J. Anim. Sci. 2002. 80:3046-3052
© 2002 American Society of Animal Science
Prediction of intramuscular fat percentage in live swine using real-time ultrasound1
D. W. Newcom,
T. J. Baas2 and
J. F. Lampe
Department of Animal Science, Iowa State University, Ames 50011
2 Correspondence:
109 Kildee Hall (phone: 515-294-6728; fax: 515-294-5698; E-mail:
tjbaas{at}iastate.edu).
 |
Abstract
|
|---|
Purebred Durocs (n = 207) were used to develop a model to predict loin intramuscular fat percentage (PIMF) of the longissimus muscle in live pigs. A minimum of four longitudinal, real-time ultrasound images were collected 7 cm off-midline across the 10th to the 13th ribs on the live animal. A trained technician used texture analysis software to interpret the images and produce 10 image parameters. Backfat and loin muscle area were measured from a cross-sectional image at the 10th rib. After harvest, a slice from the 10th to the 11th rib loin interface was used to determine carcass loin intramuscular fat percentage (CIMF). The model to predict loin intramuscular fat percentage was developed using linear regression analysis with CIMF as the dependent variable. Initial independent variables were off-test weight, live animal ultrasonic 10th rib backfat and loin muscle area, and the 10 image parameters. Independent variables were removed individually until all variables remaining were significant (P < 0.05). The final prediction model included live animal ultrasound backfat and five image parameters. The multiple coefficient of determination and root mean square error for the prediction model were 0.32 and 1.02%, respectively. An independent data set of Duroc (n = 331) and Yorkshire (n = 288) pigs from two replications of the National Pork Board’s Genetics of Lean Efficiency Project were used for model validation. Results showed the Duroc pigs provided the best validation of the model. The product moment correlation and rank correlation coefficients between PIMF and CIMF were 0.60 and 0.56, respectively, in the Duroc population. Results show real-time ultrasound image analysis can be used to predict intramuscular fat percentage in live swine.
Key Words: Carcass Quality • Pigs • Ultrasound
|
Introduction
|
|---|
Greater levels of loin intramuscular fat (IMF), or marbling, have been associated with improved customer acceptance of the pork longissimus dorsi muscle (NPPC, 1995). Levels of IMF below 2.5% have been associated with poorer eating quality traits (Wood, 1985; Enser and Wood, 1991). With the evolution of niche and value-added markets, improved IMF percentage could be one of the criteria used in evaluation of specialty meat products. Accurate and reliable estimates of IMF percentage in breeding stock are needed to increase genetic improvement and to meet the specifications of these niche markets. However, progeny and sib-testing programs have been the only means to evaluate IMF percentage; both are costly in terms of time and money.
The use of real-time ultrasound to estimate backfat thickness and loin muscle area has been well documented in both swine and beef cattle research (Houghton and Turlington, 1992). Use of this technology has helped producers develop leaner, more muscular animals. Selection for increased lean meat content has caused a reduction in IMF percentage (Schwörer et al., 1995) and, ultimately, has had a detrimental impact upon the eating quality traits of pork (Barton-Gade, 1990).
Researchers at Iowa State University have developed image collection and interpretation equipment and prediction equations to estimate IMF percentage in live cattle (Amin et al., 1997; Hassen et al., 2000; Wilson et al., 2001). These equations are currently being utilized by 16 beef breed associations, including Angus, which has the highest number of animals registered. Previous research with swine has shown that estimation of IMF in the live pig is possible using real-time ultrasound (Ragland, 1998; Newcom et al., 2001). Therefore, the objective of this study was to develop and validate a model to predict loin IMF percentage in live swine.
|
Materials and Methods
|
|---|
Model Development
Developmental data were obtained from purebred Duroc barrows and gilts born in the spring of 2000 (n = 67) and 2001 (n = 140) at the Iowa State University Bilsland Memorial Swine Breeding Farm. Pigs were weighed off-test and scanned 5 d prior to harvest with an Aloka 500V SSD ultrasound machine fitted with a 3.5 MHz, 12.5-cm linear-array transducer (Corometrics Medical Systems, Inc., Wallingford, CT). Gain settings for the Aloka ultrasound machine were: Overall, 90; Near, -25; and Far, 2.1. Focal lengths were set at 1 and 2. Off-midline backfat and loin muscle area were measured from a cross-sectional image taken at the 10th rib. A sound transmitting guide (Superflab, Mick Radio Nuclear Instruments, Inc., Bronx, NY) conforming to the pigs’ back was attached to the ultrasound probe and vegetable oil was used as conducting material between the probe and skin.
A minimum of four longitudinal images (Figure 1
) were collected 7 cm off-midline across the 10th to 13th ribs, digitized, and saved to a computer for later interpretation. The probe was used without a guide, and vegetable oil was again used as a couplant. A trained technician used texture analysis software (Amin et al., 1997) to define Fourier, gradient, histogram, and co-occurrence parameters (Table 1) (Hassen et al., 2001) within a defined region of interest for each ultrasound image. The region of interest was a 100- x 100-pixel region placed as close to the 10th to 11th rib interface as possible. The technician made a visual assessment of each image for acceptability. After harvest, a slice of the longissimus muscle from the 10th to 11th rib interface was analyzed for carcass IMF percentage by the method outlined in Bligh and Dyer (1959).
View larger version (131K):
[in this window]
[in a new window]
|
Figure 1. Example ultrasound image used for prediction of intramuscular fat percentage. aFat layers. bTrapezius muscle. cRegion of interest: 100 x 100 pixel area, 10 image parameters generated from this region. d10th rib.
|
|
Image parameters from the longitudinal images were averaged by animal as recommended by Hassen et al. (1999) and included with off-test weight, backfat, and loin muscle area as independent variables in the IMF percentage prediction model development. The REG procedure of SAS (SAS Inst., Inc., Cary, NC) was used to perform a linear regression with carcass IMF percentage as the dependent variable. Independent variables (image parameters, off-test weight, backfat, loin muscle area) with the highest P-value were removed individually until all variables remaining were significant (P < 0.05).
Model Validation
Purebred Duroc and Yorkshire barrows and gilts (n = 619) from two replications of the National Pork Board’s Genetics of Lean Efficiency Project were weighed off-test and scanned 5 d prior to harvest. Cross-sectional and longitudinal images were collected in the same manner as in the developmental data. Model validation procedures were completed using three data sets: Validation 1 (use of all data), Validation 2 (use of Duroc data only), and Validation 3 (use of Yorkshire data only).
Models were analyzed for predictive ability using the difference between predicted and carcass IMF percentage (predicted - carcass), the absolute difference between predicted and carcass IMF percentage, the standard error of prediction (SEP), and both Pearson product moment (PCOR) and Spearman rank (RCOR) correlations between predicted and carcass IMF percentage (Hassen et al., 2001). Regression of carcass on predicted IMF percentage was used to test how well the model fit the validation data. The slope and intercept of these lines were expected to be 1.00 and 0.00, respectively.
where predicted = IMF predicted from ultrasound; carcass = chemical IMF from longissimus muscle; bias = mean difference between predicted and carcass IMF percentage.
Models were also analyzed for ability to correctly classify animals within a carcass IMF percentage class (not subjective marbling score). Each animal was classified by carcass IMF percentage into classes defined as Class 1 (
2.0%), Class 2 (> 2.0% and 3.0%), Class 3 (> 3.0% and 4.0%), Class 4 (> 4.0% and 5.0%), Class 5 (> 5.0% and 6.0%), and Class 6 (> 6.0%). Predicted IMF percentage values were classified in the same manner.
|
Results and Discussion
|
|---|
Developmental Data
Descriptive statistics for the developmental data are shown in Table 2. The mean carcass IMF percentage for the developmental data was 3.76%, and values ranged from 1.39 to 8.14%. However, most of the pigs (82%) were between 2.0 and 5.0%. The final model to predict IMF percentage in live pigs included 10th-rib off-midline backfat and five image parameters. Multiple coefficients of determination and root mean square error (RMSE) for the prediction model were 0.32 and 1.02%, respectively. The PCOR and RCOR between carcass and predicted IMF percentage for the developmental data were 0.56 and 0.55, respectively. Regression of carcass on predicted IMF percentage (Figure 2) showed an intercept of 0.00 and a slope of 1.00.
View larger version (23K):
[in this window]
[in a new window]
|
Figure 2. Regression of carcass on predicted intramuscular fat percentage from developmental data. aDevelopmental data = purebred Durocs from Iowa State University Swine Breeding Farm. bPIMF = predicted intramuscular fat percentage. cCIMF = carcass intramuscular fat percentage.
|
|
Results agree with Ragland (1998), who evaluated several models of prediction, including various combinations of all image parameters or only significant image parameters and backfat and loin muscle area measured by real-time ultrasound from nine breeds of pigs. He reported R2 values of 0.28 to 0.48 and RMSE values of 0.88 to 1.03% from his developmental data. Hassen et al. (2001) reported R2 and RMSE values of 0.69 to 0.72 and 0.84 to 0.91%, respectively, for prediction models in beef cattle from images collected using an Aloka 500V machine (Corometrics Medical Systems, Inc.).
Validation Data
Results from the model validation are shown in Table 3. The mean carcass IMF percentages for Validations 1, 2, and 3 were 2.79, 3.37, and 2.12%, respectively. The carcass IMF values across the validation data ranged from 0.88 to 8.50. The mean difference between predicted and carcass IMF percentage from Validations 1, 2, and 3 was 1.18, 0.75, and 1.68%, respectively, indicating that the prediction model gives a predicted value that, on average, overestimates carcass IMF percentage in all three validation data sets. The predicted IMF percentage also had less variability across the validation data sets than did carcass IMF, which was expected because regression procedures tend to regress predicted values toward the mean. The difference in mean carcass IMF values between the developmental data and the validation data ranged from 0.39 to 1.64%, which could be the cause of this overprediction. The mean absolute difference ranged from 0.91 to 1.74%. Ragland (1998) reported mean absolute differences between carcass and predicted IMF percentage ranging from 0.63 to 0.73, depending on the parameters in the model. Hassen et al. (2001) reported values of 0.42 and 0.83% for the mean difference and absolute difference between predicted and carcass IMF percentage, respectively, from four prediction models tested in beef cattle.
Standard errors of prediction, PCOR and RCOR are shown in Table 4. The SEP takes into account the mean difference between predicted IMF and carcass IMF (bias) and indicates that the predicted IMF percentage is within the reported SEP value of the carcass IMF percentage for 67% of the observations. The SEP for the validation data ranged from 0.80 to 0.93%, which means that after adjusting for the bias, the predicted value is within 0.80 to 0.93% of the carcass IMF percentage, 67% of the time. This is similar to the SEP from Hassen et al. (2001), who reported a value of 0.84%.
View this table:
[in this window]
[in a new window]
|
Table 4. Standard errors of prediction, Pearson product moment correlations, and Spearman rank correlation coefficients from validation dataa
|
|
Correlations between carcass IMF and predicted IMF were moderate (Table 4
). The PCOR for Validations 1, 2, and 3 were 0.56, 0.60, and 0.46, respectively, and for RCOR were 0.55, 0.56, and 0.55, respectively. Ragland (1998) reported PCOR and RCOR values ranging from 0.52 to 0.71 and 0.57 to 0.70, respectively. Hassen et al. (2001) reported PCOR and RCOR values of 0.88 and 0.88, respectively. The differences observed between the beef and swine results could be due to differences in machine settings or hide thickness, which changes the speed of the sound waves going into and coming out of the animal, creating a darker image. Darker images may cause differences in the image parameters used in the prediction model.
Regression of carcass IMF on predicted IMF (Figures 3 to 5) showed y-intercepts that ranged from -0.09 to -0.73, with regression coefficients from 0.58 (Validation 3) to 0.89 (Validation 1). Two outliers (high carcass IMF, low predicted IMF) in Figures 3 and 5 slightly skew the regressions. Removal of the two outliers changed the R2 from 0.31 to 0.36 in Validation 1 and from 0.22 to 0.36 in Validation 3 (data not shown). Ragland (1998) reported intercepts that ranged from 0.08 to 0.40 and slopes from 0.82 to 0.96, values similar to those reported in the present study. The R2 values in the current study ranged from 0.22 (Validation 3) to 0.36 (Validation 2). These values are similar to those found by Ragland (1998), who reported R2 values from 0.27 to 0.50.
View larger version (22K):
[in this window]
[in a new window]
|
Figure 3. Regression of carcass on predicted intramuscular fat percentage from validation data set 1. aVAL 1 = validation data from National Pork Board’s Genetics of Lean Efficiency Project, Duroc and Yorkshire pigs. bPIMF = predicted intramuscular fat percentage. cCIMF = carcass intramuscular fat percentage.
|
|
View larger version (19K):
[in this window]
[in a new window]
|
Figure 5. Regression of carcass on predicted intramuscular fat percentage from validation data set 3. aVAL 3 = validation data from National Pork Board’s Genetics of Lean Efficiency Project, Yorkshire pigs only. bPIMF = predicted intramuscular fat percentage. cCIMF = carcass intramuscular fat percentage.
|
|
The frequency distribution of the mean absolute difference between predicted and carcass IMF values across the validation data sets is shown in Table 5. The percentage of pigs for which predicted IMF was within 0.5% of carcass IMF varied across validation data sets, from 6 (Validation 3) to 39% (Validation 2). The percentage of pigs predicted within 1.0% of their carcass IMF ranged from 17% in Validation 3 to 65% in Validation 2. A greater proportion of predicted IMF values within ±1% of the carcass IMF value may have been expected as Validation 2 contained pigs of the same breed and having a similar mean carcass IMF percentage as data used in model development. The pigs in Validation 3 were Yorkshires only and had a lower mean carcass IMF when compared to Durocs only. Ville et al. (1997), using a 4 MHz Piglog 105 (SFK Technology, Soborg, Denmark), found the regression to predict IMF from image analysis from pigs at 60 and 100 kg liveweight was not significant. However, they did find the calibration equation used by their ultrasound machine overestimated IMF by 1%.
View this table:
[in this window]
[in a new window]
|
Table 5. Frequency of the absolute difference between predicted and carcass intramuscular fat percentage from validation dataa
|
|
Ragland (1998) reported estimated IMF was within 0.50% of carcass IMF 44 to 51% of the time, depending on the model. Predicted IMF was within 1.0% of carcass IMF in 72 to 86% of the observations. The mean IMF value from the data used to develop his model was less than the mean carcass IMF value observed in the present study (2.39 vs 3.76). Hassen et al. (2001) reported that the cumulative frequency of cattle predicted within 0.50, 0.75, and 1.0% of their carcass IMF value was 34, 54, and 71%, respectively.
Distribution of classes for carcass IMF and predicted IMF for the three validation data sets is shown in Table 6. The percentage of pigs falling into Classes 1, 2, and 3 for carcass IMF ranged from 76 to 97%, but the percentage of pigs falling into Classes 1, 2, and 3 for predicted IMF ranged from 42 to 63%. The percentage of pigs in the same class for both carcass IMF and predicted IMF was 19% (117/622), 31% (104/332), and 5% (13/288) for Validations 1, 2, and 3, respectively. The percentage classified into adjacent classes was 41% (252/622), 47% (157/332), and 33% (94/288) for Validations 1, 2, and 3, respectively. Of the pigs not correctly classified by more than one class, the percentage from carcass IMF Classes 1, 2, and 3 was 97%, 93%, and 99% for Validations 1, 2, and 3, respectively. These results demonstrate the overprediction of animals with a lower carcass IMF value. These results also show the underprediction of animals with a greater carcass IMF value (> 5.0). The validation data lacked a significant number of pigs with greater carcass IMF values (> 5.0%), which may have contributed to this underprediction.
View this table:
[in this window]
[in a new window]
|
Table 6. Distribution of classes for carcass intramuscular fat percentage (CIMF) and predicted intramuscular fat percentage (PIMF) from validation dataa
|
|
|
Implications
|
|---|
The results of this study indicate that estimation of intramuscular fat percentage in the live pig using real-time ultrasound is feasible. The ability to measure intramuscular fat in the live pig will allow identification of superior breeding animals and the use of intramuscular fat in single- or multiple-trait selection programs. Research to refine or enhance the current ultrasound intramuscular fat prediction model will be necessary as advances in image analysis capabilities and ultrasound technologies occur. Future model development and validation must evaluate populations with both lesser and greater amounts of loin intramuscular fat percentage and be tested across different breeds or genetic lines.
View larger version (21K):
[in this window]
[in a new window]
|
Figure 4. Regression of carcass on predicted intramuscular fat percentage from validation data set 2. aVAL 2 = validation data from National Pork Board’s Genetics of Lean Efficiency Project, Duroc pigs only. bPIMF = predicted intramuscular fat percentage. cCIMF = carcass intramuscular fat percentage.
|
|
|
Footnotes
|
|---|
1 Journal paper no. J-19800 of the Iowa Agric. and Home Econ. Exp. Sta., Ames, IA, Project no. 3456, and supported by Hatch Act and State of Iowa funds. 
Received for publication April 9, 2002.
Accepted for publication July 11, 2002.
|
Literature Cited
|
|---|
Amin, V., D. Wilson, and G. Rouse. 1997. USOFT: An ultrasound image analysis software for beef quality research. A. S. Leaflet R1437. Iowa State University, Ames.
Barton-Gade, P. A. 1990. Pork quality in genetic improvement programmes—the Danish experience. Proc. Natl. Swine Improv. Fed. Ann. Mtg. Des Moines, IA.
Bligh, E. G., and W. J. Dyer. 1959. A rapid method for total lipid extraction and purification. Can. J. Biochem. Physiol. 37:911–917.[Medline]
Enser, M., and J. D. Wood. 1991. Factors controlling fat quality in pigs. Page 32 in EAAP Publ. No. 42, Berlin.
Hassen, A., G. H. Rouse, D. E. Wilson, A. Trenkle, R. L. Willham, D. Beliele, and C. Crawley. 1996. Validation of a model for prediction of percent intramuscular fat on live feedlot cattle. A. S. Leaflet R1328. Iowa State University, Ames.
Hassen, A., D. Wilson, V. Amin, G. Rouse, and C. Hays. 2000. Predicting percentage of intramuscular fat using two types of real-time ultrasound equipment. A. S. Leaflet R1732. Iowa State University, Ames.
Hassen, A., D. E. Wilson, V. Z. Amin, G. H. Rouse, and C. L. Hays. 2001. Predicting percentage of intramuscular fat using two types of real-time ultrasound equipment. J. Anim. Sci. 79:11–18.[Abstract/Free Full Text]
Hassen, A., D. E. Wilson, V. R. Amin, and G. H. Rouse. 1999. Repeatability of ultrasound-predicted percentage of intramuscular fat in feedlot cattle. J. Anim. Sci. 77:1335–1340.[Abstract/Free Full Text]
Houghton, P. L., and L. M. Turlington. 1992. Application of ultrasound for feeding and finishing animals: A review. J. Anim. Sci. 70:930–941.[Abstract]
NPPC. 1995. Genetic Evaluation/Terminal Line Program Results. R. Goodwin and S. Burroughs, ed. Natl. Pork Producer’s Counc., Des Moines, IA.
NPPC. 2000. Pork Composition and Quality Assessment Procedures. E. P. Berg, ed. Natl. Pork Producer’s Counc., Des Moines, IA.
Newcom, D., A. Hassen, T. J. Baas, D. E. Wilson, G. H. Rouse, and C. L. Hays. 2001. Prediction of percent intramuscular fat in live swine. J. Anim. Sci. 79 (Suppl. 2):8 (Abstr.).
Schwörer, D. A., A. Rebsamen, and D. Lorenz. 1995. Selection of intramuscular fat in Swiss pig breeds and the importance of fatty tissue quality. Proc. 2nd Dummerstorf Muscle Workshop on Growth and Meat Quality, Rostock.
Ragland, K. D. 1998. Assessment of intramuscular fat, lean growth, and lean composition using real-time ultrasound. Ph.D. Diss. Iowa State University, Ames.
Villé, H., G. Rombouts, P. Van Hecke, S. Perremans, G. Maes, G. Spincemaille, and R. Geers. 1997. An evaluation of ultrasound and nuclear magnetic resonance spectroscopy to measure in vivo intramuscular fat content of longissimus muscle of pigs. J. Anim. Sci. 75:2942–2949.[Abstract/Free Full Text]
Wilson, D. E., G. H. Rouse, and A. Hassen. 2001. Ultrasound prediction model for % intramuscular fat in beef cattle. A. S. Leaflet R1756. Iowa State University, Ames.
Wood, J. D. 1985. Consequences of changes in carcass composition on meat quality. Page 157 in Recent Advances in Animal Nutrition. W. Haresign and D. J. A. Cole, ed. Butterworths, London.
This article has been cited by other articles:
|
|
|
|
 
S. M. Moore, K. J. Stalder, D. C. Beitz, C. H. Stahl, W. A. Fithian, and K. Bregendahl
The correlation of chemical and physical corn kernel traits with growth performance and carcass characteristics in pigs
J Anim Sci,
March 1, 2008;
86(3):
592 - 601.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
C. R. Schwab, T. J. Baas, K. J. Stalder, and J. W. Mabry
Deposition rates and accretion patterns of intramuscular fat, loin muscle area, and backfat of Duroc pigs sired by boars from two time periods
J Anim Sci,
June 1, 2007;
85(6):
1540 - 1546.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
D. W. Newcom, T. J. Baas, K. J. Stalder, and C. R. Schwab
Comparison of three models to estimate breeding values for percentage of loin intramuscular fat in Duroc swine
J Anim Sci,
April 1, 2005;
83(4):
750 - 756.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
D. W. Newcom, T. J. Baas, C. R. Schwab, and K. J. Stalder
Genetic and phenotypic relationships between individual subcutaneous backfat layers and percentage of longissimus intramuscular fat in Duroc swine
J Anim Sci,
February 1, 2005;
83(2):
316 - 323.
[Abstract]
[Full Text]
[PDF]
|
|
|
Top
Abstract
Introduction
