1
Predicting beef cuts composition, fatty acids and meat quality
2
characteristics by spiral computed tomography
3
4
N. Prieto1*, E. A. Navajas1, R. I. Richardson2, D. W. Ross1, J. J. Hyslop3, G. Simm1
5
and R. Roehe1
6
7
1
8
Edinburgh EH9 3JG, UK.
9
2
Sustainable Livestock Systems Group, Scottish Agricultural College, West Mains Road,
University of Bristol, Division of Farm Animal Science, Langford, Bristol, BS40 5DU,
10
UK.
11
3
12
UK.
Select Services, Scottish Agricultural College, West Mains Road, Edinburgh EH9 3JG,
13
14
15
16
17
18
19
*CORRESPONDING AUTHOR: Nuria Prieto. Scottish Agricultural College (SAC),
20
Bush Estate, Edinburgh EH26 0PH, UK. Tel.: +44 131 535 3361, fax: +44 131 535
21
3121. E-mail nuria.prieto@eae.csic.es; Nuria.Prieto@sac.ac.uk
1
22
Abstract
23
The potential of X-ray computed tomography (CT) as a predictor of cuts
24
composition and meat quality traits using a multivariate calibration method (partial least
25
square regression, PLSR) was investigated in beef cattle. Sirloins from 88 crossbred
26
Aberdeen Angus (AAx) and 106 Limousin (LIMx) cattle were scanned using spiral CT.
27
Subsequently, they were dissected and analyzed for technological and sensory
28
parameters, as well as for intramuscular fat (IMF) content and fatty acid composition.
29
CT-PLSR calibrations, tested by cross-validation, were able to predict with high
30
accuracy the subcutaneous fat (R2, RMSECV = 0.94, 34.60 g and 0.92, 34.46 g),
31
intermuscular fat (R2, RMSECV = 0.81, 161.54 g and 0.86, 42.16 g), total fat (R2,
32
RMSECV = 0.89, 65.96 g and 0.93, 48.35 g) and muscle content (R2, RMSECV = 0.99,
33
58.55 g and 0.97, 57.45 g) in AAx and LIMx samples, respectively. Accurate CT
34
predictions were found for fatty acid profile (R2 = 0.61 - 0.75) and intramuscular fat
35
content (R2 = 0.71 - 0.76) in both sire breeds. However, low to very low accuracies
36
were obtained for technological and sensory traits with R2 ranged from 0.01 to 0.26.
37
The image analysis evaluated in this study provides the basis for an alternative approach
38
to deliver very accurate predictions of cuts composition, IMF content and fatty acid
39
profile with lower costs than the reference methods (dissection, chemical analysis),
40
without damaging or depreciating the beef cuts.
41
Keywords: computed tomography, carcass composition, beef quality, technological
42
parameters, sensory characteristics, fatty acids.
43
Introduction
44
Consumers prefer leaner meat with the minimal fat level required to maintain
45
juiciness and flavour, a preference thought to be due to health concerns (Ngapo, Martin
46
& Dransfield, 2007). In addition, consistent quality, less wastage, convenience and ease
2
47
in cooking and high level of choice or flexibility in available cuts are of concern to
48
consumers (Aaslyng, 2009). Hence, cattle breeders need to address carcass composition
49
and meat quality traits, which will determine consumer acceptance of beef. Overall,
50
meat quality is difficult to define because it is a combination of microbiological,
51
nutritional, technological and organoleptic components. Moreover, the term “quality” of
52
carcasses has different meanings depending on local customs in different countries of
53
the world (Hocquette & Gigli, 2005). Hence, it becomes necessary to move focus from
54
the aggregate “quality” to investigate individual components of meat quality, such as
55
visual aspects (e.g. the colour of lean) or eating quality (tenderness, juiciness and
56
flavour), which in turn are affected by intramuscular fat and fatty acid composition
57
(Aaslyng, 2009).
58
Measurements of meat quality traits present particular problems for improvement, as
59
direct measurements require destruction of the sample. Muscle quality is generally
60
considered to be difficult, if not impossible, to measure in the live animal and is
61
expensive and time-consuming to measure completely in samples from the carcass
62
(Clutter, 1995). Tools to predict carcass composition for grading and classification of
63
carcasses generally use dissected composition as a reference, which is usually obtained
64
by manual dissection performed by skilled technicians. Beside the valuable and accurate
65
information provided, it is also a destructive, time-consuming and therefore a costly
66
method. Hence these methods are difficult and expensive to use in research programmes
67
or breeding programmes involving many animals, and impossible to use routinely in
68
commercial operations (Kempster, Cuthbertson & Harrington, 1982).
69
Because of these restrictions, alternative methods have been used in beef cattle to
70
predict meat quality attributes, such as near infrared (NIR) spectroscopy (Andrés,
71
Murray, Navajas, Fisher, Lambe & Bünger, 2007; Prieto, Andrés, Giráldez, Mantecón,
3
72
& Lavín, 2008; Prieto et al., 2009a). Moreover, partial dissection using sample joints
73
(Kempster & Jones, 1977), visual assessment of fatness and conformation (Kempster et
74
al., 1982), ultrasound scanning in live animals (Realini, Williams, Pringle & Bertrand,
75
2001) or video-image analysis (VIA) of carcasses (Allen & Finnery, 2001) and live
76
animals (Sakowski, Sloniewski & Reklewski, 2002; Hyslop, Ross, Schofield, Navajas,
77
Roehe & Simm, 2008) have been used as a means of assessing carcass characteristics at
78
slaughter.
79
More recently, the use of X-ray computed tomography (CT) in carcasses has been
80
investigated in pigs, sheep and beef cattle. CT scanning is a non-invasive technique that
81
can provide in vivo predictions of carcass composition, which are used in pig and sheep
82
breeding programmes (Simm, Lewis, Collins & Nieuwhof, 2001; Aass, Hallenstvedt,
83
Dalen, Kongsro & Vangen, 2009). Very accurate in vivo predictions of muscle, fat and
84
bone weight were reported in both species (sheep: Jones, Lewis, Young & Wolf, 2002;
85
Lambe, Young, Mclean, Conington & Simm, 2003; Macfarlane, Lewis, Emmans,
86
Young & Simm, 2006; pigs: Szabo, Babinszky, Verstegen, Vangen, Jansman & Kanis,
87
1999). Very accurate predictions of carcass tissue weights were also reported from the
88
CT scanning of carcasses of pigs (Dobrowolski, Romvari, Allen, Branscheid & Horn,
89
2003; Vester-Christensen et al., 2009) and sheep (Johansen, Egelandsdal, Røe, Kvaal &
90
Aastveit, 2007; Kongsro, Røe, Aastveit, Kvaal & Egelandsdal, 2008). In beef cattle,
91
although the size of the CT scanner gantry prevents CT scanning of live beef cattle or
92
whole carcasses, Navajas et al. (2010, in press) reported that it could be used as an
93
economical and faster alternative to total dissection for determining carcass composition
94
based on the scanning of primal cuts. This allows a non-invasive assessment of
95
composition without affecting the value of primal cuts. More comprehensive and faster
96
scanning is possible due to the development of CT technology, such as spiral CT
4
97
scanning (SCTS), which has been recently investigated in animal and meat science.
98
Predictions of beef and sheep carcass composition as well as muscle volume and
99
weights and muscularity in sheep, based on in vivo or post-slaughter SCTS, were found
100
to be very accurate (Navajas et al., 2006, 2007, 2010, in press). Although multivariate
101
analysis was used to predict sheep carcass composition from CT images (i.e. Johansen
102
et al., 2007; Kongsro, Røe, Kvaal, Aastveit & Egelandsdal, 2009), it has not been
103
applied for estimating beef carcass composition by SCTS.
104
The prediction of meat quality using CT scanning has been investigated based on
105
the average CT muscle density, calculated as the average values of the pixels segmented
106
as muscle in the CT images. In sheep, Karamichou, Richardson, Nute, McLean and
107
Bishop (2006a) found strong negative genetic correlations of CT muscle density with
108
IMF content and taste panel scores for flavour, juiciness and overall palatability;
109
although no genetic association with tenderness was identified. Associations of variable
110
magnitude were reported with different fatty acids (Karamichou, Richardson, Nute,
111
Gibson & Bishop, 2006b). A more sophisticated approach was used by Lambe, Jopson,
112
Navajas, McLean, Johnson and Bünger (2009) to quantify the association between CT
113
parameters and IMF in sheep. By fitting parameters of a mixture of four normal
114
overlapping distributions for the full tissue density the accuracy increased by 10%
115
compared to those using average muscle density.
116
Chemical and physical differences in the tissues between live animals and carcasses
117
are expected due to the post-mortem transformation process, particularly in the case of
118
muscle/meat (i.e. lower water content due to drip losses, differences in tissue density
119
because of low temperatures, histological differences due to ageing, etc.) (Lawrie,
120
1998). CT scanning of meat may capture the changes of tissue densities and properties
121
and therefore improve the predicting ability of CT data for both composition and quality
5
122
traits compared to measurements in the live animal. In the case of beef, moderate to low
123
phenotypic correlations were found between average CT muscle density of beef primals
124
and IMF in a preliminary study by Navajas et al. (2009). To the best of our knowledge,
125
there are no studies testing the use of SCTS to predict quality parameters of beef using a
126
multivariate analysis.
127
The aim of this study was to investigate, using a multivariate approach, the potential
128
of SCTS tissue density values as predictors of beef cuts composition and beef quality
129
characteristics in crossbred Aberdeen Angus and Limousin cattle. Beef quality traits
130
included in this study were technological parameters, eating quality traits, fatty acid
131
profile and intramuscular fat content.
132
2. Material and methods
133
2.1. Animals and management
134
This study was carried out as part of a larger trial in which a total of 88 Aberdeen
135
Angus (AAx) and 106 Limousin (LIMx) crossbred heifers and steers were slaughtered
136
in the autumn/winter months of 2006, 2007 and 2008. The AAx and LIMx animals had
137
average live weights of 582 and 609 kg and average ages at slaughter of 546 and 544
138
days, respectively.
139
Within the 194 animals, 144 animals were slaughtered in 2006 and 2007 and
140
produced within a two-breed reciprocal crossbreeding rotation using Aberdeen Angus
141
and Limousin breeds at the SAC Beef Research Centre (BRC). The 144 animals from
142
the SAC BRC were finished during the final 2-4 months of their production cycle on
143
similar diets consisting of 1st cut grass silage and a barley based concentrate (50:50 on a
144
dry matter basis) which was offered ad libitum as a completely mixed ration on a daily
145
basis. The ration analysis averaged 381 g.kg-1 dry matter (DM), 12.0 MJ.kg-1 DM
146
metabolisable energy and 139 g.kg-1 DM crude protein. All animals remained on these
6
147
diets for a minimum of eight weeks after which they were selected for slaughter
148
according to standard commercial practice (target grades R4L or better). The remaining
149
50 animals were slaughtered in 2008, sourced from different commercial farms and
150
sired by either Aberdeen Angus or Limousin sires but the breed of the dam was
151
unknown. These 50 animals were selected in the commercial abattoir where all
152
slaughtering took place on the basis of sire breed, sex and the fact that both farm of
153
origin was known and the individual sire identity was recorded on the animal passport.
154
Although the ration formulation was not known, their ages and slaughter dates suggest
155
that their finishing management was likely to be similar to that of the BRC animals.
156
2.2. Meat samples
157
After slaughter, the left carcass sides were kept and chilled for 48 h, until quartering
158
between the 10th and 11th ribs. After quartering, carcass sides were split into 20 primal
159
cuts, as illustrated in Figure 1. From the sirloins, two other cuts were obtained which
160
will be referred to as 11–12th rib sirloin and 13th rib sirloin, whilst the remaining
161
lumbar section of this cut will be referred to as lumbar sirloin. M. longissimus thoracis
162
et lumborum of these cuts was chosen for assessing all the traits in the present study
163
since most meat quality studies (e.g. Prieto et al., 2009) chose it for being the most
164
homogeneous and representative muscle of the carcass. Colour was measured after 45
165
min blooming on the 11–12th rib sirloin. Lumbar sirloins and 11–12th rib sirloins were
166
vacuum packed in the abattoir and transported to the SAC-BioSS CT unit in Edinburgh,
167
where they were CT scanned, and then sent to the University of Bristol for dissection
168
and meat quality analysis. Cuts were kept and transported at temperatures of 1-2 °C.
169
The 13th rib sirloins were not vacuum-packed as they were retained for textural slice
170
shear force (SSF) measurements that were taken at approximately 72 h after slaughter.
171
[Figure 1 near here, please]
7
172
After dissection, samples of the M. longissimus thoracis of the 11–12th rib sirloins
173
were vacuum packed and aged at 1 ºC to 14 days post-mortem for assessment by a
174
trained sensory panel. From the dissected M. longissimus lumborum of the sirloins, a 75
175
mm-long piece of the cranial end was separated, vacuum packed, aged for 10 days and
176
used to assess instrumental texture by Volodkevitch shear jaws. From an adjacent
177
section, 25 mm-thick steaks were vacuum packed and used to determine texture by a
178
second SSF, after an ageing period of 14 days. The next 25 mm of the lumbar sirloin
179
was taken, vacuum packed and frozen for subsequent analysis of fatty acid composition
180
and intramuscular fat content.
181
2.3. CT scanning and data
182
At the SAC-BioSS CT unit in Edinburgh, beef cuts were CT fully-scanned using a
183
Siemens Somatom Esprit scanner. The X-ray tube operated at 130 kV and 100 mAs,
184
using Pitch 2. The diameter of the CT images was 450 mm. CT scanning method was
185
SCTS, in which the X-ray tube rotates continuously in one direction whilst the table on
186
which the cuts were positioned is mechanically moved through the X-ray beam. The
187
transmitted radiation takes on the form of a helix or spiral (Jackson & Thomas, 2004).
188
This technology captures very detailed information from a volume of contiguous slices,
189
rather than by collecting individual cross-sectional images. SCTS were collected of each
190
of the sirloins with cross-sectional images that were 8 mm thick. A pilot trial was
191
carried out to evaluate the protocol and check that the sizes of the primal cuts were such
192
that effective and useful CT scanning was possible. Given the size of the primal cuts, a
193
thickness of 8 mm gave a good balance between the quality of the images and the time
194
required for the actual scanning. Furthermore, with this slice thickness, it was possible
195
to have one spiral sequence per cut for most of the cuts, and reduce the risk of
8
196
overheating the CT tube. The average numbers of cross-sectional images were 19 and
197
54 for the 11–12th rib and lumbar sirloins, respectively.
198
The principle of CT is based on the attenuation of X-rays through tissues and
199
objects depending on their different densities. These differences are reflected in
200
different CT values, which are measured in Hounsfield units (HU) (Hounsfield, 1992).
201
The frequency distributions of pixel values from -256 HU upwards were obtained for
202
these cuts using STAR 4.9 CT image analysis software (Mann, Glasbey, Navajas,
203
McLean & Bünger, 2008). Alternatively, the histogram of HU values could have been
204
obtained directly from the CT scanner or from any standard image processing software.
205
The pixel distribution values for the range of CT densities that correspond to the soft
206
tissues (fat and muscle) were considered in this study. The range was defined between -
207
254 and 133 HU using as reference values the CT tissue thresholds estimated as part of
208
the development of image analysis to predict carcass composition, described by Navajas
209
et al. (2010).
210
2.4. Dissection of sirloins
211
After CT scanning, beef cuts were transported to the University of Bristol where
212
they were dissected into subcutaneous, kidney knob and channel, intermuscular and
213
thoracic fat, muscle, cutaneous trunci, bone and ligaments. The composition traits
214
included in this study were the weights of subcutaneous fat, intermuscular fat, total fat
215
(in this case as subcutaneous fat plus intermuscular fat) and muscle (muscle plus
216
cutaneous trunci) of the 11–12th rib sirloin and lumbar sirloin.
217
2.5. Sensory analysis
218
Sensory analysis was carried out by a 10-person trained taste panel (BSI, 1993). The
219
samples were defrosted overnight at 4 ºC and then cut into steaks 20 mm thick. Steaks
9
220
were grilled to an internal temperature of 74 ºC in the geometric centre of the steak
221
(measured by a thermocouple probe) after which, all fat and connective tissue was
222
trimmed and the muscle cut into blocks of 2 cm3. The blocks were wrapped in pre-
223
labelled foil, placed in a heated incubator and then given to the assessors in random
224
order chosen by a random number generator. The assessors used 8-point category scales
225
to evaluate the following traits: tenderness (1 – extremely tough, 8 – extremely tender),
226
juiciness (1 – extremely dry, 8 – extremely juicy), beef flavour intensity (1 – extremely
227
weak, 8 – extremely strong), abnormal flavour intensity (1 – extremely weak, 8 –
228
extremely strong) and overall liking (1 – dislike very much, 8 – like very much).
229
2.6. Physical analyses
230
Meat colour as L* (lightness), a* (red-green) and b* (yellow-blue) (CIE, 1978) was
231
measured at 48 hours post mortem after blooming for 45 minutes, with a portable
232
Minolta® colorimeter (CM-2002, D45 illuminant and 10 º observer; Konica-Minolta
233
Sensing, Inc., Germany).
234
The slice shear force test was performed on hot cooked meat, according to
235
Shackelford, Wheeler and Koohmaraie (1999). Meat was cooked in a pre-warmed clam
236
shell grill (George Foreman brand) where temperature was monitored continuously,
237
using a stabbing temperature probe inserted into the geometric centre of the steak during
238
the cooking process, until it reached 71 ºC, when the steak was removed from the grill
239
and monitoring continued until temperature plateaued at approximately 76 ºC. The
240
weight before and after cooking was used for calculation of cooking loss. For this test, a
241
single meat sample of 50 mm by 10 mm was sheared orthogonal to muscle fibre
242
orientation and the maximum shear force noted. A Stevens CR Texture Analyser (Stable
243
micro-systems, UK) was used for 14 days test and a Lloyd Texture Analyser (Lloyd
244
Instruments, UK) for 72 hours test; both instruments equipped with a custom-designed
10
245
accessory, featuring a flat, blunt-end blade as described by Shackelford et al. (1999).
246
Particular care was taken to avoid fat or connective tissue at the point of shearing.
247
For the Volodkevitch shear force test, the samples were cooked in a water bath at 80
248
ºC until a centre temperature of 78 ºC was reached. From each of these cooked sections,
249
10 replicate blocks (20 x 10 x 10 mm) were cut parallel to the fibre direction and
250
sheared across the fibres with the Volodkevitch jaw (stainless steel probe shaped like an
251
incisor) on a Stevens CR Texture Analyser (Stable micro-systems, UK).
252
2.7. Fatty acid and intramuscular fat analyses
253
Fatty acids analysis was carried out by direct saponification as described in detail by
254
Teye, Sheard, Whittington, Nute, Stewart and Wood (2006). Samples were hydrolysed
255
with 2M KOH in water:methanol (1:1) and the fatty acids extracted into petroleum
256
spirit, methylated using diazomethane and analysed by gas liquid chromatography.
257
Samples were injected in the split mode, 70:1, onto a CP Sil 88, 50 m  0.25 mm fatty
258
acid methyl esters (FAME) column (Chrompack UK Ltd, London) with helium as the
259
carrier gas. The output from the flame ionization detector was quantified using a
260
computing integrator (Spectra Physics 4270) and linearity of the system was tested
261
using saturated (FAME4) and monounsaturated (FAME5) methyl ester quantitative
262
standards (Thames Restek UK Ltd, Windsor, UK). Total IMF content was calculated
263
gravimetrically as total weight of FA extracted.
264
2.8. Data analysis
265
The effect of breed cross (AAx or LIMx) on beef cuts composition and beef quality
266
traits was estimated using the general linear models (GLM) procedure of the SAS
267
package (SAS, 2003). The data were subjected to one way analysis of variance
268
according to the following model:
269
Yi = µ + Bi + εi,
11
270
where Yi was the dependent variable, µ was the overall mean, Bi was the fixed effect of
271
breed, εi was the random error and i was the number of observations.
272
Partial least square regression (PLSR) was used to carry out the prediction equations
273
where frequency distributions of pixel values of cross-sectional images from either the
274
11–12th rib sirloin or lumbar sirloin were used as predictor variables (X) and beef cuts
275
composition, fatty acids and meat quality characteristics as predicted variables (Y). The
276
specific cut used to estimate each trait was chosen according to the place where the
277
reference method was performed. In this sense, the frequency distribution of pixel
278
values of the 11–12th rib sirloins were used as predictor variables for predicting colour
279
and sensory parameters, and those from the lumbar sirloins were taken into account for
280
the instrumental texture, fatty acids and intramuscular fat content. CT information from
281
both cuts was used to predict dissection tissue weights.
282
For carcass tissues, pixel value segments may vary between and within animals
283
depending on the density and mixture of tissues (intramuscular fat within muscle).
284
Dobrolowski et al. (2003) reported a problem with adapting certain pixel values as
285
estimates for various body components in grading. This was explained by a non-exact
286
delimitation of muscle tissue density ranges due to influence of intramuscular fat. Using
287
multivariate calibration of dissected carcass tissues against the intensity histogram may
288
deal with these problems, yielding more correct ranges for tissues and more exact
289
estimations. Internal full leave-one-out cross-validation was performed in order to avoid
290
over-fitting the PLSR equations. Thus, the optimal number of factors in each equation
291
was determined as the number of factors after which the standard error of cross-
292
validation no longer decreased substantially. Calibration and validation were performed
293
using The Unscrambler program (version 8.5.0, Camo, Trondheim, Norway).
294
The predictive ability of the PLS calibration models was evaluated in terms of
12
295
coefficient of determination (R2) and root mean square error of cross-validation
296
(RMSECV). RMSECV is regarded as a measure of precision and accuracy of prediction
297
and is defined by:
298
RMSECV 
299
where n is the number of samples in the calibration set, the yi represents the real
300
(measured) responses and the y icv represents the estimated responses obtained via cross-
301
validation (Cederkvist, Aastveit & Naes, 2005).
302
3. Results and discussion
1 n  cv
 ( y  yi ) 2
n i 1 i
303
Ranges, means, standard deviations (SD) and coefficients of variation (CV) of the
304
parameters studied in AAx and LIMx samples are summarized in Table 1. Some
305
samples resulted in reference values considerably different to the rest of the population,
306
either the texture value because they were undercooked or the intramuscular fat content
307
which was not in agreement with the live weight and age of the animal; therefore they
308
were considered as errors of laboratory and outliers accordingly. Hence, four outliers
309
were deleted from the whole AAx beef sample population and three were eliminated
310
from LIMx data; thus samples from 187 animals (84 AAx and 103 LIMx) were used to
311
carry out the predictions.
312
[Table 1 near here, please]
313
Generally, the values of the parameters in both AAx and LIMx beef samples were
314
within the normal range of variation reported by other authors (Prieto et al., 2008;
315
Sierra, Aldai, Castro, Osoro, Coto-Montes & Oliván, 2008). Most carcass and meat
316
quality characteristics showed a large variability among samples with coefficients of
317
variation (CV = SD/average) higher than 20% in both AAx and LIMx samples; except
13
318
for muscle weight, L* and a* colour, sensory traits and PUFA which showed CV in the
319
range from 6.6 to 16.5% in both breed crosses. Statistically significant differences
320
(P<0.001) between the two breed crosses were observed in muscle weight and most FA
321
as well as IMF content (Table 1). However, non-significant differences between breed
322
crosses were found for PUFA, since on a standard diet the variability in supply of these
323
FA would be small and many PUFA, especially the longer chain PUFA, are in the
324
phospholipids fraction which does not vary much as the animal increase in fatness.
325
Tissue weights were heavier in the lumbar sirloins than in the 11–12th rib sirloins
326
due to the higher weight of the former cut (Table 1). Dissection data from both cuts
327
showed that LIMx carcasses tend to be leaner with higher muscle and lower
328
subcutaneous, intermuscular and total fat weights than AAx carcasses. A similar trend
329
was observed in the composition of the lumbar sirloin, where a higher concentration of
330
fatty acid was measured in AAx samples which could be a consequence of these
331
animals presenting greater total IMF content (P<0.001). The higher level of fat in AAx
332
is reflected in the Figure 2, which shows the average pixel frequency distribution for the
333
range of CT values of fat and muscle in the AAx and LIMx crossbred 11–12th rib
334
sirloin and lumbar sirloin. The first peak in the figure corresponds to fat and the second
335
peak to muscle. The AAx samples showed a higher frequency of pixels for the lower
336
CT values, which is representative of the fat densities, than the LIMx samples, the latter
337
showing higher within the muscle region. Additionally, the average pixel frequency
338
distribution of fat and muscle in both breeds was much higher for lumbar sirloin than
339
for 11–12th rib sirloin due to higher weights of the former cut.
340
[Figure 2 near here, please]
341
In a preliminary study we observed that when pixel frequency values for the CT
342
values of both muscle and fat tissue densities were used as predictor variables (X) to
14
343
estimate beef cuts composition, fatty acids and meat quality characteristics (predicted
344
variables, Y), the prediction equations were more accurate than when using only the
345
those representative of the muscle or the complete range (muscle, fat and bone) of CT
346
tissue densities as predictor. Thus, we used in the following analyses the pixel
347
frequency values for the CT values of the soft tissues (fat plus muscle). Nevertheless,
348
the differences in predictability using only the pixel histogram data for the muscle
349
density values were very small. Fitting sex, hot carcass weight and batch in the
350
prediction equations did not improve the accuracy compared to regression models that
351
only included the pixels within the range of CT density for fat and muscle as the only
352
predictors. Therefore, we took into account only the CT information as predictor
353
variables in all following analyses.
354
3.1. Tissue weights of cuts
355
As presented in Tables 2 and 3 for AAx and LIMx, respectively, the high coefficient
356
of determination and low RMSECV of the regression between tissue weights of cuts
357
obtained by dissection and the pixel histogram values within the range for soft tissues
358
showed a high accuracy of the multivariate approach to predict subcutaneous fat (R2,
359
RMSECV = 0.94, 34.60 g and 0.92, 34.46 g), intermuscular fat (R2, RMSECV = 0.81,
360
161.54 g and 0.86, 42.16 g), total fat (R2, RMSECV = 0.89, 65.96 g and 0.93, 48.35 g)
361
and muscle (R2, RMSECV = 0.99, 58.55 g and 0.97, 57.45 g) in AAx and LIMx cuts,
362
respectively. In general, the results were very similar for both breeds where the carcass
363
component predicted with the highest accuracy by CT was the muscle weight in both
364
sirloin cuts. This may be explained by the fact that muscle was the biggest component
365
of both sirloin cuts with lowest variation. In addition, the inclusion of the CT values that
366
correspond to both muscle and fat tissues as predictors may capture variations in muscle
15
367
due to the content of intramuscular fat, which may improve the accuracy of prediction
368
of muscle weight (Dobrolowski et al., 2003).
369
Figure 3 shows the regression coefficients from the PLSR for the estimation of
370
tissue weights using the CT values as predictors. The highest regression coefficients for
371
the predictions of the fat depots (subcutaneous, intermuscular and total) and muscle
372
were located within the range of CT values reported by Navajas et al. (2010) as the best
373
predictors of total carcass fat and muscle weights, respectively (fat: from -254 to 29
374
HU; muscle, from 30 to 133 HU).
375
[Figure 3 near here, please]
376
The accurate estimation of tissue weights in both AAx and LIMx cuts is in
377
agreement with those showed by Navajas et al. (2010, in press). The use of CT scanning
378
in beef was only recently investigated for the prediction of carcass tissue weights based
379
on SCTS scans collected for each primal cut. Values of R2 between 0.95 and 0.96 were
380
reported for the prediction of total carcass fat and muscle weights (Navajas et al., 2010,
381
in press).
382
Weights of fat depots of different cuts from in vivo CT scanning of sheep were
383
investigated by Kvame, McEwan, Amer and Jopson (2004), who reported ranges of
384
accuracies (R2) of 0.82 to 0.97 and 0.87 to 0.98, for intermuscular and subcutaneous fat,
385
respectively. Lambe et al. (2003) predicted total carcass intermuscular and subcutaneous
386
fat weights in sheep with R2 values of 0.95 and 0.96. Both studies predicted tissue
387
weights by fitting areas of the different tissues segmented in individual cross-sectional
388
images located at specific anatomical locations (reference scans, see Navajas et al.
389
2006).
390
Recent studies on total sheep carcass composition (Johansen et al., 2007; Kongsro et
391
al., 2009) obtained very accurate predictions of total carcass fat and muscle weights
16
392
using PLS as the method of analysis (R2 values fat 0.96, muscle 0.97, Johansen et al,
393
2007; fat 0.92, muscle 0.94, Kongsro et al., 2009). Very accurate prediction of lean
394
meat percentage was also reported for CT scanning of pig carcasses (R2: 0.9994, Vester-
395
Christensen et al., 2009).
396
When comparing our results with those obtained by alternative methods used in beef
397
to estimate or measure carcass composition such as ultrasound, TOBEC (total body
398
electrical conductivity) or VIA (video-image analysis), it is seen that the accuracy
399
obtained with CT was mostly substantially higher. For example, measurements of
400
ultrasound on the carcass of back fat showed R2 of 0.31 (May et al., 2000) and in the
401
live animals the saleable meat was predicted with R2 ranged from 0.58 to 0.84 (Greiner,
402
Rouse, Wilson, Cundiff & Wheeler, 2003; May et al., 2000; Realini et al., 2001). Allen
403
and McGeehin (2001) estimated the weights of lean by TOBEC with R2 of 0.78 and
404
Allen and Finnery (2001) reported correlations between saleable meat yield by
405
dissection and predicted by three VIA machines ranged between 0.80 and 0.82.
406
The results presented here indicate that CT scanning of vacuum packed cuts may
407
deliver very accurate information of their composition and, therefore, of beef carcass
408
composition. The method used requires the development of a prediction equation for
409
each CT scanned cut or primal cut, whilst Navajas et al. (2010) focused on the
410
composition of the entire carcass, therefore the method was developed with that
411
different objective. The predictions of total carcass tissue weights were based on the
412
estimations of the best thresholds for the CT tissue densities that maximised the
413
accuracy of the whole carcass composition (Navajas et al., 2010). Future studies should
414
compare the two alternatives in terms of accuracy, speed to deliver the data and costs,
415
considering as objectives total carcass composition and the composition of cuts or
416
primal cuts, given their different prices and meat quality attributes, using also
17
417
commercial de-boned primal cuts. Nevertheless, independent of the image analysis
418
methodology, it would be possible to obtain the composition data without damaging or
419
devaluing the CT scanned cuts or primal cuts, as they are CT scanned in vacuum packs
420
and kept at low temperatures.
421
3.2. Meat technological parameters and eating quality
422
In relation to the technological parameters, neither the colour values (L*, a* and b*;
423
R2 = 0.04-0.19, RMSECV = 1.81-2.44) nor the instrumental texture (SSF 3 d pm, SSF
424
14 d pm and Volodkevitch shear force; R2 = 0.03-0.26, RMSECV = 11.80-70.87 N)
425
could be predicted with any reasonable accuracy in AAx and LIMx samples. To the best
426
of our knowledge, there are no studies testing the ability of CT to predict these or
427
similar parameters in beef. In sheep, very low phenotypic correlations were reported by
428
Karamichou et al. (2006a) for the association of average CT muscle density with shear
429
force (r = -0.16) or colour parameters (r < 0.10). Similarly, the CT prediction ability for
430
sensory attributes was low in our study in both breeds with R2 and RMSECV from 0.01
431
to 0.17 and 0.45 to 0.78, in AAx and LIMx, respectively. Karamichou et al. (2006a)
432
showed low phenotypic correlations between toughness, flavour, juiciness and overall
433
liking and average CT muscle density (r = 0.15, -0.20, -0.16 and -0.29, respectively).
434
There are studies in the literature showing the difficulty of predicting both technological
435
and sensory parameters by means of other technologies. For example, Andrés et al.
436
(2007) showed low ability of NIR spectroscopy to predict sensory traits (R2 = 0.13-0.38)
437
in lamb and Prieto et al. (2008 and 2009a) found low to moderate NIR ability to
438
estimate instrumental texture (R2 = 0.17-0.54) in beef. It is worth noting that CT is a
439
secondary method, so that it is not independent of the disadvantages arising from the
440
reference method used for calibration such as the low precision of the reference method
441
(e.g. instrumental texture) or the subjectivity of the assessors when scoring the sensory
18
442
attributes in albeit scientifically-constructed consumer taste panels. Furthermore, a
443
narrow range of intensity scores for sensory attributes could reduce CT prediction
444
ability since it is necessary to have a wide range in the reference values to maximise CT
445
predictability.
446
3.3. Intramuscular fat and fatty acid contents
447
As far as the fatty acid content is concerned, the ability of CT to predict the most
448
abundant fatty acids was acceptable showing R2 ranges of 0.65-0.75 and 0.61-0.69 in
449
AAx and LIMx, respectively; with corresponding RMSECV from 83.79 to 245.25 and
450
79.38 to 235.52 mg.100 g-1 muscle. Among the groups of fatty acids, the sum of the
451
saturated fatty acid (SFA) and sum of the mono-unsaturated fatty acid (MUFA) content
452
were predicted with greater accuracy in AAx (R2 = 0.71 and 0.72, RMSECV = 281.59
453
and 318.36 mg.100 g-1 muscle, respectively) than in LIMx samples (R2 = 0.67 and 0.66,
454
RMSECV = 253.07 and 279.30 mg.100 g-1 muscle, respectively). In contrast, the CT
455
predictability for the sum of poly-unsaturated fatty acid (PUFA) content was less
456
reliable for both breeds (R2 = 0.26 and 0.09, RMSECV = 21.49 and 25.88 mg.100 g-1
457
muscle, AAx and LIMx, respectively), probably due to less variability in the sample
458
population (CV = 12.5 and 14.9%, AAx and LIMx samples, respectively, Table 1) and
459
much lower concentrations (about 5.8-7.4% of total fat). As animals mature and deposit
460
more fat, the relative proportion of PUFA decreases (Warren, Scollan, Enser, Hughes,
461
Richardson & Wood, 2008). The narrow ranges of concentration could be because of
462
PUFA are mainly located in membrane phospholipids, strictly controlled by a complex
463
system of enzymes and relatively constant between individuals (Scollan, Hocquette,
464
Nuernberg, Dannenberger, Richardson & Moloney, 2006). Karamichou et al. (2006b)
465
showed low phenotypic correlations between average CT muscle density, measured in
466
vivo, and individual fatty acids (r = 0.01-0.35) and groups of fatty acids (r = 0.26-0.35)
19
467
in sheep. Navajas et al. (2009) in preliminary study in beef indicated higher correlations
468
for SFA (r = 0.55-0.64) and MUFA (r = 0.55-0.64) in beef, but they were still lower
469
than those found in the present study. In this study, the predictions of fatty acids
470
composition were more accurate in AAx than in LIMx samples, which agree with the
471
highest correlations in AAx samples indicated by Navajas et al. (2009), probably due to
472
a higher concentration of fatty acids in the former breed.
473
The intramuscular fat content was successfully predicted by muscle and fat CT
474
values in the present study, with R2 of 0.76 and 0.71 and RMSECV of 567.39 and
475
539.15 mg FA.100 g-1 muscle in AAx and LIMx samples, respectively. This finding is
476
very important because IMF is a main contributing factor to eating quality traits like
477
juiciness and flavour (Wood et al., 2003). These results are slightly better than those
478
showed by Lambe et al. (2009) in lamb (R2 = 0.69), who fitted four overlapping normal
479
distributions. The results of the current study show higher accuracies than those found
480
by Karamichou et al. (2006a and b) and Navajas et al. (2009) (r = -0.57 and -0.55 to -
481
0.66, respectively) between IMF and average CT muscle density in sheep and beef
482
samples. It has to be pointed out that Karamichou et al. (2006a and b) and Navajas et al.
483
(2009) used the average CT muscle density as predictor variable. The approach of the
484
present study using a multivariate calibration method such as PLSR allowed all the
485
pixel values to be considered within the range of both muscle and fat tissues. Hence,
486
more information could be captured from the CT images, which may explain the most
487
reliable prediction of IMF content and fatty acid profile by SCTS. In the present study,
488
the spiral CT was defined by cross-sectional image that were 8 mm thick. Thinner
489
cross-sectional images may capture CT information of higher quality regarding tissue
490
composition which may increase the potential of CT information as predictor of beef
491
quality.
20
492
Comparing the regression coefficients from PLSR between the fatty acids and IMF
493
content (Y, predicted variables) and CT values (X, predictor variables) (Figure 3), it can
494
be observed that the highest coefficients were located within the range of CT values that
495
Navajas et al. (2010) reported as the best predictors of muscle tissue (30-133 HU). This
496
is reasonable given that the range of CT values used to predict muscle weights by
497
Navajas et al. (2010) included also the fat present in muscle tissue. The fatty acid profile
498
analysed in this study corresponded to the intramuscular fat. On the other hand, it is
499
likely that pixels frequency values at lower CT densities (first peak in Figure 2) were
500
more associated with the other fat depots. This may explain the lower regression
501
coefficients in this range of CT density values.
502
In the literature, other alternative technologies have been used to predict the fatty
503
acids and IMF content. For instance, NIR spectroscopy has successfully predicted the
504
IMF content in meat of different species (beef, sheep, pork and poultry), with most of
505
the studies reporting R2 from 0.90 to 0.99 for the prediction of this characteristic (Prieto,
506
Roehe, Lavín, Batten & Andrés, 2009b). Although these values of accuracy are above
507
those reported here and in other studies for CT, other factors such as the possibility of
508
providing a more comprehensive simultaneous assessment of carcass composition and
509
meat quality attributes by one method should also be considered when evaluating
510
measuring techniques.
511
CT scanning is one of the methods that show the highest accuracy as a predictor of
512
carcass composition in pigs, sheep and cattle. However, CT is regarded as an expensive
513
tool and the time required for carcass evaluation is somewhat higher than that of other
514
on-line methods, which make its application difficult at line speed with the current
515
technology. Although recent advances in CT scanning such as multi-slice scanning,
516
combined with spiral scanning can improve substantially the speed of CT scanners
21
517
(Kongsro et al., 2009), the utilization of CT scanning on-line is still a challenge,
518
particularly for beef carcasses. Nevertheless, it can be a very useful tool when used as
519
dissection reference and be of high value for the calibration of other faster on-line
520
methods or for genetic improvement of product quality.
521
In beef cattle breeding programmes, CT scanning can provide valuable post-
522
slaughter information on beef carcass composition and meat quality. The results of the
523
current study suggest that information on beef cuts composition can be complemented
524
with information on eating quality. The same raw CT information acquired for the
525
prediction of joint composition could be used for the prediction of intramuscular fat,
526
fatty acid composition and other meat quality attributes, given that the genetic
527
correlations are stronger than the phenotypic ones, as reported by Karamichou et al.
528
(2006a and b) in sheep. CT scanning can be valuable tool for the effective
529
implementation of the evaluation of selection candidates based on CT information
530
collected in relatives (i.e. progeny test) with lower total cost. In addition, it can be very
531
useful for the collection of carcass and meat quality data in large reference populations
532
required for genomic studies. Further studies comparing the genetic progress and costs
533
resulting from different methods would produce the optimal combination of methods
534
from a biological and economic perspective.
535
Conclusion
536
The results of this research show that multivariate analysis of SCTS of beef cuts
537
provides very accurate estimations of tissue weights in AAx and LIMx beef cuts.
538
Moreover, this image analysis approach has proven to yield accurate predictions for the
539
IMF content and its fatty acid composition (excepting PUFA content) in both breeds
540
without damaging or devaluing the cuts. The accuracy of prediction was higher in AAx
541
than in LIMx beef samples, probably due to a higher concentration of IMF in the
22
542
former. The reliability of SCTS predictions of technological and sensory parameters
543
was, however, very low. The CT predictions of beef cuts composition and meat quality
544
traits based on the same CT images and centralised image analysis procedures may be
545
valuable and cost-effective information for beef cattle breeding programmes. The use
546
multivariate analysis with the objective of predicting total beef carcass composition is
547
an approach that may also improve further the contribution of CT scanning, which has
548
not been explored yet.
549
Acknowledgements
550
We are grateful to the Scottish Government for funding the research and Scotbeef,
551
QMS, BCF and Signet for their substantial support. Also, we thank SAC colleagues
552
Kirsty McLean, Laura Nicoll, Claire Anderson, Mhairi Jack, Ann McLaren, Ruth Turl,
553
Elizabeth Goodenough, John Gordon, Lesley Deans, Alex Moir and Cameron Craigie
554
with their help in the experimental work; University of Bristol technical staff Duncan
555
Marriott, Anne Baker and Sue Hughes, for texture and sensory analysis, Bristol
556
colleagues Kathy Hallett, Fran Taylor and Fran Whittington for the fatty acid analysis
557
and the Bristol dissection team of Dave Brock, Jackie Bayntun, Carol Ebdon, Anne
558
Laws and Sally Osborne. N. Prieto is grateful to the Ministry of Science and Innovation
559
(MICINN), Spain, for financial assistance via a post-doctoral grant.
560
References
561
Aaslyng, M. D. (2009). Trends in meat consumption and the need for fresh meat and
562
meat products of improved quality. In J. P. Kerry, & D. Ledward, Improving the
563
sensory and nutritional quality of fresh meat. Cambridge: Woodhead Publishing.
564
Aass, L., Hallenstvedt, E., Dalen, K., Kongsro, J., & Vangen, O. (2009).
565
Datatomography (CT) as a method to measure meat and fat quality in live pigs. In
566
Proceedings of the 2nd workshop on the use of computer tomography (CT) in pig
23
567
carcass classification. Other CT applications: live animals and meat technology.
568
Monells, España. (http://www.recercat.net/handle/2072/39299?locale=es).
569
570
571
572
Allen, P., & McGeehin, B. (2001). Measuring the lean content of carcasses using
TOBEC. National Food Centre Research Report, Teagasc, 40, 1–19.
Allen, P., & Finnery, N. (2001). Mechanical grading of beef carcasses. National Food
Centre Research Report, Teagasc, 45, 1–26.
573
Andrés, S., Murray, I., Navajas, E. A., Fisher, A. V., Lambe, N. R., & Bünger, L.
574
(2007). Prediction of sensory characteristics of lamb meat samples by near infrared
575
reflectance spectroscopy. Meat Science, 76, 509–516.
576
577
BSI (1993). Assessors for sensory analysis. Guide to selection, training and monitoring
of selected assessors. BSI 7667. London: British Standards Institution.
578
Cederkvist, H. R., Aastveit, A. H., & Naes, T. (2005). A comparison of methods for
579
testing differences in predictive ability. Journal of Chemometrics, 19, 500–509.
580
CIE (1978). International Commission on Illumination, recommendations on uniform
581
colour spaces, colour, difference equations, psychometric colour terms. Paris: CIE
582
publication.
583
584
Clutter, A. C. (1995). Molecular genetics and meat quality. Iowa: National Swine
Improvement Federation.
585
Dobrowolski, A., Romvari, R., Allen, P., Branscheid, W., & Horn, P. (2003). X-ray
586
computed tomography as an objective method of measuring the lean content of a pig
587
carcass – A study in the framework of the European EUPIGCLASS project. In
588
Proceedings of the 43rd international congress of meat science and technology (pp.
589
371–372).
24
590
Greiner, S. P., Rouse, H. G., Wilson, D. E., Cundiff, L. V., & Wheeler, T. L. (2003).
591
Prediction of retail product weight and percentage using ultrasound and carcass
592
measurements in beef cattle. Journal of Animal Science, 81, 1736–1742.
593
Hyslop, J. J., Ross, D. W., Schofield, C. P., Navajas, E. A., Roehe, R., & Simm, G.
594
(2008). An assessment of the potential for live animal digital image analysis to
595
predict the slaughter liveweights of finished beef cattle. In Proceedings of the
596
British Society of Animal Science, 50, Scarborough, UK.
597
Hocquette, J. F., & Gigli, S. (2005). The challenge of quality. In J. F. Hocquette, & S.
598
Gigli, Indicators of milk and beef quality. The Netherlands: Wageningen Academic
599
Publishers.
600
Hounsfield, G. N. (1992). Computed medical imaging. In J. Lindsten, Nobel lectures in
601
physiology or medicine 1971-1980. Singapore: World Scientific Publishing
602
Company.
603
Jackson, S., & Thomas, R. (2004). Cross-sectional imaging. Livingstone: Churchill.
604
Johansen, J., Egelandsdal, B., Røe, M., Kvaal, K., & Aastveit, A. H. (2007). Calibration
605
models for lamb carcass composition analysis using Computerized Tomography
606
(CT) imaging. Chemometrics and Intelligent Laboratory Systems, 87, 303-311.
607
Karamichou, E., Richardson, R. I., Nute, G. R., McLean, K. A., & Bishop, S. C.
608
(2006a). Genetic analyses of carcass composition, as assessed by X-ray computer
609
tomography and meat quality traits in Scottish Blackface sheep. Animal Science, 82,
610
151-162.
611
Karamichou, E., Richardson, R. I., Nute, G. R., Gibson, K. P., & Bishop, S. C. (2006b).
612
Genetic analyses and quantitative trait loci detection, using a partial genome scan,
25
613
for intramuscular fatty acid composition in Scottish Blackface sheep. Journal of
614
Animal Science, 84, 3228–3238.
615
Kempster, A. J., Cuthbertson, A., & Harrington, G. (1982). Carcasse evaluation in
616
livestock breeding, production and marketing (p. 306). London: Granada Publishing
617
Limited.
618
Kempster, A. J., & Jones, D. W. (1977). Relationships between the lean content of
619
joints and overall lean content in steer carcasses of different breeds and crosses.
620
Journal of Agricultural Science, 88, 193–201.
621
Kongsro, J., Røe, M., Aastveit, A. H., Kvaal, K., & Egelandsdal, B. (2008). Virtual
622
dissection of lamb carcasses using computer tomography (CT) and its correlation to
623
manual dissection. Journal of Food Engineering, 88, 86-93.
624
Kongsro, J., Røe, M., Kvaal, K., Aastveit, A. H., & Egelandsdal, B. (2009). Prediction
625
of fat, muscle and value in Norwegian lamb carcasses using EUROP classification,
626
carcass shape and length measurements, visible light reflectance and computer
627
tomography (CT). Meat Science, 81, 102-107.
628
Kvame, T., McEwan, J. C., Amer, P. R., & Jopson, N. B. (2004). Economic benefits in
629
selection for weight and composition of lambs cuts predicted by computer
630
tomography. Livestock Production Science, 90, 123-133.
631
Jones, H. E., Lewis, R. M., Young, M. J., & Wolf, B. T. (2002). The use of X-ray
632
computer tomography for measuring the muscularity of live sheep. Animal Science,
633
75, 387-399.
634
Lambe, N. R., Young, M. J., Mclean, K. A., Conington, J., & Simm, G. (2003).
635
Prediction of total body tissue weights in Scottish Blackface ewes using computed
636
tomography scanning. Animal Science, 76, 191–197.
26
637
Lambe, N. R., Jopson, N. B., Navajas, E. A., McLean, K., Johnson, P. L., & Bünger, L.
638
(2009). In-vivo prediction of meat quality in lambs using CT scanning. In
639
Proceedings of the 2nd workshop on the use of computer tomography (CT) in pig
640
carcass classification. Other CT applications: live animals and meat technology.
641
Monells, España. (http://www.recercat.net/handle/2072/39299?locale=es).
642
643
Lawrie, R. A. (1998). Lawrie’s Meat Science. Cambridge: Woodhead Publishing
Limited.
644
Macfarlane, J. M., Lewis, R. M., Emmans, G. C., Young, M. J., & Simm, G. (2006).
645
Predicting carcass composition of terminal sire sheep using X-ray computed
646
tomography. Animal Science, 82, 289-300.
647
Mann, A. D., Glasbey, C. A., Navajas, E. A., McLean, K. A., & Bünger, L. (2008).
648
STAR: Sheep Tomogram Analysis Routines (Version 4.9). BioSS software
649
documentation.
650
May, S. G., Mies, W. L., Edwards, J. W., Harris, J. J., Morgan, J. B., Garrett, R. P.,
651
Williams, F. L., Wise, J. W., Cross, H. R., & Savell J. W. (2000). Using live
652
estimates and ultrasound measurements to predict beef carcass cutability. Journal of
653
Animal Science, 78, 1255–1261.
654
Navajas, E. A., Glasbey, C. A., McLean, K. A., Fisher, A. V., Charteris, A. J. L.,
655
Lambe, N. R., Bünger, L., & Simm, G. (2006). In vivo measurements of muscle
656
volume by automatic image analysis of spiral computed tomography scans. Animal
657
Science, 82, 545–553.
658
Navajas, E. A., Lambe, N. R., McLean, K. A., Glasbey, C. A., Fisher, A. V., Charteris,
659
A. J. L., Bünger, L., & Simm G. (2007). Accuracy of in vivo muscularity indices
27
660
measured by computed tomography and their association with carcass quality in
661
lambs. Meat Science, 75: 533–542.
662
Navajas, E. A., Glasbey, C. A., Fisher, A. V., Ross, D. W., Hyslop, J. J., Richardson, R.
663
I., Simm, G., & Roehe, R. (2010). Assessing beef carcass tissue weights using
664
computed tomography spirals of primal cuts. Meat Science, 84, 30-38.
665
Navajas, E. A., Richardson, R. I., Glasbey, C. A., Prieto, N., Ross, D. W., Hyslop, J. J.,
666
Simm, G., & Roehe, R. (2009). Associations between beef density by X-ray
667
computed tomography, intramuscular fat and fatty acid composition: preliminary
668
results. In Proceedings of the Bristish Society of Animal Science, 117, Southport,
669
UK.
670
Navajas, E. A., Richardson, R. I, Fisher, A. V., Hyslop, J. J., Ross, D. W., Prieto, N.,
671
Simm, G., & Roehe, R. Predicting beef carcass composition using tissue weights of
672
a primal cut assessed by computed tomography. Animal (In press).
673
674
Ngapo, T. M., Martin, J. F., & Dransfield, E. (2007). International preferences for pork
appearance: 1. Consumer choices. Food Quality and Preference, 18, 26-36.
675
Prieto, N., Andrés, S., Giráldez, F. J., Mantecón, A. R., & Lavín, P. (2008). Ability of
676
near infrared reflectance spectroscopy (NIRS) to estimate physical parameters of
677
adult steers (oxen) and young cattle meat samples. Meat Science, 79, 692–699.
678
Prieto, N., Roehe, R., Lavín, P., Batten, G., & Andrés, S. (2009b). Application of near
679
infrared reflectance spectroscopy to predict meat and meat products quality: a
680
review. Meat Science, 83, 175-186.
681
Prieto, N., Ross, D. W., Navajas, E. A., Nute, G. R., Richardson, R. I., Hyslop, J. J.,
682
Simm, G., & Roehe, R. (2009a). On-line application of visible and near infrared
28
683
reflectance spectroscopy to predict chemical-physical and sensory characteristics of
684
beef quality. Meat Science, 83, 96-103.
685
Realini, C. E., Williams, R. E., Pringle, T. D., & Bertrand, J. K. (2001). Gluteus medius
686
and rump fat depths as additional live animal ultrasound measurements for
687
predicting retail product and trimmable fat in beef carcasses. Journal of Animal
688
Science, 79, 1378–1385.
689
Sakowski, T., Sloniewski, K., & Reklewski, Z. (2002). Using digital image analysis and
690
ultrasound measurements for a pre-slaughter evaluation of carcass qualitative traits
691
in cattle. Animal Science Papers and Reports, 20, 111–123.
692
693
SAS (2003). SAS/STAT® User's Guide (Version 9.1). Cary, North Carolina: Statistical
Analysis System Institute, Inc.
694
Scollan, N., Hocquette, J. F., Nuernberg, K., Dannenberger, D., Richardson, I., &
695
Moloney, A. (2006). Innovations in beef production systems that enhance the
696
nutritional and health value of beef lipids and their relationship with meat quality.
697
Meat Science, 74, 17-33.
698
Shackelford, S. D., Wheeler, T. L., & Koohmaraie, M. (1999). Evaluation of slice shear
699
force as an objective method of assessing beef longissimus tenderness. Journal of
700
Animal Science, 77, 2693-2699.
701
Sierra, V., Aldai, N., Castro, P., Osoro, K., Coto-Montes, A., & Oliván, M. (2008).
702
Prediction of the fatty acid composition of beef by near infrared transmittance
703
spectroscopy. Meat Science, 78, 248-255.
704
Simm, G., Lewis, R. M., Collins, J. E., & Nieuwhof, G. J. (2001). Use of sire
705
referencing schemes to select for improved carcass composition in sheep. Journal of
706
Animal Science, 79, 225-259.
29
707
Szabo, C. S., Babinszky, L., Verstegen, M. W. A., Vangen, O., Jansman, A. J. M., &
708
Kanis, E. (1999). The application of digital imaging techniques in the in vivo
709
estimation of the body composition of pigs: a review. Livestock Production Science,
710
60, 1-11.
711
Teye, G. A., Sheard, P. R., Whittington, F. M., Nute, G. R., Stewart, A., & Wood, J. D.
712
(2006). Influence of dietary oils and protein level on pork quality. 1. Effects on
713
muscle fatty acid composition, carcass, meat and eating quality. Meat Science, 73,
714
157-165.
715
Vester-Christensen, M., Erbou, S. G. H., Hansen, M. F., Olsen, E. V., Christensen, L.
716
B., Hviid, M., Ersbøll, B. K., & Larsen, R. (2009). Virtual dissection of pig
717
carcasses. Meat Science, 81, 699-704.
718
Warren, H. E., Scollan, N. D., Enser, M. B., Hughes, S. I., Richardson, R. I., & Wood, J.
719
D. (2008). Effects of breed and a concentrate or grass silage diet on beef quality in
720
cattle of 3 ages. I. Animal performance, carcass quality and muscle fatty acid
721
composition. Meat Science, 78, 256-269.
722
Wood, J. D., Richardson, R. I., Nute G. R., Fisher, A. V., Campo, M. M., Kasapidou,
723
E., Sheard, P. R., & Enser, M. (2003). Effects of fatty acids on meat quality: a
724
review. Meat Science, 66, 21-32.
30
725
Table 1. Descriptive statistics for dissection components and meat quality traits in
726
Aberdeen Angus (AAx, n = 84) and Limousin (LIMx, n = 103) crossbred cattle.
AAx
l
LIMx
SDa CVb(%)
Range
Mean
Subcutaneous fat
145-510
319
94.6
Intermuscular fat
165-570
389
Total fat
310-1055
Muscle
SDa CVb(%)
Range
Mean
29.6
168-560
293
95.3
32.6
ns
103.4
26.6
220-640
371
82.3
22.2
ns
708
167.6
23.7
425-1075
664
161.3 24.3
ns
1080-2065
1480
232.3
15.7
1080-2395
Subcutaneous fat
340-1275
754
248.7
33.0
310-1030
659
192.2 29.2
*
Intermuscular fat
485-1636
1014
262.0
25.8
570-1370
901
190.1 21.1
*
Total fat
825-2794
1768
446.6
25.3
1005-2300
1560
329.7 21.1
*
Muscle
3145-7105
5669
694.9
12.3
4965-8297
6169
716.0 11.6 ***
L* colour
31.5-42.0
37.6
2.47
6.6
33.4-53.9
38.7
2.94
7.6
a* colour
16.7-27.2
23.7
2.16
9.1
19.7-28.9
23.9
2.02
8.4
**
ns
b* colour
3.4-11.9
8.4
1.79
21.4
4.4-15.4
8.5
1.93
22.8
ns
Slice shear force 3 d pm (N)
90.4-385.0
191.5
67.44
35.2
89.3-385.8
191.4 73.33 38.3
ns
Slice shear force 14 d pm (N)
61.4-273.3
122.7
35.11
28.6
62.6-242.1
127.2 34.04 26.8
ns
Volodkevitch shear force (N)
23.4-80.3
48.0
13.27
27.7
17.0-98.3
48.7 14.48 29.7
ns
Tenderness
3.0-6.5
4.8
0.73
15.3
3.2-6.7
4.8
0.69
14.2
ns
Juiciness
3.9-5.8
4.9
0.43
8.9
3.8-5.9
4.8
0.44
9.2
ns
Flavour
3.0-5.7
4.3
0.57
13.3
2.7-5.3
4.3
0.52
12.1
ns
C16:0 (palmitic)
248-1510
830
276.1
33.3
240-1519
653
C18:0 (stearic)
166-683
408
119.7
29.3
156-673
330
247.8 38.0 ***
110.5 33.5 ***
C18:1 (oleic)
346-2069
1183
401.9
34.0
336-1978
927
SFAc (saturated)
442-2325
1350
421.9
31.3
422-2339
1057
MUFA (monounsaturated)
415-2487
1441
471.3
32.7
423-2376
1119
PUFAe (polyunsaturated)
141-284
177
22.2
12.5
120-276
176
432.9 38.7 ***
26.3 14.9 ns
1186-5397
3229
968.0
30.0
1171-5405
2600
874.8 33.7 ***
Dissection components (g)
11–12th rib sirloin
1701 281.0 16.5 ***
Lumbar sirloin
Technological parameters
Sensory traits
Fatty acids (mg.100 g-1 muscle)
d
IMFf (mg FA.100 g-1 muscle)
727
728
a
363.9 39.3 ***
388.0 36.7 ***
Standard deviation, bcoefficient of variation, cC12:0 + C14:0 + C16:0 + C18:0, dC16:1 + Ct18:1 + C9c18:1 + C11c18:1 + C20:1, eC18:2n-6 + C18:3n-3 +
C20:3n-6 + C20:4n-6 + C20:5n-3 + C22:4n-6 + C22:5n-3 + C22:6n-3, fintramuscular fat, lsignificance: ns = P>0.05, * = P<0.05, ** = P<0.01, *** = P<0.001.
31
729
Table 2. Prediction of dissection components and meat quality characteristics in
730
Aberdeen Angus crossbred beef samples using the pixel distribution values for the range
731
of CT densities that correspond to the soft tissues (fat and muscle) of cross-sectional
732
images from cuts of the sirloin.
R2
p
RMSEC
RMSECV
Dissection components (g)
Subcutaneous fat(a/b)
5/3
0.94/0.87
23.61/87.81
34.60/101.20
Intermuscular fat(a/b)
3/6
0.77/0.81
48.63/115.23
59.25/161.54
Total fat(a/b)
3/3
0.89/0.86
55.29/164.60
65.96/194.38
Muscle(a/b)
7/4
0.99/0.93
24.23/129.92
58.55/171.96
L* colour
3
0.12
2.30
2.44
a* colour
1
0.04
2.10
2.18
b* colour
1
0.05
1.73
1.81
Slice shear force 3 d pm (N)
1
0.06
64.93
67.42
Slice shear force 14 d pm (N)
2
0.13
32.56
34.24
Volodkevitch shear force (N)
1
0.26
11.36
11.80
Tenderness
1
0.01
0.72
0.78
Juiciness
1
0.04
0.42
0.45
Flavour
1
0.05
0.56
0.58
C16:0
8
0.74
140.15
167.62
C18:0
8
0.65
70.83
83.79
C18:1
8
0.75
201.38
245.25
SFA
8
0.71
223.75
281.59
MUFA
8
0.72
248.08
318.36
PUFA
6
0.26
18.99
21.49
IMF (mg FA.100 g-1 muscle)
8
a
b
11–12th rib sirloin, lumbar sirloin.
0.76
471.99
567.39
Technological parameters
Sensory traits
Fatty acids (mg.100 g-1 muscle)
733
32
734
Table 3. Prediction of dissection components and meat quality characteristics in
735
Limousin crossbred beef samples using the pixel distribution values for the range of CT
736
densities that correspond to the soft tissues (fat and muscle) of cross-sectional images
737
from cuts of the sirloin.
R2
p
RMSEC
RMSECV
Dissection components(g)
Subcutaneous fat(a/b)
5/3
0.92/0.77
26.97/91.08
34.46/99.73
Intermuscular fat(a/b)
5/3
0.86/0.76
30.78/93.04
42.16/117.04
Total fat(a/b)
3/3
0.93/0.89
41.75/110.55
48.35/132.98
Muscle(a/b)
2/2
0.97/0.97
51.76/116.73
57.45/126.79
Technological parameters
L* colour
4
0.19
2.18
2.38
a* colour
4
0.18
1.81
1.92
b* colour
4
0.19
1.72
1.82
Slice shear force 3 d pm (N)
3
0.16
66.91
70.87
Slice shear force 14 d pm (N)
3
0.10
32.09
33.31
Volodkevitch shear force (N)
1
0.03
14.19
14.61
Tenderness
1
0.03
0.67
0.70
Juiciness
1
0.02
0.43
0.46
Flavour
4
0.17
0.47
0.50
C16:0
8
0.69
137.96
158.03
C18:0
8
0.61
68.66
79.38
C18:1
7
0.66
212.29
235.52
SFA
8
0.67
221.04
253.07
MUFA
7
0.66
250.13
279.30
PUFA
2
0.09
24.92
25.88
IMF (mg FA.100 g-1 muscle)
8
a
b
11–12th rib sirloin, lumbar sirloin.
0.71
468.52
539.15
Sensory traits
Fatty acids (mg.100 g-1 muscle)
33
738
Figure 1. Diagram showing the beef cuts used in this study. Full and dotted lines
739
indicate division between primal cuts and the subdivisions, respectively.
740
741
742
34
743
Figure 2. Average pixel frequency values for the range of computed tomography
744
densities corresponding to the soft tissues (fat and muscle) in 11–12th rib sirloins and
745
lumbar sirloins of Aberdeen Angus (AAx) and Limousin (LIMx) crossbred animals.
140000
AAx 11–12th rib sirloins
120000
LIMx 11–12th rib sirloins
Pixel frequency values
AAx lumbar sirloins
100000
LIMx lumbar sirloins
80000
60000
40000
20000
0
-254 -230 -206 -182 -158 -134 -110 -86
-62
-38
-14
CT values (Hounsfield units)
746
35
10
34
58
82
106 130
747
Figure 3. Regression coefficients between the dissection components and meat quality
748
characteristics in Limousin and Aberdeen Angus crossbred beef samples and the pixel
749
distribution values for the range of CT densities that correspond to the soft tissues (fat
750
and muscle) of cross-sectional images from cuts of the sirloin.
Intermuscular fat
Subcutanea fat
0.006
0.005
0.004
0.003
0.002
126
86
106
66
46
26
6
-14
-34
CT values (Hounsfield units)
Muscle
Total fat
0.012
0.012
0.010
0.010
0.008
0.008
0.006
0.006
0.004
0.004
0.002
0.002
126
106
86
66
46
26
6
-14
-34
106
106
126
86
88
106
66
66
86
46
46
70
26
6
-14
-34
-54
-94
26
CT values (Hounsfield units)
-114
-134
-154
-174
-194
-214
-234
-254
126
106
86
66
46
26
6
-14
-34
-54
-74
-94
-114
-134
-154
-174
-194
-214
0.012
0.010
0.008
0.006
0.004
0.002
0.000
-0.002
-0.004
-0.006
-0.008
-0.010
-74
C18:0
C16:0
-234
-54
CT values (Hounsfield units)
CT values (Hounsfield units)
-254
-74
-94
-114
-134
-154
-174
-194
-214
-0.004
-234
-0.002
-254
0.000
126
106
86
66
46
26
6
-14
-34
-54
-74
-94
-114
-134
-154
-174
-194
-214
-234
-254
0.000
0.030
0.025
0.020
0.015
0.010
0.005
0.000
-0.005
-0.010
-0.015
-0.020
-0.025
-54
-0.002
CT values (Hounsfield units)
-0.002
-74
-94
-114
-134
-154
-174
-194
-214
-234
-0.001
-254
0.000
126
86
106
66
46
6
26
-14
-34
-54
-74
-94
-114
-134
-154
-174
-194
-214
-234
0.001
-254
0.007
0.006
0.005
0.004
0.003
0.002
0.001
0.000
-0.001
-0.002
-0.003
CT values (Hounsfield units)
C18:1
SFA
0.04
0.04
0.03
0.03
0.02
0.02
0.01
0.01
126
6
-14
-34
-54
-74
-0.02
-0.02
-0.03
-0.03
-0.04
CT values (Hounsfield units)
CT values (Hounsfield units)
MUFA
PUFA
0.04
0.0015
0.03
-0.02
-0.0010
-0.03
-0.04
-0.0015
CT values (Hounsfield units)
CT values (Hounsfield units)
CT values (Hounsfield units)
124
106
88
70
52
34
16
-2
-20
-38
-56
-74
36
-92
-110
-128
-146
-164
-182
-200
-218
-236
0.10
0.08
0.06
0.04
0.02
0.00
-0.02
-0.04
-0.06
-0.08
-254
IMF
124
52
34
16
-2
-20
-38
-56
-74
-92
-110
-128
-146
-164
-182
-200
-218
-0.0005
-236
126
106
86
66
46
26
6
-14
-34
-54
-74
-94
-114
-134
-154
-174
-194
-214
0.0000
-234
0.0005
0.00
-254
0.01
-254
0.0010
0.02
-0.01
-94
-114
-134
-154
-174
-194
-214
-234
126
106
86
66
46
26
6
-14
-34
-54
-74
-94
-114
-134
-154
-174
-194
-214
-234
-254
-0.01
-0.01
-254
0.00
0.00