 0.148 (51.4,194.6) 0.449

+ 0.0573(±0.0123)∗P2 − 0.0010(±0.0004)∗P22

Shoulder −2.249(±0.610) + 3.136(±1.248)∗logWT 0.084 (68.3,146.4) 0.634 0.150 (50.8,196.7) 0.430− 1.109(±0.634)∗logWT2 + 0.0228(±0.0023)∗P2

Middle −1.027(±0.105) + 0.941(±0.093)∗logWT 0.090 (66.5,150.4) 0.720 0.157 (49.2,203.2) 0.508+ 0.0296(±0.0027)∗P2

Loin −1.111(±0.075) + 0.895(±0.081)∗logWT 0.089 (66.9,149.1) 0.754 0.137 (53.8,185.9) 0.653+ 0.0332(±0.0025)∗P2

Belly −1.105(±0.102) + 1.203(±0.096)∗logWT 0.104 (62.4,160.2) 0.698 0.205 (39.6,252.2) 0.386+ 0.0240(±0.0030)∗P2

Ham −0.985(±0.115) + 0.706(±0.109)∗logWT 0.082 (69.1,144.7) 0.642 0.134 (54.7,182.8) 0.452+ 0.0273(0.0023)∗P2

BoneHalf-carcass −0.607(±0.057) + 0.914(±0.039)∗logWT 0.033 (86.1,116.1) 0.746 0.044 (82.0,121.9) 0.675

− 0.0033(±0.0009)∗P2Shoulder −0.592(±0.042) + 0.936(±0.043)∗logWT 0.039 (84.0,119.0) 0.680 0.040 (83.5,119.7) 0.815

− 0.0039(±0.0010)∗P2Middle −0.696(±0.050) + 0.899(±0.049)∗logWT 0.047 (80.9,123.6) 0.683 0.066 (74.3,134.7) 0.452

− 0.0057(±0.0014)∗P2Loin −0.608(±0.040) + 0.857(±0.052)∗logWT 0.059 (76.6,130.3) 0.614 0.067 (74.0,135.2) 0.651

− 0.0044(±0.0016)∗P2Belly −0.887(±0.067) + 0.871(±0.076)∗logWT 0.081 (69.3,144.2) 0.469 0.128 (56.1,178.3) 0.142

− 0.0035(±0.0023)∗P2Ham −0.720(±0.054) + 0.960(±0.050)∗logWT 0.039 (83.9,119.2) 0.698 0.065 (74.6,134.0) 0.340

− 0.0014(±0.0011)∗P2

AError standard deviation.BEstimated 95% proportionate probability interval (%).CAnalogue of % variance accounted for.

BEV was included. Predictive error around the DXA modelfor shoulder fat was about 40% within experiments and 70%when BEV was included. The DXA models for predictingfat in the loin and belly were less precise than DXA modelsfor predicting fat in other primal cuts. However, the DXAmodels for predicting dissected fat were more precise thanmodels developed from a combination of weight and P2 for allprimal cuts, except for the loin (Table 2) within experiments,but not when BEV was included. In addition, when BEVwas included, precision of models for predicting dissected fatfrom a combination of weight and P2, was particularly poor

in all primal cuts relative to DXA and this was especiallypronounced in the belly (0.205 v. 0.117 kg for weight plus P2or DXA alone, respectively).

Primal-cut bone

The DXA could predict dissected bone in all primal cuts towithin about 20% error within experiments. An exceptionwas the belly, in which the error was about 40%, probablydue to the low amount of bone present in this primal cut.The shoulder and loin primal cuts were most accurate afterinclusion of BEV, as error only increased by 1% and 5%,


respectively. Models developed from DXA were more precisein predicting dissected bone than were those developed froma combination of weight and P2.

Discussion

Half-carcass composition

Chemical analysis is the ‘gold standard’ measurement in bodycomposition studies and was used as the point of comparisonwith DXA-derived measurements in a previous study (Susteret al. 2003). However, if DXA were to be considered for usein carcass grading, more appropriate models may need tobe developed for dissectible lean meat yield, fat content, andbone, as these values differ from their chemically determinedcounterparts. ‘Chemically determined’ lean tissue is anaddition of all protein and water in the carcass, whereasdissected lean meat excludes protein and water from fat tissueand bone. Chemically determined lipid excludes water andconnective tissue (protein), unlike the dissected fat value.Finally, chemically determined ash constitutes an addition ofosseous and non-osseous minerals, whereas dissected bonemass constitutes osseous mineral, water, and some proteinand fat.

As expected, DXA-derived TTM was not different fromweight, an observation also made previously (Suster et al.2003). Although TTM can be easily measured and itseems unnecessary to use DXA to predict this value, itis a fundamental indicator for composition predictions.Consequently, if LTM, FTM, and BMC are measuredincorrectly, it will affect the DXA prediction for TTM.The DXA-derived LTM was overestimated relative todissected lean meat, and this overestimation was greaterthan that observed when the comparison was performedwith chemically measured lean tissue (Suster et al. 2003).The greater degree of overestimation may result from waterin fat and bone not being included in the dissected leanmeat measurement, but being included in the chemicallean tissue measurement. The relationship between DXA-derived FTM and dissected fat was steeper in gradient thanthe corresponding relationship with chemically determinedvalues (Suster et al. 2003). In the present study, fatwas overestimated in animals with dissectible fat contentbelow about 8 kg and underestimated when fat was abovethis amount. Finally, the DXA-derived BMC was grosslyoverestimated due to comparison of a bone mineral valuewith one comprising whole ‘wet’ bone.

These discrepancies between DXA-derived and bothdissected and chemically determined values occur becauseof the underlying principles (Laskey and Phil 1996; Kellyet al. 1998) behind DXA function and, most significantly,because the DXA measurements have been calibrated tothe human model. Measurements made by DXA are basedon the differential attenuation of high and low energy

X-rays by bone, lean tissue, and lipid. Each of thesecomponents attenuates X-rays at different rates. However,X-ray attenuation coefficients for LTM are essentially derivedfrom body water and estimated, by the QDR4500A software,with the assumption that the lean body mass of the subject is73.2% water (Anon. 1996). In addition, although the FTMmeasurement primarily consists of lipid, it is not totallyanhydrous. To correct for this, the QDR4500A DXA softwareassumes FTM to be a combination of lipid (91.4%) and water(8.6%) (Anon. 1996). Importantly, the estimated water that isincorporated in the DXA-derived FTM value is not measuredfrom water within body fat but rather from the whole bodywater pool. Because tissue hydration level varies amongspecies of differing age (Manners and McCrea 1963), theseassumptions may not universally apply. Therefore, both leanand fat predictions will inevitably be influenced by changesin water content, an effect also observed in other studies(Elowsson et al. 1998; Pietrobelli et al. 1998; Proctor et al.1999). This will also explain in part why the discrepancybetween the DXA and standard values is not constant butchanges with animal age or liveweight. In addition, theincreasing tissue thickness with animal age may provide someexplanation. The degree of X-ray penetration of the body,and thus the subsequent signal for the DXA measurement,decreases with body thickness, reducing the efficiencyat which attenuation can be determined (Kelly et al. 1998;Nord 1998).

The DXA models for dissected measurements were mostprecise for LTM and least precise for FTM and BMC, anobservation that is consistent with other studies (Mitchellet al. 1996; Lukaski et al. 1999; Suster et al. 2003).However, DXA models for dissected bone were clearly moreprecise than those developed for fat, a contradictory findingto that made previously using chemically determined ashvalues (Suster et al. 2003) and to other similar studiescomparing DXA-derived half-carcass values and chemicalanalysis (Mitchell et al. 1998). This probably arises fromthe inequality between the chemical and dissected valuesalready discussed. In addition, comparisons with retaildissections may be beneficial from an industry perspectivebut there is some impreciseness associated with dissectionthat needs to be addressed. In comparative studies thatutilise chemical analysis, all lipid in the half-carcass isaccounted for. However, fat tissue dissected to a retaillevel is free of intra-muscular fat and some inter-muscularfat. Therefore, any fluctuations in the intra- and inter-muscular fat content will introduce error to the DXA modelfor dissected fat. This error will also be propagated tothe model for lean meat because the excess fat will beincluded in the dissected lean meat value. However, becauseof the greater proportionate lean meat mass relative to fat,the effect of this inaccuracy would be more pronounced inthe model for fat.


Effect of subregion incorporation

Overall, in the critical evaluation of DXA for measuringprimal cut composition, there is a need to discriminatebetween errors inherent in assumptions related to X-rayabsorption ratios to estimate soft tissue composition and theerrors associated with discrimination of specific anatomicalareas with DXA image analysis and dissection. In previouswork with the live animal and whole carcass, it wasestablished that precision of DXA models for TTM wasnot affected by the incorporation of subregions into a scanimage (Suster et al. 2003). Furthermore, DXA models forTTM are not influenced by any inaccuracies associated withchemical analysis or manual dissection. Therefore, althoughDXA-derived TTM is not a valuable measurement in a directsense, it can reveal much about the efficacy of regional gridplacement to the true position of the primal cut in the scanimage. If subregions in the scan image are placed accurately,the precision of the model for predicting weight from DXAin the primal cut should be equal to that of the half-carcass.

Precision of the model for ham weight developed fromDXA-derived TTM was not different from the equivalentmodel in the half-carcass. Therefore, regional placement ofthe ham was accurate to the true position of the ham on thescan image, an expected outcome because the ham regionwas easiest to consistently delineate accurately. Models forthe middle and shoulder, and more so belly and loin, wereless precise than models for the half-carcass, indicatingthat subregional placement errors exist for the shoulder andmiddle and are larger in the belly and loin. This was inline with practical delineation difficulties encountered inthis study. For example, correct placement of the regionalline separating the shoulder from the middle was sometimesdifficult in images where boundaries between the ribsin the scan image were not visually distinct. Consistentdivision of the loin from the belly was also difficult andwas exacerbated by inaccurate middle separation from theshoulder. Therefore, care needs to be taken to ensure accurateand consistent placement of subregions to ensure that TTMand consequently composition measurement is precise in theprimal cuts.

Primal-cut composition

The DXA models for dissectible fat followed the sameorder of precision across the primal cuts as did models forweight. Within experiments, DXA models for dissectible fatwere more precise in the half-carcass than in primal cuts.Across primal cuts, DXA models for dissectible fat in theham were most precise and those for middle and shoulderand more so belly and loin were less precise. However,DXA models for lean meat and bone did deviate from thisorder of model precision across the primal cuts. Therefore,subregion delineation difficulties may be of major influencein fat measurement by DXA but not lean meat and bone

measurement. This is not unexpected because the effectof incorporating subregions into a whole body scan waspreviously shown (Suster et al. 2003) to mainly affect theDXA-derived FTM estimates.

Surprisingly, the DXA model for dissectible lean tissuewas stronger in the ham than in the half-carcass, both withinand across experiments. This is unusual since it is expectedthat although regional placement of the ham region canbe very consistent with the true position, at least somedegree of variation should still exist that will detract frommodel precision. In addition, there is less lean tissue inthe ham relative to the half-carcass, suggesting that anyerrors would be more pronounced in the ham. This beingthe case, an element of error that negates errors associatedwith subregion delineation must exist in the half-carcass thatis not present in the ham. For example, the ham containsthe highest percentage of lean meat relative to bone, whichmeans that inaccuracies, within DXA, that relate to tissuehidden by bone, are minimised. Alternatively, the complexbone structure associated with the rib cage in the middleregion introduces scanning difficulties to the half-carcass.The extent of dissection difficulty also differs across primalcuts, and this will influence model precision. For example, theham had a lower proportion of inter-muscular fat than the half-carcass, which would reduce dissection errors. Alternativelythe layered nature of the belly will present dissection difficultyin the half-carcass. These outlined problems stem from themiddle primal cut, and together with practical delineationdifficulties, explain why prediction models for the middle,belly, and loin are less precise than the half-carcass and otherprimal cuts.

Between-experiment variation

Although the pigs in this study came from only 5 differentexperiments, the statistical analysis showed that this numberwas sufficient to evaluate effects of BEV on the predictionequation. In some cases the effect of BEV was very large.Model ESD was always greater upon addition of BEV.However, the extent of increase in model ESD from theinclusion of BEV depended on the measured attribute andvaried between the half-carcass and the primal cuts. Overall,the accuracy of DXA for the prediction of weight, dissectiblelean meat, and bone was better maintained when BEV wasincluded, than was the accuracy of DXA for the predictionof fat. This is in line with observations made in a previousstudy (Suster et al. 2003).

The increase in ESD around the model resulting from theinclusion of BEV may be explained by a variety of factors.Most pertinent in the present study are errors associated withsubregion delineation, which will be exacerbated betweenexperiments. A greater increase in error is therefore expectedupon inclusion of BEV for models in regions that are difficultto consistently place in the true position of the primal cut. Thisis particularly so for the belly and loin primal cuts discussed


previously. The effect of BEV on model precision for weightcan be used to gauge the true effect of inaccuracy in subregiondelineation on BEV, as already explained. The effect of BEVon precision of the model for weight was the smallest in thehalf-carcass and ham and was larger in other primal cuts.In general, the extent to which BEV reduced precision ofmodels for lean meat followed this same order. However,DXA estimates of FTM were unusual as, upon addition ofBEV, the models for middle, belly, and loin maintained theiraccuracy better than models for other primal cuts. Someexplanation may be in the higher amount and percentage offat in these primal cuts.

The BEV effect will also be influenced by alterationsin tissue hydration among carcasses. Tissue hydration levelvaries among species of differing age (Manners and McCrea1963), with different metabolic modifiers (Ostrowska et al.1999, 2003), and is affected in chilled carcasses throughdrip loss (Honikel et al. 1986). Since DXA estimates leantissue and, to some extent, fat tissue from water content,any fluctuation in tissue hydration will affect the modelprecision. These tissue hydration variations will inevitablybe exacerbated among experiments. Technicians assisting inthe manual dissection were also sometimes different amongexperiments and may not be consistent in the way that theyperform dissection, which will exacerbate model error onceBEV is included. However, in this study, emphasis was placedon maintaining consistency of experimental technique amongexperiments. In particular, dissections were standardised,with similar chilling time and temperature between slaughter,scanning, and dissection, to standardise drip-loss. It would beexpected that if less care is taken to maintain experimentalconsistency, a larger between-experiment variation mayensue. Thus, there is still a need to minimise between-experiment differences in technique when using DXA.

Weight and P2 fat depth

Models including weight or P2 alone in the predictionof measured attributes have not been presented in thismanuscript. In general, if weight was a good predictor ofan attribute, addition of P2 provided little improvement tomodel precision. However, if weight and P2 were both poorpredictors, the combination of the two generally improvedmodel precision.

In particular, the DXA models for the prediction ofdissectible fat were more precise than models includingweight, P2, and a combination of these when BEV wasincluded. A similar observation was made previously whencomparisons were made between live animal measurementsand chemically derived values (Suster et al. 2003).Furthermore, models developed from weight and P2 wereextremely poor, compared with DXA, in predicting fat inthe belly when BEV was included. It is likely that thisis a result of differences in fat distribution within theanimal, a factor that will be particularly pronounced among

experiments. For example, fat distribution throughout thebody is altered by porcine somatotropin (Suster et al. 2001),which was used in one of the experiments from which thesedata were generated. Because the scenario when BEV isincluded would be more typical of industry conditions thanthe within-experiment scenario, the adequacy of weight andP2 for predicting fat needs scrutiny. Surprisingly, the modelincluding a combination of weight and P2 was more precisethan the model including DXA for the prediction of leanmeat in the loin region. This may be due to difficulty inaccurate regional grid placement to the true loin regionnegatively affecting the DXA prediction, seen by higherESD in the loin model compared with the middle model.Another explanation stems from the bulk of lean meat in theloin section comprising the eye-muscle, which is enclosedby extensions on the vertebral column. Therefore, most ofthe pixels in the DXA scan image around the eye-musclewould also contain bone and may lead to scanning difficulties.Furthermore the P2 measurement is taken from the loinregion, so the effect of fat distribution on the P2 back-fatmeasurement would be minimised here.

Conclusions

In the half-carcass and primal cuts, DXA provided anaccurate measurement of dissectible body compositionwithin experiments and when BEV was included. However,DXA estimates need to be adjusted with the use ofappropriate regression equations to correct the in-builtalgorithms. In addition, care needs to be taken to ensureaccurate and consistent delineation of subregions to the trueposition of primal cuts in the DXA scan image to ensurea reliable measurement. DXA was superior to weight andP2 fat depth for the measurement of fat across all primalcuts and especially in the belly when BEV was included.However, P2 used in conjunction with weight can be useful insome instances to measure lean meat yield in the half-carcassand primal cuts, particularly in the loin. DXA is a practicaland accurate method for tracking changes in fat distributionthroughout the half-carcass under research conditions andmay be a potential method for carcass grading in the pigindustry.

Acknowledgments

The first author thanks Australian Pork Limited (APL) forfinancial assistance. The authors thank Kym Butler forbiometrical advice. The authors also thank Paul Meredithand Robert Nightingale for help with dissections.

References

Anon. (1996) ‘Hologic Inc. QDR 4500 fan beam X-ray densitometer,user’s guide.’ (Waltham: USA)

Channon HA, Baud SR, Payne AM (1999) Adoption of valuebased trading by the Australian pork industry. Pig Research andDevelopment Corporation, Final Report. Project DV 157/1320.


D’Souza DN, Pethick DW, Dunshea FR, Suster D, Pluske JR, Mullan BP(2004) The pattern of fat and lean muscle tissue deposition in femalepigs during the finisher growth phase. Livestock Production Science(in press).

Elowsson P, Forslund AH, Mallmin H, Feuk U, Hansson I, Carlsten J(1998) An evaluation of dual-energy X-ray absorptiometry andunderwater weighing to estimate body composition by meansof carcass analysis in piglets. Journal of Nutrition 128, 1543–1549.

Gardner JAA (1990) Industry structure and trends. In ‘Pig productionin Australia’. (Eds JAA Gardner, AC Dunkin, LC Lloyd) pp. 7–11.(Butterworths: Sydney)

Genstat (2000) ‘The guide to Genstat�. Part 2: Statistics.’ (Ed.RW Payne) (Lawes Agricultural Trust, Rothamsted ExperimentalStation, VSN International Ltd: Oxford, UK)

Honikel KO, Kim CJ, Hamm R, Roncales P (1986) Sarcomereshortening of pre-rigor muscles and its influence on drip loss. MeatScience 16, 267–282. doi: 10.1016/0309-1740(86)90038-0

Kelly TL, Berger N, Richardson TL (1998) DXA body composition:theory and practice. Applied Radiation and Isotopes 49, 511–513.doi: 10.1016/S0969-8043(97)00226-1

Kistorp CN, Svendsen OL (1998) Body composition resultsby DXA differ with manufacturer, instrument generation andsoftware version. Applied Radiation and Isotopes 49, 515–516.doi: 10.1016/S0969-8043(97)00063-8

Laskey MA, Phil D (1996) Dual-energy X-ray absorptiometryand body composition. Nutrition 12, 45–51. doi: 10.1016/0899-9007(95)00017-8

Lukaski HC, Marchello MJ, Hall CB, Schafer DM, Siders WA(1999) Soft tissue composition of pigs measured with dual X-rayabsorptiometry: comparison with chemical analyses and effects ofcarcass thicknesses. Nutrition 15, 697–703. doi: 10.1016/S0899-9007(99)00154-9

Manners MJ, McCrea MR (1963) Changes in the chemical compositionof sow reared piglets during the 1st month of life. The British Journalof Nutrition 17, 495–513.

Marcoux M, Bernier JF, Pomar C (2003) Estimation of Canadian andEuropean yields and composition of pig carcasses by dual energyX-ray absorptiometry. Meat Science 63, 359–365. doi: 10.1016/S0309-1740(02)00094-3

Mitchell AD, Conway JM, Potts WJ (1996) Body composition analysisof pigs by dual-energy X-ray absorptiometry. Journal of AnimalScience 74, 2663–2671.

Mitchell AD, Scholz AM, Pursel VG, Evock-Clover CM (1998)Incremental changes in total and regional body composition ofgrowing pigs measured by dual-energy X-ray absorptiometry.Growth, Development, and Aging 60, 95–105.

Moore DS (1995) ‘The basic practice of statistics.’ (WH Freeman andCompany: New York)

Nord RH (1998) DXA body composition properties: inherent inthe physics or specific to scanner type? Applied Radiationand Isotopes 49, 517–518. doi: 10.1016/S0969-8043(97)00064-X

Ostrowska E, Muralitharan M, Cross RF, Bauman DE, Dunshea FR(1999) Dietary conjugated linoleic acids increase lean tissue anddecrease fat deposition in growing pigs. Journal of Nutrition 129,2037–2042.

Ostrowska E, Suster D, Cross RF, Muralitharan M, Bauman DE,Dunshea FR (2003) Conjugated linoleic acid alters bodycomposition in pigs: evaluation by Dual-Energy X-RayAbsorptiometry (DXA). The British Journal of Nutrition 89,219–229. doi: 10.1079/BJN2002765

Pietrobelli A, Wang Z, Formica C, Heymsfield SB (1998) Dual-energyX-ray absorptiometry: fat estimation errors due to variationin soft tissue hydration. American Journal of Physiology 274,E808–E816.

Proctor DN, O’Brien PC, Atkinson EJ, Nair KS (1999) Comparison oftechniques to estimate total body skeletal muscle mass in peopleof different age groups. American Journal of Physiology 277,E489–E495.

Suster D (2004) Evaluation of dual energy X-ray absorptiometry forpredicting whole body and carcass composition of pigs. PhD thesis,University of Melbourne, Australia.

Suster D, Leury BJ, Hewitt R, Kerton DJ, Dunshea FR (2001)Porcine somatotropin (Reporcin�) decreases carcass and bellyfat in the finisher gilt. In ‘Manipulating pig production VIII’.(Ed. PD Cranwell) p. 68. (Australasian Pig Science Association:Werribee, Vic.)

Suster D, Leury BJ, Ostrowska E, Butler KL, Kerton DJ, Wark JD,Dunshea FR (2003) Accuracy of dual energy X-ray absorptiometry(DXA), weight and P2 back fat to predict whole body andcarcass composition in pigs within and across experiments.Livestock Production Science 84, 231–242. doi: 10.1016/S0301-6226(03)00077-0

Tothill P, Avenell A, Love J, Reid DM (1994) Comparisonsbetween Hologic, Lunar and Norland dual-energy X-rayabsorptiometers and other techniques used for whole-body softtissue measurements. European Journal of Clinical Nutrition 48,781–794.

Manuscript received 3 March 2004, accepted 9 July 2004

http://www.publish.csiro.au/journals/ajar

The accuracy of dual energy X-ray absorptiometry (DXA), weight, and P2 back fat to predict half-carcass and primal-cut composition in pigs within and across research experiments

Documents