Les Cahiers de l’IFIP...Les objectifs de cet article sont de pro poser une méthode de...

Les Cahiers de l’IFIP - Vol 3 - n° 1 - 2016 59

2016

-Ifip

-Insti

tut d

u po

rc -

All r

ights

rese

rved

2016

-Ifip

-Insti

tut d

u po

rc -

All r

ights

rese

rved

Revue R&D de la filière porcine françaiseVol 3 - N° 1 - 2016

Les Cahiers de l’IFIP

Keywords: computed tomography, dissection, reference, comparison, classification, lean meat contentMots clés : scanner, dissection, référence, comparaison, classification, pourcentage de muscle

The grading of pig carcasses participates in profitability and orientation of production. Framed by the European regulations, it is based on the prediction of lean meat content. The reference content is measured by manual dissection or since 2009 by X-ray tomography, if it gives results comparable to dissection. The objectives of this article are: to propose a method of comparison, to apply it to the French procedure and discuss its equivalence on the European herd. The approach is essentially based on a test conducted on 63 half-carcasses representative of French livestock in 2008, in the framework of the control of classification equations. The orthogonal regression proved to be a suitable method of comparison. The procedure of computed tomography chosen by France, developed independently of dissection, was comparable to dissection. The essential difference is a mean bias of 0.6 points. The relationship between the two references is not sensitive to the nature of livestock, particularly in terms of gender and halothane genotype. This tomographic method can replace dissection in France and Europe as a reference for the classification of pigs and more generally for measuring body composition.

Le classement des carcasses de porc participe à la rentabilité et à l’orientation de la production. Encadré par la règlementation euro-péenne, il est basé sur la prédiction d’une teneur en viande maigre. Cette teneur de référence est mesurable par dissection manuelle ou depuis 2009 par tomographie RX, si celle-ci donne des résultats comparables à la dissection. Les objectifs de cet article sont de pro-poser une méthode de comparaison, de l’appliquer à la procédure française et de discuter de son équivalence sur le cheptel européen. L’approche repose essentiellement sur un essai réalisé sur 63 demi-carcasses représentatives du cheptel français en 2008, dans le cadre du contrôle des équations de classement. La régression orthogonale s’est avérée une méthode de comparaison adaptée. La procédure de tomographie RX choisie par la France, développée indépendamment de la dissection, s’est révélée comparable à la dissection. La différence essentielle est un biais moyen de 0,6 point. La relation entre les deux références n’est pas sensible à la nature du cheptel, notamment en matière de sexe et de génotype halothane. Cette méthode tomographique peut remplacer la dissection en France et en Europe comme référence pour la classification des porcs et plus généralement pour la mesure de la composition corporelle.

Gérard DAUMAS, Mathieu MONZIOLSIFIP‐Institut du Porc, BP 35104, 35651 Le Rheu Cedex, France

[email protected]

X-ray Computed Tomography: reference for pig classification

La tomographie à rayons X : référence pour la classification des porcs

Les Cahiers de l’IFIP - Vol 3 - n° 1 - 201660

2016

-Ifip

-Insti

tut d

u po

rc -

All r

ights

rese

rved


Introduction

Payment for pig farmers depends on the body composi-tion of their animals. Within the single market European rules govern classification of carcasses. The composition of carcasses must be estimated from authorized methods predicting lean meat content. The measurement of this reference content was based until 2008 only on manual dissection by knife. Since January 2009 Computed Tomography (CT) is also allowed provided it yields results comparable to dissection. The medical imaging and in particular CT has been used since the early 1980s in animals (Skerjvold et al., 1981). The measurement of body composition, in particular, is one of the main applications of this technology. The program EUPIGCLASS (http://www.eupigclass.net/), whose objective was to strengthen the harmonization of pig classification methods at the European scale, stu-died the potential of CT for replacing dissection as refe-rence measurement method of the lean meat content. The positive results of the program (Dobrowolski et al., 2004; Romvari et al, 2006) have led most major areas of European production to be equipped with this techno-logy (Germany, Denmark, Spain, France). The French pig chain, thanks to Inaporc, the Office of Livestock and IFIP, has invested in a mobile CT in 2008 (Daumas and Monziols, 2008).Firstly, CT equipped Member States have developed their own acquisition protocols and methods of measurement (Christensen and Borggaard, 2005; Christensen et al., 2006; Judas, Höreth, Dobrowolski, 2006; Judas et al. 2006; Font i Furnols et al., 2009; Vester-Christensen et al, 2009). All these methods were based on scanning carcasses and a cali-bration by regression on lean meat content measured after dissection.For its part, France has delayed a few months a control test of classification methods (Daumas et al., 2010), to inte-grate the recently purchased scanner and acquire elements of comparison to manual dissection, which could be used after the publication of the new regulation.This was published on 16 December 2008 (EC Regulation No. 1249/2008). It has introduced two novelties. The first was to add to the TMP reference (Daumas, 2008) a second reference, corresponding to the lean meat content in a side without head. The second innovation was the addition of the following paragraph:«The dissection referred to in the second subparagraph may also be replaced by assessing the lean meat percentage by means of total dissection with a computer tomography apparatus on the condition that satisfactory comparative dissection results are provided.»Obviously such writing makes the CT status very pre-carious. In particular, in the absence of clear rules to judge the degree of satisfaction of the comparison

between CT and dissection, number of interpretations can be made.Under these conditions, unlike other Member States, France has decided to develop a CT method, independent of the dissection and robust, and promote a European refe-rence lean meat content measured by CT. The IFIP began by developing such a method (Monziols and Daumas, 2010; Daumas and Monziols, 2011a, b, c).Then the IFIP has organized days on the subject in 2010 (FAnI Days), published the proceedings (IFIP, 2010) and actively worked to the establishment of a European Network on this topic. These efforts were successful in 2011, with funding by the COST Office of a network on farm animal imaging (FAIM 2011-2015; http://www.cost.eu/COST_Actions/ fa / FA1102). The IFIP has assured the animation of the group on body composition and campai-gned in favor of a European harmonization. Unfortunately, this had not fully succeeded and some participants felt that obstacles had to be raised in a draft European research project. Pending this eventual project, a description of the national methods was compiled in a brochure (Daumas, Donko, Maltin, Bünger, 2015).Under these conditions, the IFIP had to propose criteria to compare CT and dissection, show that the tomographic procedure of IFIP was equivalent on the French herd and discuss its equivalence on the European herd.The approach is based on statistical bibliography, on the analysis of comparative trial data generated in France in 2008 and the discussion of testing abroad.

Evolution of references of lean meat content and related procedures The single market within the EU implies the uniqueness of Community references. For pig classification the national references were dropped in favor of a single EC reference in the eighties. The EC reference was defined as the percen-tage of muscle in the carcass. At that time, only dissection allowed to separate muscles from other tissues.The EC reference, based on total dissection, has in prac-tice almost never been used by the Member States (MS), as very costly, very tedious and very time consuming. An EC trial in the early 90s allowed to test a simplified dissection method, and then to agree on a new definition of lean meat content.The simplified dissection was to dissect the four main parts (ham, loin, shoulder, belly) from a standard cutting, inspired by the German DLG cutting (Scheper and Scholz, 1985). Then, the lean meat content was defined as the ratio between the muscle weight of the four main cuts and the carcass weight. Finally, in order to maintain consis-tency of quotations and the EUROP grid, the Commission requested that a scale factor was introduced. Thus the multiplicative factor of 1.3 was used. The formula of


2016

-Ifip

-Insti

tut d

u po

rc -

All r

ights

rese

rved

2016

-Ifip

-Insti

tut d

u po

rc -

All r

ights

rese

rved


this new definition, called TVM in France (Daumas and Dhorne, 1994, 1996) was:

TVM = 1,3 x 100 TL + ∑ muscle weight (H,L,S,B)

half carcass weight

where TL, H, L, S, B represent the respective weight of the following cuts: TenderLoin, Ham, Loin, Shoulder, Belly. The tenderloin is likened to pure muscle. In practice, only a half-carcass is dissected, the other half being assumed equivalent. TVM and simplified dissection have been described in detail in a procedure (Walstra and Merkus, 1996).

The first form of the new reference was applied between 1 January 1995 and 30 June 2006. The Commission Regulation (EEC) N° 2967/85 was amended by the Commission Regulation (EEC) N° 3127 / 94.Then, following the results of the European research pro-ject EUPIGCLASS, who pointed to the lack of reproduci-bility of the measurement procedure of the TVM (Nissen et al., 2006), it was recommended to change the definition of the reference, at least to limit the impact of cutting errors. For this, the TVM denominator (weight of car-cass) was replaced by the weight of the dissected parts, that is to say, those of the numerator. The new criterion was therefore the percentage of muscle in the four main cuts (tenderloin included). After a discussion on the new scale factor, the multiplicative factor of 0.89 was chosen. The formula of this new definition, called TMP in France (Daumas, 2008), was:

TMP = 0,89 x 100 TL+∑ muscle weight (H,L,S,B)

TL+∑ weight (H,L,S,B)

where TL, H, L, S, B represent the respective weight of the following cuts: TenderLoin, Ham, Loin, Shoulder, Belly.

The second amendment by the Commission Regulation (EEC) N° 1197/2006 entered into force on 1 July 2006. This reference, based on the simplified dissection, is still in force today, in the framework of Regulation (EC) N° 1249/2008, which replaced on 1 January 2009 the afore-mentioned Commission Regulations.The initial procedure (Walstra and Merkus, 1996) was partly invalid, as based on TVM. However, this ratio having the same numerator than TMP, that is to say the weight of tenderloin and the muscle weight of the four main cuts, cutting and dissection procedure was still applicable.Although the procedure, supported by photographs, spe-cifies a dissection separating every muscle, the trials in EUPIGCLASS showed that this anatomical dissection was not performed with the same rigor by all Member States. Moreover, the procedure indicated a few guidelines for the dissection of each cut, without mentioning all the muscles to dissect.

Also in France, the steering committee for pig classification, chaired at the time by Ofival (currently FranceAgriMer) decided to improve the reproducibility of the European pro-cedure for dissection. The first step was to recognize, identify and accurately name all the muscles involved. The second step was to identify the dissection steps of each cut and create paper and electronic media. This work was spread over seve-ral years as part of an excellent collaboration between Ofival, IFIP and the chair of anatomy of ENVN (currently Osiris Nantes). A lot of documents were produced.The first works were worn by a student of ENVN, as part of his internship thesis (Chatellier, 2003) and his thesis for the veterinary doctorate (Chatellier, 2004). A CD on cutting and dissection of the four parts was then adapted and dis-tributed by IFIP (Nictou et al., 2005). It was also presented at a conference (Betti et al., 2005).The thesis of another student (Venet, 2007) then turned to the other cuts to design a complete dissection procedure for all of the half-carcass. A CD was produced.Then, paper media on the dissection of the four parts were disseminated, first in a series of detachable boards per cut (Daumas et al, 2010; Nictou et al, 2010a, b, c, d.). This series was later compiled into a single document (Nictou et al., 2014). It is this document which is the most success-ful procedure of implementation of the European partial dissection.Another of the conclusions of EUPIGCLASS was recom-mending the eventual replacement of the dissection by an instrumental tomographic technique. Gradually, some MS were equipped with CT. Following unsuccessful attempts to find EU funding on this theme, each MS has continued to develop a national approach.Since 1 January 2009, Regulation (EC) N° 1249/2008 has introduced an additional reference, the % of muscle in the carcass, as well as the possibility of using tomography pro-vided to give results comparable to the dissection. However, to date no European procedure is available, neither on total dissection nor on tomography nor to ensure equivalence between tomography and dissection.

Material and methods

Experimental design

A trial was conducted with two objectives:•Check the methods of classification at that time,•Acquire data with CT and compare them with those of

dissection.

Only the second objective is addressed in this article.To limit the costs of checking equations, the operation’s steering committee had decided to set the sample size to sixty carcasses. The test was conducted during six weeks, from late September to early November 2008.


2016

-Ifip

-Insti

tut d

u po

rc -

All r

ights

rese

rved


To improve the representativeness of the sample, stratifica-tion was made according to two factors: the slaughter pigs region and sex, according to the proportions estimated in the French pig population. Two regions were selected: the great West and the rest of France; they were each repre-sented by a slaughterhouse. Castrated males and females were selected in the same proportions.As the genetic effect is being revealed very low in the pre-vious update of classification methods in 2005, this factor has been removed from the design. However, the pig genotype in halothane locus (Hal) was considered a posteriori, to study the influence of this major gene on the prediction of TMP.

Classification methods measurements

Measurements of two classification methods were collected at slaughter («hot measures»):•Those of CGM, the method in force in 2008 in all slaugh-

terhouses of more than 200 pigs per week (Decision 2008/677 / EC)

•Those of the manual method ZP, in force in 2008 in all slaughterhouses less than 200 pigs per week (Decision 2006/784 / EC).

CGM measures have been taken by a single operator on the motionless carcasses placed on a bypass rail.

ZP measures have been taken by two operators on the motionless carcasses placed on a bypass rail.

Cutting and dissection

After 24h chilling, the left half carcasses were cut accor-ding to the standardized European procedure (Figure 3) (Walstra and Merkus 1996).

Figure 3: Standardized European cutting

After CT examination (see next section), the four main cuts (loin, belly, shoulder and ham) were dis-sected according to standard EU procedure (Walstra and Merkus 1996). The dissection team was trained in advance, using the CD designed for this purpose. The dissection boards of each cut, later broadcast (Nictou et al., 2014), were posted around the dissection table. To verify the anatomical integrity of each muscle, the dissection of each cut was divided into four stages. At each step, an intermediate weighing was performed. An example of this approach is illustrated in Figure 4, which shows the separation of internal and external intercostal muscles of the belly. The entire muscle groups per cut and the names of the muscles was collected in Daumas et al. (2010).

Figure 4: Example of dissection of belly intercostal muscles

The strict application of this approach was then used to produce a burst of every cut, showing the quality of the dissection. Figure 5 is an illustration for the belly.

G2M26 cm

fentesite costal (3/4 DC)

Figure 1: Fat (G2) and muscle (M2) measurements of the CGM method

MG

G = minimum thickness of fat (including rind) covering the gluteus medius muscle (hatched);

M = minimum thickness of the lumbar muscle between the anterior extremity of muscle gluteus medius (hatched) and the dorsal surface of the spinal canal.

Figure 2: Measurement sites of fat (G) and muscle (M) thicknesses at the splitline of the ZP method


2016

-Ifip

-Insti

tut d

u po

rc -

All r

ights

rese

rved

2016

-Ifip

-Insti

tut d

u po

rc -

All r

ights

rese

rved


IFigure 5: Belly burst

Acquisition and analysis of tomographic images

Before dissection, the cuts were transferred to the trailer containing the IFIP CT (Figure 6), parked at the end of shipping dock of the slaughterhouse, and then they were scanned (Figure 7).The CT was a Siemens Emotion Duo (Siemens, Erlangen, Germany). The acquisition protocol, the same for the four cuts, was as follows: 130 KV voltage, current 40 mAs, slice thickness 3 mm, helical mode. The acquisi-tion took about 1mn30 by cut.The images (Figure 8) were analyzed by automatic pro-cessing using a software developed by IFIP (Monziols, Faixo et al., 2013).The volume of each cut was obtained by thresholding, that is to say, by selecting the amount of pixels above -500 Hounsfield units (HU). Muscle volume of each piece was also obtained by thresholding, recruiting in this case the pixels of the image whose signal was between 0 and 120 HU (Figure 9). This choice, which is crucial, was motivated by the following arguments.First, the 0 of the Hounsfield scale is a natural threshold between muscle and fat, this value corresponding to the CT calibration with respect to pure water, whose density is 1. Second, the distribution of the CT signal of a pure tissue, such as muscle tissue, is considered following a Gaussian, so a symmetrical distribution. Third, several authors have reported an average or a mode of muscle pig carcasses muscles around 60 HU (Font i Furnols et al., 2009; Vester-Christensen et al., 2009). Finally, the upper threshold of 120 HU was inferred from the symmetrical 0-60 interval.

For each cut, the number of voxels muscle was multiplied by the volume of voxels (FoV / Matrix x slice thickness = 2.86 mm³), to obtain the muscle volume.This volume was then multiplied by a fixed density of 1.04 (ICRU, 1989) to calculate the muscle weight.

Figure 6: CT within the trailer (examination room)

Figure 7: Scanning of the four cuts

Figure 8: Example of tomographic slice (ham)

Figure 9: Example of tomographic slice after thresholding (ham)


2016

-Ifip

-Insti

tut d

u po

rc -

All r

ights

rese

rved


TMP calculation

TMP is defined in paragraph 2 of Annex IV of Regulation (EC) N° 1249/2008. Its formula is presented below for the TMP from the manual dissection by knife, called TMPdis:

TMPdis = 0,89 x 100 TL + ∑ muscle weight (H,L,S,B)


where FM, J, L, E, P represent the respective weight of the following cuts: Filet Mignon, ham, loin, shoulder, belly.

The formula of TMP from CT (TMPct) is similar, but the calculation of muscle weight was detailed:

TMPct = 0,89 x 100 TL+∑ muscle volume (H,L,S,B)


Statistical analyses

Influential data were detected by robust regression. The procedure recommended in the guide of good statistical practices for classifying pigs (Causeur et al., 2003) was followed. The estimator LTS (Least Trimmed Squares) of truncated least squares proposed by Rousseeuw (1984) was implemented using SAS version 9.4 (SAS Inst. Inc., 2012) and the option «Method = LTS» of ROBUSTREG procedure. The estimator S, introduced by Rousseeuw and Yohai (1984), having a better efficiency, is now recom-mended in the R software documentation. We used it thanks to the SAS option «Method = S».Descriptive statistics were calculated with the MEANS pro-cedure and the correlation with Proc CORR.The main objective of the analysis is to compare two methods, dissection and CT. If dissection is currently reco-gnized as a reference, CT is also intended to be a reference. Accordingly, the two variables play a symmetrical role and the classical OLS (Ordinary Least squares) regression is theoretically inadequate.In the case of two continuous variables, Ludbrook (2002) advises Ordinary Least Products (OLP) regression; it is to minimize the sum of products of horizontal and vertical deviations from the regression line. According to the author, the method developed by Bland and Altman (1983), and more widely by Bland and Altman (1986, 1999), would also be used in case of there would be no proportional bias. This method is to plot the differences of the two variables against their means.Another aspect to consider in the regression is the impor-tance of measurement errors. In classical regression, the predictors X are assumed to be without measurement error (or their errors are assumed to be negligible). In OLP regression, also called orthogonal regression (Carroll and Ruppert, 1996), both variables can be prone to measurement error. But we do not have reliable estimates of these errors.

However, the relative importance of these errors can guide the choice of the statistical method. Two common strategies are to make one of two assumptions (McArdle, 2003):•Equal errors,•Error variances proportional to total variances.

The two methods used are then respectively, the Major Axis (MA) and the Reduced Major Axis (RMA). They were tested both.

The bMA slope of the major axis of TMPdis (Yd) on the TMPct (Yt) is equal to (Borcard, 2005):

bMA = d ± √d² + 42

bRMA = ca( )

with d = [s²(Yd) - s²(Yt)] / [r s(Yd) s(Yt)]and where s (Y) represents the standard deviation of Y and r the correlation between Yd and Yt.The sign after d is the sign of the correlation r (+ here).Knowing that the line passes through the centroid [m(Yd), m(Yt)] of the scatter of points, the intercept kMA of the major axis is easily deduced by the equation: kMA = m(Yd) - bMA m(Yt), where m(Y) represents the average of Y.

As for the bRMA slope of the reduced major axis of the TMPdis (Yd) on the TMPct (Yt), it is equal to:

bMA = d ± √d² + 42

bRMA = ca( )

where a is the slope of the least squares regression of Yd on Yt and c represents the slope of the least squares regression of Yt on Yd. This slope corresponds to the bisecting line of the OLS regression line of Yt on Yd and the OLS regression line of Yd on Yt in the same plot.

Knowing that, there too, the line passes through the cen-troid [m(Yd), m(Yt)] of the scatter of points, the intercept kRMA of the reduced major axis is easily deduced by the equation:

kRMA = m(Yd) - bRMA m(Yt)The least squares regressions were calculated by the Proc REG and testing with the test instruction.

The residual error was then decomposed by Theil (1966), quoted by Benchaar (1998). This decomposition allows to separate the error in central tendency (or bias) (ECT), the error due to regression (ER) and the error due to distur-bances (ED). By keeping the same notation, the formulas thus write:ECT = [m(Yt) - m(Yd)]²ER = s²(Yt) (1 - a)²ED = (1 - r²) s²(Yd)These errors were then expressed as a percentage of the residual error.


2016

-Ifip

-Insti

tut d

u po

rc -

All r

ights

rese

rved

2016

-Ifip

-Insti

tut d

u po

rc -

All r

ights

rese

rved


Given the results, the graph Bland-Altmann was then per-formed.Finally, the effects of sex and halothane genotypes factors, as well as fat and muscle thicknesses covariates were tested by the Proc GLM.

Results

Table 1 shows the sizes of cross terms of the two factors: gender and halothane genotype. The sex ratio is balanced globally, according to the stratification performed. The proportion of heterozygous Nn is overrepresented and conversely the proportion of normal homozygotes NN is underrepresented.

Tableau 1: Number of carcasses per sex and halothane genotype

NN Nn nn TOTALFemales 6 24 2 32Castrated males 14 17 0 31TOTAL 20 41 2 63

Table 2 shows some centering and dispersion criteria of the main variables for the entire sample, which includes 63 observations.

No influential data was detected in the regression of TMPdis on TMPct neither by the LTS method nor by the S method. Analyses were therefore carried out on the entire sample.The standard deviations of the two methods were very close, respectively 3.65 and 3.59 for TMPdis and TMPct. An F-test failed to reject the hypothesis of equality of variances.The average of the two methods, although quite close, res-pectively 60.74 and 61.31 for TMPdis and TMPct, were statistically different.The correlation between the two methods was estimated at 0.989.The coefficients of the OLS regression of TMPdis on TMPct were: TMPdis = 1,007 TMPct - 1,00 (1)

The standard errors of the slope and the constant were 0.019 and 1.17 respectively. The equality tests of the slope

to 1 and the constant to 0 were not significant. As for the residual standard deviation, it was 0.54.

The figure 10 illustrates the results of this approach com-monly used.

53

55

57

59

61

63

65

67

53 55 57 59 61 63 65 67

TMPd

is (%

)

TMPrx (%)

TMPRX

Régression

53 55 57 59 61 63 65 67 69

53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68

TMP

préd

it (%

)

TMPrx (%)

pMCO pAMR pAM pCorbiais

-2

-1,5

-1

-0,5

0

0,5

1

53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68

dif T

MP

(%)

moyTMP (%)

Figure 10: OLS regression

Below are shown the coefficients of the ordinary least squares (OLS) regression of TMPct on TMPdis, of which the slope of the line steps in the calculation of the slope of the reduced major axis: TMPct = 0.972 TMPdis + 2.28 (2)

From both previous equations (1) and (2) was derived the following equation for the reduced major axis: TMPdis = 0.982 TMPct + 0.50

As for the method of the major axis, it gave the following equation: TMPdis = 1.018 TMPct - 1.68

53

55

57

59

61

63

65

67

53 55 57 59 61 63 65 67

TMPd

is (%

)

TMPrx (%)

TMPRX

Régression

53 55 57 59 61 63 65 67 69

53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68

TMP

préd

it (%

)

TMPrx (%)


-2

-1,5

-1

-0,5

0

0,5

1

53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68

dif T

MP

(%)

moyTMP (%)

Figure 11: Major Axis (pAM), Reduced Major Axis (pAMR), OLS regression line (pMCO) and line after mean bias correction

(pCorbiais)

Table 2: Descriptive statistics

(n=63) Mean Standard deviation Minimum MaximumTMPdis (%) 60,74 3,65 52,68 67,01TMPrx (%) 61,31 3,59 54,56 67,73ZP fat depth (mm) 13,8 3,7 6,0 23,1ZP muscle depth (mm) 74,1 6,6 60,6 57,7CGM fat depth (mm) 13,3 3,3 8,0 21,0CGM muscle depth (mm) 59,0 5,7 45,0 70,0Carcass weight (kg) 91,7 8,2 77,8 112,8


2016

-Ifip

-Insti

tut d

u po

rc -

All r

ights

rese

rved


Figure 11 clearly shows the very close proximity of these two axes and thus also of the OLS line, which has an intermediate position, as well as the line after correction of the mean bias (TMPdis = TMPct - 0,58). The highest differences between predicted values, at the extremes of the range, were only 0.2.

The residual error of the OLS regression of TMPdis on TMPct (Equation 1) was then decomposed and expressed in %. The error of central tendency (bias) accounted for 54% and the error due to disturbances for 46%, the error due to regression being negligible. Accordingly, in this case, a correction by regression should give similar results as a correction of the average bias.

Moreover, as the proportional bias (due to regression) is negligible, comparing the two methods can be illus-trated by the Bland-Altman method (Figure 12). The graph shows that the difference, about -0.6, is relati-vely fixed and does not depend on the TMP average. In addition, the variability of this difference seems quite normal.

53

55

57

59

61

63

65

67

53 55 57 59 61 63 65 67

TMPd

is (%

)

TMPrx (%)

TMPRX

Régression

53 55 57 59 61 63 65 67 69

53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68

TMP

préd

it (%

)

TMPrx (%)


-2

-1,5

-1

-0,5

0

0,5

1

53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68

dif T

MP

(%)

moyTMP (%)

Figure 12: Plot of TMP differences (difTMP) against TMP means (moyTMP)

The Sex and Halothane effects were first tested in a simple analysis of variance model. No effects were signi-ficant.Then the fat and muscle thickness of each classifica-tion method were introduced as covariates in a general linear model also containing interactions with the Sex and Halothane effects.For the ZP method, although muscle thickness was significant, it did contribute to a reduction of only 0.05 of the residual standard deviation.For the CGM method, the interaction Sex * G2 was significant at the 5% threshold. As the interaction was small, we tested its withdrawal. The Sex factor was no longer significant. The thickness of the muscle (M2) then remained significant at the 5% level. For the CGM also, the complete model contributed to decrease the residual standard deviation of only 0.05.

Discussion

Robust regression versus average bias correction

The methods of orthogonal regression, major axis and reduced major axis, gave very close results on the French sample of 63 pigs. These results are also very close to the OLS regression and an average bias correction. This is explained by the fact that the slope is not significantly dif-ferent from 1.The only correction of the mean bias, mentioned in the current working version of the draft Regulation on the classification of carcasses does not change much for the IFIP procedure. Indeed, it was designed independently of the dissection, the slope of the regression is close to 1 and the average bias (-0.6) is low. The average bias is to compare to biases between Member States arising from the application of the European dissection method. These biases are currently unknown among the 28 Member states. In 2006, Nissen et al. had published an estimate with 8 Member States on the basis of prior definition of lean meat content (TVM). The extremes reached 2 points. On the one hand, the transition to TMP had to reduce differences, but on the other hand, the current number of Member States, a practice not always conform to the procedure and the lack of regular practice had to help increase differences. The need to further standardize the reference dissection has been recently reminded by Font-i-Furnols et al., 2016.Comparison of the results with other Member States is difficult because none has built a TMPct regardless of dissection. All have estimated a TMPdis by regression on variables extracted from CT. By construction, the average bias is zero on the calibration sample. Rarely, a validation set was used.Some foreign methods suffer from significant error. In this case, a correction by regression seems more appropriate than a simple correction of the mean bias.The draft regulation provides for a correction from a sub-sample of at least 10 carcasses covering all stratification factors. In France, the only stratification factor used is sex. Genetics has been checked retrospectively by the halothane status. Neither effect was significant on the relationship between the two methods.The appearance of the entire male is the main change in the French herd since the trial in 2008. At present, its pro-portion is fairly stable at around 12%. With such a small proportion, to have a noticeable effect on the French cor-rection, the effect should be very important. Now, entire males, moreover pure Pietrain, thus very lean, exhibited peaks and dispersions of muscle in the range [0, 120] HU close to those of fatter pigs (Daumas et al. (2014). It is the-refore unlikely.


2016

-Ifip

-Insti

tut d

u po

rc -

All r

ights

rese

rved

2016

-Ifip

-Insti

tut d

u po

rc -

All r

ights

rese

rved


The Dutch Engel et al. (2012) studied the effect of gender, comparing the three sexual types, on classification equa-tions, which are more sensitive by nature to changes in livestock. They concluded that the equations by classifica-tion instrument should remain valid when castrated males would (gradually) be replaced by entire males.Accordingly, the IFIP procedure seems applicable in France, without new dissections, using the correction based on the 2008 sample.

Regarding the European herd, except Italy, the controllable factors of differentiation of national herds are also gender and genotype. The few countries that stratify according to the genotypes mention crossbreds. The effects on the equations are rarely reported. The halothane gene seems a better indicator of genetic differences. Our results showed no significant halothane and no significant differences between NN and Nn alleles. In our opinion, the nn allele, very few in the sample, is rare in Europe.Accordingly, the IFIP procedure seems applicable also in other European countries, without further dissection, using the correction based on the 2008 sample.

Italy is a special case in Europe, with a herd overwhelmin-gly composed of heavy pigs, which carcass weight ranges between 110 and 180 kg (Rossi et al., 2014). The average carcass weight is around 40 kg higher than elsewhere. However, sexual types are identical: castrated males and females, and genetic changes with a constant progression of crossbreds, approaching the other countries. However, Daumas et al. (2014) showed that the extreme values of muscle on the histograms of Hounsfield values of Italian heavy pigs did not differ substantially from those of European pigs and even very lean pigs, as entire males pure Pietrain.This leads us to think that the IFIP procedure is also appli-cable as such in Italy without new dissections.

Cuts scanning versus carcasses scanning

In addition, European regulations currently interpreted by the authorities as limiting the possibility of using CT to scan the carcass. While scanning a carcass is easier and faster than scanning the cuts, this is not consistent with the definition of lean meat content most used in the EU, namely the TMP. As TMP corresponds to the muscle percentage into the four main cuts, it makes more sense to scan these four parts. This is what is done in France. Unfortunately, the Commission then rejected the equiva-lence between CT and dissection, simply because the parts, not the half-carcass were scanned. This position lies not on scientific basis, but only on the lobbying of some national experts, wishing to impose as the only reference the percen-tage of muscle in the carcass. Also, France was forced to

use CT, not as equivalent reference to dissection, but as a national quick method through double sampling. This had two consequences: firstly, the integration of the prediction error of TMPdis by TMPct in the prediction error of the classification methods, and secondly, the need for a mini-mum size of 50 partially dissected carcasses.The other Member States contented themselves with scan-ning half carcasses. But, as all, except Germany, wanted to predict the TMP, they tried to show the equivalence between their scan procedure of half carcasses and the dissection of the four main cuts. However, equivalence between the mea-sures of different entities is far from obvious.First, the two lean meat contents authorized by the regu-lation, have an average difference of about 1 point (unpu-blished). The difference also depends on the nature of lives-tock. This became clear during discussions on the scale factor to apply in order to ensure an equivalent level over Europe between TMP and TVM. This debate, in which several Member States have proposed their own patch, ulti-mately led to the 0.89 factor.Next, the applied regression equations are inherently sen-sible to livestock. The regression of TMCdis (lean meat content in the carcass by dissection) on TMPct according to the IFIP procedure, applied to the so-called «Italian light pigs», which carcass weight ranges between 70 and 110 kg (Rossi et al., 2014), i.e. a weight comparable to the French pigs, gave the following equation (Monziols, Rossi et al., 2013): TMCdis = 0.92 TMPct + 0.14with a residual standard deviation: RSD = 0.91.

Comparing with the equation (1), which RSD was 0.43, exemplifies the extent of the effect of a different entity.

Improvements of the scanning procedure

Ensuring equivalence between CT and dissection on the same entity, regardless of the entity, seems a reasonable approach, natural and pretty obvious.Does the comparison give the same results regardless of the entity? No, that’s unlikely. Obviously, it depends on the CT procedure used. The objectives of the IFIP procedure were to be simple and accurate. The accuracy is better on the cuts than on the side (Daumas and Monziols, 2011a and b). In our opinion, this is explained by the fact that fatty and streaky parts are more difficult both to analyze and dissect. Thus, the error is larger for the belly than for the other three cuts (Daumas and Monziols, 2011c). This is probably also the case for the necks and some other minor parts.Abroad, the procedures are so far not very sophisticated and errors are greater than or at best equal to that of the IFIP (Christensen and Borggaard, 2005; Christensen et al., 2006; Judas et al. 2006; Font i Furnols et al., 2009; Vester-Christensen et al, 2009).


2016

-Ifip

-Insti

tut d

u po

rc -

All r

ights

rese

rved


The accuracy of the IFIP procedure could be improved. Three tracks seem promising: better removal of the rind, better understanding of tissue’s densities, closer analysis of Hounsfield distributions.Firstly, as rind has a similar density as muscle, it is therefore difficult to separate it completely by a simple threshold on the Hounsfield distribution. The rind can be separated by mathematical morphology, but this may come at the expense of a fair measure of the measured muscle volume. Thus, according to the Danish method, muscle density would be 1.11 (Vester-Christensen et al. 2009), which is unrealistic. More accurate methods taking into account the location of the rind on the surface would be possible, but difficult to automate.Secondly, muscle density is part of the TMPct calculation. For simplicity, it was considered fixed (1.04) (ICRU, 1989). Not only is this average poorly understood but also the variability of this density. A project with measurements by densitometry has been proposed by the authors to explore this route.Thirdly, a more refined analysis of Hounsfield distribu-tions, especially concerning the mixed voxels between fat and muscle, is possible. It could be the same for all the products scanned or adapted to each product.The adaptation of the IFIP procedure to scanned products is possible. But it is reserved primarily for products where the error is likely to be highest and where the issue deserves more complex method.

Comparison for classification in France

Finally, it is interesting to note that the mean bias between TMPdis and TMPct (-0.58) is of the same order of magni-tude as the correction introduced in the last update of the TMPdis prediction equation by Image-Meater (0.65) (Blum et al., 2014). The slope of the regression line between TMPdis and TMPct being close to 1, the equation before correction can be interpreted as a prediction equation of TMPct by Image-Meater. And this is precisely the type of equation desired by French professionals.

Conclusion

The present study demonstrates that the IFIP CT pro-cedure is a technique quite suitable for measuring weight and muscle content of the main parts of pigs. It is intended to replace the dissection for calibration of grading methods.In general, the orthogonal regression is more suitable than the classical regression and Bland-Altman method to com-pare CT and dissection as reference methods.In the absence of criteria and decision thresholds, regardless of the statistical method used, it is difficult to affirm or deny the equivalence of methods. However, the proximity of the slope to 1 and the low average bias lead us to believe that

the IFIP CT procedure can be considered equivalent to the dissection. Moreover, the method is robust to changes in sexual and genetic types.The orthogonal regression, least squares regression and mean bias correction have been very close for the IFIP pro-cedure. This is due to the very high correlation between TMPct and TMPdis and the close proximity between their variances, on one hand, and their mean on other hand. In the framework of pig classification, which is regulated, the bias may be corrected by any of these methods. It could also not be corrected, because of a level probably lower than that of the dissection uncertainty in the EU.For the other Member States, which do not have a CT method to measure the muscle percentage, the orthogonal regression is more difficult to implement. Classical regres-sion (OLS) seems to be a lesser evil. When measuring on a different sample, an adjustment would be necessary.About the content, any correction against dissection appears to be a solution of the past, because it freezes the existing bias between Member States, due to insufficient reproducibility of the dissection as currently practiced. The future is the development of a European CT procedure. The IFIP procedure is a step in this development.The accuracy and robustness of the IFIP procedure allow its safe use for calibration methods for classification wit-hout additional dissections, both in France and in other countries.The authors recommend a change to the draft European regulation on pig classification, on one hand by not limi-ting CT to scan only half carcasses and on the other hand, by taking into account the dissection uncertainty.Regarding body composition studies, the CT IFIP pro-cedure seems well suited to the vast majority of needs. It is clearly a major improvement over current methods: classification methods on slaughterline or cuts weighing. Only the cost can be a barrier. However, this cost must be compared to the one engaged to measure production traits. The authors recommend to scan at least a subsample, in addition to conventional measures (classification measures or cuts weights), when a study includes a body composition component.Similarly, for genetic selection, it would be desirable to improve the body composition measurements of candi-dates for selection using CT.Finally, for some specific needs, the scanning procedure could be refined.

Acknowledgements

The authors thank FranceAgriMer for its funding and implementation of the dissection, INAPORC for his involvement in pig classification and active members of the body composition group of the COST FAIM Action for the exchange of ideas and sharing results.


2016

-Ifip

-Insti

tut d

u po

rc -

All r

ights

rese

rved

2016

-Ifip

-Insti

tut d

u po

rc -

All r

ights

rese

rved


References

• Altman D.G., Bland J.M., 1983. Measurement in medicine: the analysis of method comparison studies. The Statistician 32, 307-317.• Benchaar C., Rivest J., Pomar C., Chiquette J., 1998. Prediction of methane production from dairy cows using existing mechanistic models and regression

equations. Journal of Animal Science, 76, 617–627.• Betti E., Guintard C., Nictou A., Grondin G., Chatellier S., Daumas G., 2005. A CD-ROM as a tool for the European reference dissection method of pig carcass:

practical and interactive anatomical guide of the four main joints (ham, shoulder, loin, belly). Proceedings of the Third Meeting of Young Veterinary Anatomists, Ghent-Antwerp, Belgium, 13-15 July 2005.

• Bland J.M., Altman D.G., 1986. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet, i, 307-310.• Bland J.M., Altman D.G., 1999. Measuring agreement in method comparison studies. Statistical Methods in Medical Research 8: 135-160.• Blum Y., Monziols M., Causeur D., Daumas G., 2014. Recalibrage de la principale méthode de classement des porcs en France. Journées Rech. Porcine,

46, 39-43.• Borcard D., 2005. Régression linéaire simple de modèle II. Cours, 13p. Téléchargé le 23/06/2016 à l’URL: biol09.biol.umontreal.ca/CoursPL/Regression.pdf• Carroll R., D. Ruppert, 1996. The use and misuse of orthogonal regression in measurement error models. Am. Statistician 50: 1–6.• Causeur D., Daumas G., Dhorne T., Engel B., Font i Furnols M., Hojsgaard S., 2003. Statistical handbook for assessing pig classification methods: recom-

mendations from the “EUPIGCLASS” project group. EC working document, 132 p. http://ec.europa.eu/agriculture/markets/pig/handbook.pdf• Chatellier S., 2003. Guide anatomique des 4 pièces principales (jambon, épaule, longe, poitrine) du porc charcutier : support à la méthode européenne

de référence de dissection pour le classement des carcasses. Mémoire de stage de thèse, ENVN Nantes, 53p et annexes.• Chatellier S., 2004. Guide anatomique des 4 pièces principales (jambon, épaule, longe, poitrine) du porc charcutier : support à la méthode européenne

de référence de dissection pour le classement des carcasses. Thèse de Dr vétérinaire, ENVN Nantes, 471p et annexes.• Christensen L.B., Borggaard C., 2005. Challenges in the approval of CT as future reference for grading of farmed animals. 51st ICoMST, Baltimore,

Maryland USA, 260-269.• Christensen L.B., Lyckegaard A., Borggaard C., Romvari R., Olsen E.V., Branscheid W., Judas M., 2006. Contextual volume grading vs. spectral calibration,

52nd ICoMST, Dublin, Ireland.• Commission des C.E., 1985. Règlement (CEE) n°2967/85 de la Commission du 24 octobre 1985, qui établit les modalités d’application de la grille com-

munautaire de classement des carcasses de porc.• Commission des C.E., 1990. Recherche concernant l’harmonisation des méthodes de classement des carcasses de porc dans la Communauté. Document

(CEE) VI/3860/89 Rév.6.• Commission européenne, 1994. Règlement (CE) n° 3127/94 du 20 décembre 1994 modifiant le règlement (CEE) n° 2967/85 établissant les modalités

d’application de la grille communautaire de classement des carcasses de porcs. JO L 330 du 21.12.1994, p. 43.• Commission européenne, 2006. Règlement (CE) n° 1197/2006 du 7 août 2006 portant modification du règlement (CEE) n° 2967/85 établissant les moda-

lités d’application de la grille communautaire de classement des carcasses de porcs. JO L 217 du 8.8.2006, p. 6.• Commission européenne, 2006. Décision 2006/784/CE de la Commission du 14 novembre 2006 relative à l’autorisation de méthodes de classement des

carcasses de porcs en France. JO L 318 du 17.11.2006, p. 27.• Commission européenne, 2008. Décision 2008/677/CE de la Commission du 28 juillet 2008 modifiant la décision 2006/784/CE relative à l’autorisation

d’une méthode de classement de carcasses de porcs en France. JO L 221 du 19.8.2008, p. 30.• Commission européenne, 2008. Règlement (CE) n° 1249/2008 du 10 décembre 2008 portant modalités d’application des grilles communautaires de

classement des carcasses de bovins, de porcins et d’ovins et de la communication des prix y afférents. JO L 337 du 16.12.2008, p. 3.• Commission européenne, 2013. Décision 2013/282/UE d’exécution de la Commission du 11 juin 2013 modifiant la décision 2006/784/CE en ce qui

concerne la formule d’une méthode de classement des carcasses de porcs autorisée en France. JO L 161 du 13.6.2013, p. 10.• Cook G.L., Yates C.M., 1992. A report to the Commission of the European Communities on research concerning the harmonisation of methods for grading

pig carcases in the Community.• Daumas G., 2008. Taux de muscle des pièces et appréciation de la composition corporelle des carcasses. Journées Rech. Porcine, 40, 61-68.• Daumas G., Causeur D., Prédin J., 2010. Validité de l’équation française de prédiction du taux de muscle des pièces (TMP) des carcasses de porc par la

méthode CGM. Journées Rech. Porcine, 42, 229-230.• Daumas G., Dhorne T., 1994. Nouvelles équations françaises de prédiction du taux de muscle des carcasses de porc. Journées Rech. Porcine en France,

26, 151-156.• Daumas G., Dhorne T., 1996. Historique et futur du classement objectif des carcasses de porc en France. Journées Rech. Porcine en France, 28, 171-180.• Daumas G., Donko T., Monziols M., Kongsro J., Candek-Potokar M., Allen P., Scholz A., Bünger L., 2014. A pragmatic short-term approach to establish

a computed tomography (CT) based reference method for the measurement of lean meat percentage (LMP) in pig carcasses. In: Maltin CA, Craigie C, Bunger L eds. FARM ANIMAL IMAGING Copenhagen 2014. Edinburgh: SRUC. 2014: 52-57.

• Daumas G., Donko T., Maltin C., Bünger L., 2015. Imaging facilities (CT & MRI) in EU for measuring body composition. 50p.• Daumas G., Monziols M., 2008. Un scanner à rayons X au service de la filière. TechniPorc, 31, N°4, 9-14.• Daumas G., Monziols M., 2011a. A simple and accurate Computed Tomography approach for measuring the lean meat percentage of pig carcasses.

Abstracts of the poster session of the 2011 annual meeting of CMSA-ASCV, p4. http://www.cmc-cvc.com/english/documents/CMSAabstracts2011.pdf (Accessed 14/09/11).

• Daumas G., Monziols M., 2011b. An accurate and simple computed tomography approach for measuring the lean meat percentage of pig cuts. Proceedings of the 57th ICoMST, 7-12 August 2011, Ghent, Belgium. Paper 061.

• Daumas G., Monziols M., 2011c. Comparison between computed tomography and dissection for calibrating pig classification methods. Proceedings of the 57th ICoMST, 7-12 August 2011, Ghent, Belgium. Paper P044.

• Daumas G., Nictou A., Guintard C., Betti E., 2010. Dissection européenne de la carcasse de porc : variabilité de la composition anatomique en muscles des 4 pièces majeures. TechniPorc, 33, N°6, encart de 6p.

• Dobrowolski A., Branscheid W., Romvári R., Horn P., Allen P., 2004. X-ray computed tomography as possible reference for the pig carcass evaluation. Fleischwirtschaft, 84 (3), 109-112.

• Engel B., Lambooij E., Buist W.G., Vereijken P., 2012. Lean meat prediction with HGP, CGM and CSB-Image-Meater, with prediction accuracy evaluated for different proportions of gilts, boars and castrated boars in the pig population. Meat Sci. 90, 338-344.


2016

-Ifip

-Insti

tut d

u po

rc -

All r

ights

rese

rved


How to cite

• Daumas G., Monziols M., 2016. X-ray Computed Tomography: reference for pig classification. Cahiers IFIP, 3(1), 59-70.

• Font-i-Furnols M., Čandek-Potokar M., Daumas G., Gispert M., Judas M., Seynaeve M., 2016. Comparison of national ZP equations for lean meat percen-tage assessment in SEUROP pig classification. Meat Science 113, 1-8. DOI information: 10.1016/j.meatsci.2015.11.004.

• Font i Furnols M., Teran M.F., Gispert M., 2009. Estimation of lean meat content in pig carcasses using X-ray Computed Tomography and PLS regression. Chemometr. Intell. Lab. Syst., 98 (1), 31-37.

• ICRU, 1989. Tissue substitutes in Radiation dosimetry and measurement. ICRU report 44.• IFIP-Institut du porc, 2010. Proceedings of the Farm Animal Imaging Congress, June 17, 2010, Rennes (France), 27p.• Judas M., Höreth R., Dobrowolski A., 2006. Computertomographie als Methode zur Analyse der Schlachtkörper von Schweinen. Fleischwirtschaft 86(12),

102-105.• Judas M., Höreth R., Dobrowolski A., Branscheid W., 2006. The measurement of pig carcass lean meat percentage with X-Ray Computed Tomography.

ICoMST Proc., 52th International Congress of Meat Science and Technology, Dublin, Ireland, 2006, pp 641-642.• Ludbrook J., 2002. Statistical techniques for comparing measurers and methods of measurement: a critical review. Clinical and Experimental

Pharmacology and physiology, 29, 527-536.• McArdle B.H., 2003. Lines, models, and errors: Regression in the field. Limnol. Oceanogr., 48(3), 1363–1366.• Monziols M., Daumas G., 2010. Comparaison entre la tomographie à rayons X et la dissection pour mesurer la teneur en muscle des pièces. Journées

Rech. Porcine, 42, 231-232.• Monziols M., Faixo J., Zahlan E., Daumas G., 2013. Software for Automatic Treatment of Large Biomedical Images Databases. Proc. SCIA, Workshop on

Farm Animal and Food Quality Imaging, Espoo, Finland, 17th June 2013, pp. 17-22.• Monziols M., Rossi A., Daumas G., 2013. Impact of pig population (light or heavy) on computed tomography (CT) and dissection relationship for lean

meat percentage measurement. In: C. Maltin, C. Craigie and L. Bünger (Eds), Farm Animal Imaging Kaposvár 2013, 22-26.• Nictou A., Guintard C., Betti E., Daumas G., 2005. Guide pratique de la dissection européenne de la carcasse de porc. CD-ROM. Ed. ITP, Paris.• Nictou A., Guintard C., Betti E., Daumas G., 2010a. Dissection européenne de la carcasse de porc : la poitrine. TechniPorc, 33, N°2, 27-32.• Nictou A., Guintard C., Betti E., Daumas G., 2010b. Dissection européenne de la carcasse de porc : le jambon. TechniPorc, 33, N°3, encart de 6p.• Nictou A., Guintard C., Betti E., Daumas G., 2010c. Dissection européenne de la carcasse de porc : l’épaule. TechniPorc, 33, N°4, encart de 6p.• Nictou A., Guintard C., Betti E., Daumas G., 2010d. Dissection européenne de la carcasse de porc : la longe. TechniPorc, 33, N°5, encart de 6p.• Nictou A., Guintard C., Betti E., Daumas G., 2014. Dissection anatomique des quatre pièces principales de découpe. In: Ifip (Ed), Mémento viandes et

charcuteries, 2014, 24 p.• Nissen P.M., Busk H., Oksama M., Seynaeve M., Gispert M., Walstra P., Hansson I., Olsen E., 2006. The estimated accuracy of the EU reference dissection

method for pig carcass classification, Meat Sci., 73, 22-28.• Romvári R., Dobrowolski A., Repa I., Allen P., Olsen E., Szabó A., Horn P., 2006. Development of a CT calibration method for the determination of lean

meat content in pig carcass. Acta Veterinaria Hungarica 54(1):1-10.• Rossi A., Bertolini A., Gorlani E., 2014. Nuova classificazione delle carcasse: via libera dall’Ue. Agricoltura, febbraio/marzo 2014, 64-65.• Rousseeuw P.J., 1984. Least Median of Squares Regression. Journal of the American Statistical Association, 79, 871–880. • Rousseeuw P.J., Leroy A.M., 1987. Robust Regression and Outlier Detection, New York: John Wiley & Sons. • Rousseeuw P.J., Yohai, V., 1984. Robust Regression by Means of S-Estimators. In J. Franke, W. Härdle, and R. D. Martin, eds., Robust and Nonlinear Time

Series Analysis, number 26 in Lecture Notes in Statistics, 256–274, Berlin: Springer-Verlag.• SAS Institute Inc., 2012. SAS /STAT Software Release 9.4, Cary, NC, USA.• Scheper J., Scholz W., 1985. DLG-Schnittführung für die Zerlegung der Schlachtkörper von Rind, Kalb, Schwein und Schaf. Arbeitsunterlagen DLG,

Frankfurt/M.• Skjervold H., Gronseth K., Vangen O., Evensen A., 1981. In vivo estimation of body composition by computerized tomography. Zeitschrift für Tierzüchtung

und Züchtungsbiologie, 98, 77-79.• Theil H., 1966. Applied economic forecasting. Amsterdam: North-Holland Publishing Company.• Venet J., 2007. Contribution à la réalisation d’un cédérom sur la dissection européenne de référence du porc charcutier en vue du classement : échine,

jarrets avant et arrière, gorge, partie arrière de la poitrine et côtes, Thèse de Doctorat vétérinaire, Nantes, 37 p.• Vester-Christensen M., Erbou S.G.H., Hansen M.F., Olsen E.V., Christensen L.B., Hviid M., Ersbøll B.K., Larsen R., 2009. Virtual Dissection of Pig Carcasses.

Meat Sci 81:699-704.• Walstra P., Merkus G.S.M., 1996. Procedure for assessment of the lean meat percentage as a consequence of the new EU reference dissection method in

pig carcass classification. Report ID-DLO 96.014, March 1996, 22 p.

Les Cahiers de l’IFIP...Les objectifs de cet article sont de pro poser une méthode de...

Documents

Transcript of Les Cahiers de l’IFIP...Les objectifs de cet article sont de pro poser une méthode de...