Predicting primal weight and primal yield in pork carcasses from a large-scale survey at major meat processing centers in Japan
Abstract
Pork primal weight and primal yield are important indicators for pig breeding, feeding management, commercial distribution systems, and meat processing. Here, we aimed to determine whether primal weight and primal yield could be predicted through non-destructive measurements of pork carcass traits. A total of 4397 carcasses (1958 gilts and 2439 barrows) from eight major meat processing centers were used, and the mean primal weight and primal yield were 56.0 kg and 73.9%, respectively. Significant sex differences were observed for all primal and carcass traits (P < 0.001), except for carcass weight. A maximum of 12 variables were examined, and primal weight was predicted with very high accuracy (R = 0.95, RMSE = 1.7, RPD = 3.0) using four variables. Primal yield was predicted with relatively good accuracy (R = 0.71, RMSE = 2.3, RPD = 1.4) using three variables, and these same variables were also effective for predicting primal weight. These prediction formulas were sufficiently accurate without accounting for the effect of sex. Overall, our results demonstrate that primal weight and primal yield can be accurately predicted using four variables, “carcass weight,” “backfat thickness above M. gluteus medius,” “spinous process length of 13th thoracic vertebra,” and “length from 1st thoracic vertebra to backfat,” without accounting for the effect of sex.
1 INTRODUCTION
In Japan, growing-finishing pigs are slaughtered and dismembered at slaughterhouses before being processed into carcasses. Pork carcasses are divided into primals by meat wholesalers and meat processors and then sold to restaurants, mass merchandisers, and meat retailers for butchering. The value of a pig carcass is determined by its weight, degree of fat attachment and deposition, muscle accumulation, and the quality of fat and muscles. Because Japanese consumers often eat pork with lean meat and fat together (Okumura et al., 2022), a moderate amount of fat attached to the muscles is an important edible part of meat products. However, carcasses with excessive fat deposition are low-grade, and excess fat is trimmed when the carcass is cut into primals.
Pork is principally distributed in Japan in the form of primal cuts (Figure 1a) (Kaku et al., 2002). The yield rate of primal cuts to retail meat is as high as 90% (Ministry of Agriculture, Forestry and Fisheries, 2024), which means that most of the primal cuts comprise edible parts. Therefore, the weight of pork primal cuts and the ratio of primal weight to carcass weight (i.e., the primal yield) are important indicators for commercial transactions and production efficiency. In some of the large-scale meat processing centers in Europe using AutoFOM (Frontmatec, Denmark), the muscle and fat content in the carcass is automatically calculated, and their values are used for grading (Brøndum et al., 1998; Danish Agriculture & Council, 2021). However, in Japan, meat amount indicators such as the yield percentage and loin area are used as grading items only for cattle, and these items are being promoted in breeding selection for beef cattle (MAFF, 2020), but for pigs, carcass weights and backfat thicknesses are the only indicators for estimating meat amount in current state. The reason for this is that pork carcasses are generally not cut in cross-section (Japan Meat Grading Association, 2016); thus it is impossible to measure loin area or other objective parameters to assess the edible meat amount in the carcass. As a result, carcasses of the same weight can vary substantially in the same weight of the retail cuts dissected from them (Gardner et al., 2021). Therefore, it would be very useful in Japan if the parameters for evaluating the amount of meat in carcasses could be incorporated into the grading.
When evaluating meat amount in pork carcasses, the most important traits are primal weight and primal yield, which are considered to be closely related to the amount of edible meat (MAFF, 2024). Accurate predictions of primal weight and yield could be easily and inexpensively predicted on the basis of carcass characteristics could provide useful information for distributors to evaluate the economic value of meat; this would also aid the selection of pork meat with high uniformity to manufacture more stable products, which could enhance revenue for meat packers (Overholt et al., 2016). Furthermore, pig producers could use these primal weight and yield data to improve breeding and feeding management.
It is well known that growth rate and feed efficiency of fattening pigs differ by sex (Lee et al., 2013; National Research Council, 2012; Suzuki & Nishi, 1992). In recent years, different nutritional management based on sex has even been proposed (Wu et al., 2017). It has been reported that the muscle percentage in carcasses was higher in gilts than in barrows (Piao et al., 2004; Unruh et al., 1996), and conversely, intramuscular fat was higher in barrows than in gilts (Ashihara et al., 2008; Stoller et al., 2003), and that net profit was higher in gilts than in barrows when there was no difference in carcass weight between gilts and barrows (Piao et al., 2004). Because these sex differences are of great interest to distributors, farmers, and researchers, this study compared sex differences in the primal cut characteristics. Accurate predictions of primal traits require data from large numbers of pig carcasses with different genetic attributes and feeding environments; in particular, sex effects have been reported for carcass yield (Miyahara et al., 2004) and major commercial cut weights (Correa et al., 2006). However, large sample sizes have not been used to generate prediction formulas in previous studies. Here, we evaluated the accuracy of primal weight and primal yield predictions on the basis of measurements of 12 carcass traits from 4397 pigs (1958 gilts and 2439 barrows) from eight meat processing centers in five regions in Japan.
2 MATERIALS AND METHODS
2.1 Measurements of pig carcass traits
This survey was conducted from January 2020 to October 2021. A total of 4397 pig carcasses (1958 gilts and 2439 barrows) slaughtered using the skinning method at eight meat processing centers in five regions (each one center in Tohoku, Chugoku, Shikoku, and Kyushu and four centers in Kanto) in Japan were used in this study. A large and diverse set of carcasses were collected to in all regions throughout the year to avoid seasonal and regional biases in the carcass trait data. Due to the agreements with the pig farmers, the breeds of pigs used in this study, the number of farms, and the number of samples per region cannot be disclosed in this paper. All the carcasses were either gilts or barrows because uncastrated boars are not usually fattened in Japan.
In the first test, 873 (378 gilts and 495 barrows) carcasses were used to develop a formula for predicting the primal weight and primal yield. In the second test, a higher quality dataset was obtained using measurements from carcasses of 2054 pigs (914 gilts and 1140 barrows); unknown samples of 1470 carcasses (666 gilts and 804 barrows) were used to validate the prediction formulas. To investigate suitable variables for predicting primal weight and primal yield, eight areas (TRAITs 1 to 8 in Figure 1b) of carcasses in addition to carcass weight were measured in the first test. A multiple correlation coefficient of approximately 0.7, which is considered the threshold for practical use, was used as the criterion for determining whether the measured variables were suitable for predicting primal weight and primal yield. However, because the initial variables did not meet the criterion of 0.7, we conducted a new test (the second test) with additional variables that were assumed to be suitable for predicting the primal traits. In the second test, data on three additional traits were used; measurements from 11 areas in addition to carcass weight were used as predictor variables: fat thickness at seven locations and length of bones and muscles at four locations (Figure 1b). Carcass traits data for prediction and validation in the second test were selected as evenly as possible to avoid seasonal and regional biases. More than 2000 samples were used for developing the prediction formulas and the remainder to be used for verification. All measurements were taken by hand to the nearest 1 mm using a 30 cm ruler.
After taking these measurements, the carcasses were divided into shoulder, tenderloin, loin, belly, and ham according to standards established by the Japan Meat Grading Association (JMGA, 2016) (Figure 1a) and weighed. The primal yield was calculated as the percentage of the obtained primal weight divided by the carcass weight. Measurements of the variables and inspection of the primal cuts were performed by staff of the JMGA.
2.2 Statistical analysis
The normality of primal weight and primal yield of all pigs in this study was confirmed by the Anderson-Darling test. To develop high-accuracy prediction formulas, we examined sex differences (gilts vs. barrows) in the carcass and primal cut characteristics. If significant differences were found between the sexes, we confirmed the accuracy of the prediction formulas for each sex. All analyses were performed using JMP 15 (SAS Institute Japan, Tokyo, Japan). For differences in traits between gilts and barrows, the homoscedasticity of the data was confirmed using an F-test, and the means of gilts and barrows were compared using a t-test; the threshold for statistical significance was P < 0.05. The significance of correlation among carcass traits was determined at a level of P < 0.05.
Prediction formulas for primal weight and primal yield were developed using multiple regression analysis. A total of 9 and 12 carcass traits were used as explanatory variables to develop the prediction formulas in the first and second tests, respectively. Backward elimination method was used to select a maximum of 3 variables (excluding carcass weight). That is, in order to reduce the effort required to measure variables and to select variables with high predictive accuracy, variables were excluded in the order of the smallest F-value obtained. The criterion for variable exclusion to prevent multicollinearity was set in advance VIF (variance inflation factor) < 3 (Zuur et al., 2010). No variables were excluded due to multicollinearity in the process of variable selection; the developed prediction formulas were confirmed to meet the VIF criteria. To verify the accuracy of the developed prediction formulas, the predicted and actual values were compared using unknown samples (Figures 3 and 4). The accuracies of the prediction formulas were evaluated using the multiple correlation coefficient, coefficient of determination, root mean square error (RMSE), and residual prediction deviation (RPD). RPD was calculated by dividing the standard deviation of measured values by RMSE.
3 RESULTS AND DISCUSSION
3.1 Carcass weight and sex differences
Histograms of primal weight and primal yield of all pigs used in this study (n = 4397) are shown in Figure 2. Their normality was confirmed as <0.001 for primal weight and <0.05 for primal yield, which is statistically significant but close to a normal distribution (Figure 2). Carcass weight and primal cut characteristics of all pigs used in this study (n = 4397; 1958 gilts and 2439 barrows) are shown in Table 1. The mean carcass weight of gilts and barrows was 75.6 kg and 76.0 kg, respectively, and no significant difference in carcass weight was observed between the sexes (P = 0.07). The standard deviations of the carcass weights were approximately 6 kg for both sexes.
| Item | Overall | Minimum | Maximum | Gilts | Barrows | P valued |
|---|---|---|---|---|---|---|
| (n = 4397) | (n = 1958) | (n = 2439) | ||||
| Carcass weight, kg | 75.8 ± 6.0 | 52.3 | 98.5 | 75.6 ± 6.1 | 76.0 ± 6.0 | 0.07 |
| Primal weight, kga | 56.0 ± 5.0 | 38.1 | 73.7 | 56.6 ± 5.0 | 55.6 ± 4.9 | *** |
| Shoulder, kg | 17.8 ± 1.7 | 11.6 | 23.6 | 17.9 ± 1.7 | 17.8 ± 1.7 | *** |
| Tenderloin, kg | 1.1 ± 0.2 | 0.6 | 1.7 | 1.1 ± 0.2 | 1.0 ± 0.2 | *** |
| Loin, kg | 9.5 ± 1.1 | 6.3 | 13.9 | 9.6 ± 1.2 | 9.3 ± 1.1 | *** |
| Belly, kg | 10.2 ± 1.1 | 6.9 | 15.0 | 10.1 ± 1.1 | 10.3 ± 1.1 | *** |
| Ham, kg | 17.4 ± 1.7 | 11.8 | 24.3 | 17.8 ± 1.6 | 17.2 ± 1.6 | *** |
| Ends and pieces weight, kgb | 19.8 ± 3.0 | 11.8 | 32.8 | 19.1 ± 2.8 | 20.4 ± 3.0 | *** |
| Primal yield, %c | 73.9 ± 3.2 | 62.3 | 84.3 | 74.8 ± 3.0 | 73.2 ± 3.2 | *** |
| Shoulder, % | 23.5 ± 1.4 | 18.5 | 27.9 | 23.7 ± 1.3 | 23.4 ± 1.4 | *** |
| Tenderloin, % | 1.4 ± 0.2 | 0.8 | 2.2 | 1.5 ± 0.2 | 1.4 ± 0.2 | *** |
| Loin, % | 12.5 ± 1.1 | 9.3 | 17.0 | 12.7 ± 1.0 | 12.3 ± 1.0 | *** |
| Belly, % | 13.5 ± 0.9 | 10.2 | 17.0 | 13.3 ± 0.9 | 13.5 ± 0.9 | *** |
| Ham, % | 23.0 ± 1.6 | 17.7 | 28.7 | 23.6 ± 1.5 | 22.6 ± 1.5 | *** |
| End and pieces, % | 26.1 ± 3.2 | 15.7 | 37.7 | 25.2 ± 3.0 | 26.8 ± 3.2 | *** |
- Note: Values represent means ± SD.
- a Primal weight is the sum of shoulder, tenderloin, loin, belly, and ham weight. Primal cut positions are illustrated in Figure 1a.
- b Ends and pieces weight was calculated as carcass weight − primal weight.
- c Primal yield was calculated as primal weight/carcass weight × 100.
- d Significant differences between gilts and barrows were tested by one-way analysis of variance.
- *** P < 0.001.
In previous studies of the carcass performance of gilts and barrows, carcass weight was reported to be greater in barrows (95.1 kg) than in gilts (94.2 kg) according to data from 8042 animals by Overholt et al. (2016); however, no differences were observed between these two (n = 240; gilts, 91.4 kg; barrows, 91.6 kg) when the pigs were slaughtered at the same body weight (Latorre et al., 2003). Barrows have been shown to grow faster than gilts in a previous study (Lee et al., 2013). The daily growth of livestock depends on their feed intake; thus, the carcass weight strongly depends on the market age and feeding period. The reason for the lack of sex differences in carcass weight in this study presumably stems from the range of carcass weights considered acceptable for grading in Japan. Generally, the age at slaughter for gilts is approximately the same or slightly greater than that for barrows.
3.2 Primal cut characteristics and sex differences
Total primal weights were 56.6 kg for gilts and 55.6 kg for barrows; thus, gilts were approximately 1 kg heavier than barrows (P < 0.001) (Table 1). The primal yield of gilts (74.8%) was also greater than that of barrows (73.2%) (P < 0.001). The shoulder, tenderloin, loin, belly, and ham weights were significantly greater in barrows than in gilts (P < 0.001; Figure 1a), and the belly weight was significantly greater in barrows than in gilts (P < 0.001). Shoulder, tenderloin, loin, and ham yields were significantly greater in gilts than in barrows (P < 0.001), and the belly yield was greater in barrows than in gilts (P < 0.001). The primal cut with the largest difference in weight between the sexes was ham (0.6 kg), and there was a 1-point difference in primal yield % between gilts and barrows (P < 0.001). In addition, large sex differences were observed in the percentage of “End and pieces (excess fat and bones)” to primal weight and primal yield, which was higher in barrows (26.8%) than in gilts (25.2%) (P < 0.001). The significant sex differences observed in primal cut characteristics (primal weight and yield) might stem from sex differences in the growth rates of each tissue (NRC, 2012). In general, muscle tissue develops faster than adipose tissue during the early stages of growth, but the rate of adipose tissue growth exceeds the rate of muscle growth in the finishing stage (Irshad et al., 2012). Sex differences begin to appear when the live weight exceeds approximately 60 kg, and the ratio of fat to protein accumulation becomes greater in barrows than in gilts (NRC, 2012). Overholt et al. (2016) suggested that barrows may require less energy to produce lean tissue than gilts (Schinckel, 1994; Thompson et al., 1996) and that their excess energy would be stored as fat.
The belly was the only primal that was heavier in barrows than in gilts in this study. Adipose tissue accumulates extensively in the belly, and intermuscular fat is often deposited on the inside of the latissimus dorsi muscle, deep pectoral muscle, and other muscles. Thus, we speculated that the greater belly weight of barrows compared with gilts stemmed from their greater fat content. Our findings demonstrated that protein accumulation was slightly higher in gilts than in barrows, and the number of ends and pieces (especially excess fat) that needed to be removed during primal cut processing was lower in gilts than in barrows. Gilt carcasses thus contained more edible parts than barrow carcasses. Given that ham weight was greater in gilts than in barrows, the hindquarters of gilts might be slightly more well-developed than those of barrows.
3.3 Carcass traits and sex differences
To evaluate sex differences in carcass traits, measurements were taken of the thickness of several areas of subcutaneous fat (backfat) and lengths of bones and muscles (Figure 1b and Table 2). All seven backfat thickness measurements of gilts were 2 to 4 mm thinner than those of barrows (P < 0.001). All four bone and muscle length measurements were greater in gilts than in barrows (P < 0.001).
| Traita | Nb | Overall | Minimum | Maximum | N | Gilts | N | Barrows | P valuec |
|---|---|---|---|---|---|---|---|---|---|
| Backfat thickness, cm | |||||||||
| 1. Tail end of M. gluteus medius | 4397 | 2.7 ± 0.7 | 0.8 | 5.5 | 1958 | 2.5 ± 0.7 | 2439 | 2.9 ± 0.7 | *** |
| 2. Above M. gluteus medius | 4397 | 2.0 ± 0.6 | 0.5 | 4.6 | 1958 | 1.9 ± 0.6 | 2439 | 2.1 ± 0.6 | *** |
| 3. Front end of M. gluteus medius | 4397 | 2.7 ± 0.7 | 0.9 | 5.3 | 1958 | 2.6 ± 0.7 | 2439 | 2.9 ± 0.6 | *** |
| 4. Thickest part of lumbar | 4397 | 3.1 ± 0.7 | 1.0 | 5.5 | 1958 | 3.0 ± 0.7 | 2439 | 3.2 ± 0.6 | *** |
| 5. 13th rib | 4397 | 2.1 ± 0.6 | 0.5 | 4.7 | 1958 | 1.9 ± 0.6 | 2439 | 2.3 ± 0.6 | *** |
| 6. 4th rib | 4397 | 3.3 ± 0.7 | 1.2 | 5.9 | 1958 | 3.1 ± 0.7 | 2439 | 3.5 ± 0.6 | *** |
| 7. Thickest part of shoulder | 4397 | 4.0 ± 0.7 | 2.0 | 6.6 | 1958 | 3.8 ± 0.7 | 2439 | 4.2 ± 0.7 | *** |
| Bone and muscle length, cm | |||||||||
| 8. 13th thoracic vertebra spinous process | 4397 | 4.2 ± 0.3 | 2.6 | 5.9 | 1958 | 4.3 ± 0.3 | 2439 | 4.2 ± 0.3 | *** |
| 9. Under M. gluteus medius – last lumbar | 3524 | 6.6 ± 0.8 | 4.2 | 10.0 | 1580 | 6.8 ± 0.8 | 1944 | 6.4 ± 0.8 | *** |
| 10. Between 9th and 13th thoracic vertebrae | 3524 | 16.1 ± 0.8 | 11.6 | 19.1 | 1580 | 16.2 ± 0.8 | 1944 | 16.0 ± 0.8 | *** |
| 11. 1st thoracic vertebra – backfat | 3524 | 11.3 ± 0.8 | 8.0 | 16.2 | 1580 | 11.4 ± 0.8 | 1944 | 11.2 ± 0.8 | *** |
- Note: Values represent means ± SD.
- a Locations from which measurements were taken are illustrated in Figure 1b.
- b N, number of animals.
- c Significant differences between gilts and barrows were tested by one-way analysis of variance.
- *** P < 0.001.
Gilts and barrows are known to differ in their fat content; we performed a cross-tabulation analysis of carcass weight and backfat thickness (above the 13th rib), which revealed the most important factors for grading pork carcasses (Table 3). The mode of carcass weight ranged from 75.0 to 80.0 kg in both gilts and barrows. The mode of backfat thickness ranged from 1.5 to 1.9 cm for gilts and 2.0 to 2.4 cm for barrows; these ranges were consistent with the carcass trait results. That is, at the market weight, no sex differences in carcass weight were observed; fat accumulation was greater in barrows than in gilts; and muscle mass (i.e., protein accumulation) was greater in gilts than in barrows.
| Carcass weight, kg | Sex | Backfat thickness at 13th rib, cm | Row totals | |||||
|---|---|---|---|---|---|---|---|---|
| ~0.9 | 1.0–1.4 | 1.5–1.9 | 2.0–2.4 | 2.5–2.9 | 3.0~ | |||
| ~64.9 | Overall | 6 | 25 | 60 | 34 | 12 | 2 | 139 |
| Gilts | 6 | 14 | 31 | 15 | 1 | 1 | 68 | |
| Barrows | 0 | 11 | 29 | 19 | 11 | 1 | 71 | |
| 65.0–69.9 | Overall | 20 | 96 | 355 | 217 | 48 | 9 | 745 |
| Gilts | 14 | 69 | 188 | 72 | 13 | 4 | 360 | |
| Barrows | 6 | 27 | 167 | 145 | 35 | 5 | 385 | |
| 70.0–74.9 | Overall | 49 | 117 | 210 | 218 | 223 | 49 | 866 |
| Gilts | 33 | 87 | 108 | 68 | 57 | 9 | 362 | |
| Barrows | 16 | 30 | 102 | 150 | 272 | 40 | 504 | |
| 75.0–80.0 | Overall | 79 | 123 | 399 | 467 | 378 | 129 | 1515 |
| Gilts | 46 | 83 | 199 | 218 | 106 | 16 | 668 | |
| Barrows | 33 | 40 | 140 | 249 | 272 | 113 | 847 | |
| 80.1–85.0 | Overall | 15 | 44 | 178 | 257 | 276 | 128 | 898 |
| Gilts | 11 | 34 | 111 | 114 | 101 | 29 | 400 | |
| Barrows | 4 | 10 | 67 | 143 | 175 | 99 | 498 | |
| 85.1~ | Overall | 0 | 9 | 47 | 62 | 74 | 42 | 234 |
| Gilts | 0 | 5 | 26 | 26 | 33 | 10 | 100 | |
| Barrows | 0 | 4 | 21 | 36 | 41 | 32 | 134 | |
| Column totals | Overall | 169 | 414 | 1189 | 1255 | 1011 | 359 | 4397 |
| Gilts | 110 | 292 | 663 | 513 | 311 | 69 | 1958 | |
| Barrows | 59 | 122 | 526 | 742 | 700 | 290 | 2439 | |
In Japan, carcass weight and backfat thickness are used for grading pork carcasses (JMGA, 2016) because the lean meat and fat content in the carcass can be predicted to some extent by measuring these two traits. According to the grading system, the carcass weights for gilts and barrows were similar in this study; however, the backfat thickness of barrows was greater than that of gilts, and fat accumulation in barrows was observed throughout the carcass, from the shoulder to loin. Our results supported the findings of Latorre et al. (2003) showing that gilts had less backfat than barrows. The results of the cross-tabulation analysis demonstrated that the mode of carcass weight was similar for gilts and barrows (75.0 to 80.0 kg), but the mode of backfat thickness was greater in barrows than in gilts (Table 3). The percentages of pigs rated as “upper grade” in 2021 was 50.3% for gilts and 47.5% for barrows, and the main cause of the difference in the grading of carcasses with a weight of 75 and 80 kg (sex not disclosed) was the presence of excess backfat (JMGA, 2021). Overall, our findings indicate that this difference in the grading of carcasses between the sexes might stem from sex differences in the development of subcutaneous adipose tissue.
Overall, the goal of the sampling in this study was to obtain carcasses with a variety of features to develop formulas for predicting primal cut traits. Although samples were not randomly selected, the broad distribution of data on carcass weight and primal weight indicates that the variance in the data could potentially be reduced (i.e., the number of carcasses significantly deviating from the mode could be decreased). The broad distribution of carcasses and primal weight data was attributed to the range of upper-grade and medium-grade carcasses, as well as genetic differences, such as differences among breeds and strains, variation in individual growth under all-in-all-out systems, and various feeding environments. Reducing the variance in carcass weight and primal weight might enhance the uniformity of pork production and enhance efficiency, including the number of types of packaging materials used for transportation and workability in meat processing, which could increase profits.
3.4 Relationships among carcass traits
To identify the carcass traits important for developing prediction formulas for primal weight and primal yield, correlations among carcass traits were investigated (Table 4). Carcass weight was significantly positively correlated (P < 0.001) with primal weight, backfat thickness, and bone and muscle lengths.
| Variablesa | (a) | (b) | (c) | (d) | (e) | (f) | (g) | (h) | (i) | (j) | (k) | (l) | (m) | (n) | (o) | (p) | (q) | (r) | (s) | (t) | (u) | (v) | (w) | (x) | (y) | (z) | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| (a) | Carcass weight | 1 | |||||||||||||||||||||||||
| (b) | Primal weight | 0.87 | 1 | ||||||||||||||||||||||||
| (c) | Shoulder weight | 0.80 | 0.91 | 1 | |||||||||||||||||||||||
| (d) | Tenderloin weight | 0.47 | 0.72 | 0.64 | 1 | ||||||||||||||||||||||
| (e) | Loin weight | 0.70 | 0.83 | 0.65 | 0.61 | 1 | |||||||||||||||||||||
| (f) | Belly weight | 0.80 | 0.75 | 0.64 | 0.37 | 0.49 | 1 | ||||||||||||||||||||
| (g) | Ham weight | 0.70 | 0.90 | 0.74 | 0.72 | 0.72 | 0.53 | 1 | |||||||||||||||||||
| (h) | Ends and pieces weight | 0.57 | 0.09 | 0.09 | −0.26 | 0.04 | 0.36 | −0.09 | 1 | ||||||||||||||||||
| (i) | Primal yield % | −0.05 | 0.45 | 0.40 | 0.61 | 0.40 | 0.07 | 0.55 | −0.85 | 1 | |||||||||||||||||
| (j) | Shoulder % | −0.03 | 0.33 | 0.58 | 0.42 | 0.12 | −0.02 | 0.28 | −0.61 | 0.72 | 1 | ||||||||||||||||
| (k) | Tenderloin % | −0.06 | 0.30 | 0.25 | 0.85 | 0.28 | −0.05 | 0.40 | −0.62 | 0.71 | 0.49 | 1 | |||||||||||||||
| (l) | Loin % | 0.07 | 0.36 | 0.18 | 0.43 | 0.76 | −0.04 | 0.36 | −0.46 | 0.60 | 0.20 | 0.44 | 1 | ||||||||||||||
| (m) | Belly % | 0.14 | 0.21 | 0.11 | 0.06 | −0.02 | 0.70 | 0.04 | −0.07 | 0.17 | 0.00 | −0.02 | −0.15 | 1 | |||||||||||||
| (n) | Ham % | −0.19 | 0.23 | 0.10 | 0.45 | 0.17 | −0.20 | 0.57 | −0.76 | 0.81 | 0.42 | 0.62 | 0.41 | −0.11 | 1 | ||||||||||||
| (o) | Ends and pieces % | 0.05 | −0.45 | −0.40 | −0.61 | −0.40 | −0.07 | −0.55 | 0.85 | −1.00 | −0.72 | −0.71 | −0.60 | −0.17 | −0.81 | 1 | |||||||||||
| (p) | TRAIT1 | 0.28 | −0.06 | −0.05 | −0.36 | −0.06 | 0.24 | −0.23 | 0.66 | −0.63 | −0.46 | −0.56 | −0.33 | 0.07 | −0.63 | 0.63 | 1 | ||||||||||
| (q) | TRAIT2 | 0.25 | −0.08 | −0.09 | −0.37 | −0.04 | 0.24 | −0.25 | 0.64 | −0.62 | −0.49 | −0.57 | −0.29 | 0.10 | −0.63 | 0.62 | 0.89 | 1 | |||||||||
| (r) | TRAIT3 | 0.28 | −0.06 | −0.06 | −0.36 | −0.05 | 0.25 | −0.22 | 0.66 | −0.62 | −0.47 | −0.56 | −0.33 | 0.08 | −0.61 | 0.62 | 0.90 | 0.91 | 1 | ||||||||
| (s) | TRAIT4 | 0.27 | −0.05 | −0.05 | −0.34 | −0.05 | 0.24 | −0.19 | 0.63 | −0.59 | −0.46 | −0.54 | −0.32 | 0.08 | −0.57 | 0.59 | 0.86 | 0.86 | 0.93 | 1 | |||||||
| (t) | TRAIT5 | 0.31 | −0.01 | −0.03 | −0.33 | 0.00 | 0.30 | −0.18 | 0.65 | −0.59 | −0.47 | −0.56 | −0.28 | 0.12 | −0.61 | 0.59 | 0.83 | 0.85 | 0.86 | 0.87 | 1 | ||||||
| (u) | TRAIT6 | 0.30 | 0.00 | 0.01 | −0.31 | 0.02 | 0.30 | −0.19 | 0.60 | −0.53 | −0.39 | −0.52 | −0.24 | 0.14 | −0.60 | 0.53 | 0.75 | 0.77 | 0.77 | 0.78 | 0.85 | 1 | |||||
| (v) | TRAIT7 | 0.34 | 0.03 | 0.04 | −0.31 | 0.03 | 0.33 | −0.17 | 0.64 | −0.56 | −0.39 | −0.55 | −0.27 | 0.14 | −0.62 | 0.56 | 0.77 | 0.78 | 0.78 | 0.79 | 0.83 | 0.91 | 1 | ||||
| (w) | TRAIT8 | 0.35 | 0.50 | 0.39 | −0.47 | 0.52 | 0.29 | 0.49 | −0.13 | 0.38 | 0.17 | 0.33 | 0.41 | 0.06 | 0.28 | −0.38 | −0.23 | −0.18 | −0.21 | −0.23 | −0.21 | −0.18 | −0.17 | 1 | |||
| (x) | TRAIT9 | 0.17 | 0.28 | 0.20 | 0.17 | 0.19 | 0.17 | 0.36 | −0.13 | 0.28 | 0.10 | 0.10 | 0.13 | 0.10 | 0.31 | −0.28 | −0.15 | −0.18 | −0.13 | −0.06 | −0.10 | −0.12 | −0.06 | 0.13 | 1 | ||
| (y) | TRAIT10 | 0.33 | 0.41 | 0.37 | 0.46 | 0.38 | 0.22 | 0.40 | −0.03 | 0.26 | 0.32 | 0.32 | 0.25 | −0.03 | 0.18 | −0.26 | −0.28 | −0.23 | −0.25 | −0.29 | −0.30 | −0.23 | −0.23 | 0.44 | −0.10 | 1 | |
| (z) | TRAIT11 | 0.37 | 0.49 | 0.43 | 0.42 | 0.48 | 0.32 | 0.44 | −0.07 | 0.33 | 0.25 | 0.25 | 0.34 | 0.10 | 0.18 | −0.33 | −0.10 | −0.07 | −0.08 | −0.09 | −0.07 | 0.00 | 0.00 | 0.46 | 0.23 | 0.23 | 1 |
- Note: Values are correlation coefficients of n = 4397 for variables (a)–(w) and n = 3524 for (x)–(z). Non-significant values (P ≥ 0.05) are indicated by the italic letters. Correlation coefficients greater than or equal to 0.50 are written in bold.
- a Variables (a)–(o) correspond to the items listed in Table 1, and variables (p)–(z) correspond to the TRAITS shown in Table 2 and Figure 1b. Definitions for each TRAIT are as follows: TRAIT1, M. gluteus medius tail end fat; TRAIT2, fat above M. gluteus medius; TRAIT3, M. gluteus medius front end fat; TRAIT4, thickest lumbar fat; TRAIT5, 13th rib fat; TRAIT6, 4th rib fat; TRAIT7, thickest shoulder fat; TRAIT8, spinous process length of 13th thoracic vertebra; TRAIT9, length from last lumbar (dorsal end) to M. gluteus medius (ventral side); TRAIT10, length between 9th (front end) and 13th (tail end) thoracic vertebra; TRAIT11, length from 1st thoracic vertebra (top end) to backfat (including supraspinous ligament).
The correlation between carcass weight and primal yield was weak (r = −0.05; P < 0.01). The correlation between backfat thickness and subcutaneous thickness was strong at seven locations, with correlation coefficients ranging from 0.75 to 0.93 (P < 0.001). The correlation between “Loin weight” and “spinous process length of the 13th thoracic vertebra” was relatively strong (r = 0.52; P < 0.001). The correlation coefficients between most of the primal weights and backfat thicknesses were negative, and those between “Belly weight” and backfat thicknesses were positive (r = 0.24–0.33) (P < 0.001). Correlation coefficients between carcass traits were similar in both sexes (data not shown).
Our results indicated that carcass weight was positively correlated with “ends and pieces weight” (excess fat and bone) (r = 0.57), and the relationship between carcass weight and primal yield was weak (r = −0.05). Increases in carcass weight thus led to increases in primal weight, but the effect is offset by the simultaneous increase in ends and pieces weight; thus, there was no significant correlation between carcass weight and primal yield. That is, the ratio of protein accumulation to fat accumulation did not differ significantly among carcasses with high and low weight. In contrast to other primals, belly weight was positively correlated with backfat thicknesses. This suggests that the greater belly weight of barrows compared with gilts stems from their greater fat accumulation.
The strong positive correlation between backfat thicknesses measured at seven sites indicated that these fat thickness measurements over a wide area from the shoulder to the loin could be used as an indicator of subcutaneous fat accumulation in pork carcasses. The moderate positive correlation (P < 0.001) between primal weight and the “spinous process length of the 13th thoracic vertebra” (TRAIT 8 in Table 4) or the lengths between each point (TRAITs 9 to 11) suggests that growth of the bone structure supporting the muscles is essential for mediating increases in primal weight. The development of the spinous processes is considered important for promoting increases in the economically important M. longissimus.
3.5 Prediction formulas for primal weight and primal yield using nine variables
In the first test, nine variables (carcass weight, TRAITs 1 to 8 in Figure 1b) were used to develop prediction formulas for primal weight and primal yield on the basis of 873 samples (378 gilts and 495 barrows). A high-accuracy prediction formula was developed for primal weight (R = 0.89), and the accuracy of the prediction formula for primal yield was moderate (R = 0.65). This indicated that additional measurements of other carcass traits (variables) that reflect meat volume are needed to improve the accuracy of primal yield in the second test. The accuracy of the prediction formulas for primal weight and partial yield was not altered much when the effect of sex was accounted for. Therefore, the prediction formula that accounts for sex differences in the first test was not used to predict primal weight and primal yield.
3.6 Prediction formulas for primal weight and primal yield using 12 variables
To increase the accuracy of the prediction formula for primal yield, three traits that were expected to reflect the amount of meat were measured, and the accuracy of the prediction formulas for predicting carcass weight and carcass yield was confirmed using 12 traits (carcass weight, TRAITs 1 to 11 in Figure 1b) as a second test (Tables 5 and 6). The explanatory variables selected for the primal weight prediction formula were “carcass weight,” “backfat thickness above M. gluteus medius (TRAIT 2),” “Spinous process length of 13th thoracic vertebra (TRAIT 8),” and “length from 1st thoracic vertebra to backfat (TRAIT 11),” which are commonly used explanatory variables (except for “carcass weight”) in primal yield prediction formulas. For primal weight, carcass weight produced the highest R2 of 0.76. The prediction formula combining carcass weight and TRAIT 2 had high accuracy (R = 0.93 and RMSE = 1.89), and more additional variables would not improve the accuracy of the estimate very much. There was no explanatory variable with particularly strong relationship with primal yield, indicating that it is necessary to use all three selected variables to create an accurate prediction formula.
| Models | Sex | N | Chosen variablesa | R | R2 | RMSE | RPD |
|---|---|---|---|---|---|---|---|
| Model 1 | Overall | 2054 | Carcass weight, TRAIT2, TRAIT8, TRAIT11 | 0.94 | 0.89 | 1.70 | 3.0 |
| Model 2 | Overall | 2054 | Carcass weight, TRAIT2, TRAIT11 | 0.94 | 0.88 | 1.77 | 2.9 |
| Model 3 | Overall | 2054 | Carcass weight, TRAIT2, TRAIT8 | 0.94 | 0.88 | 1.77 | 2.9 |
| Model 4 | Overall | 2054 | Carcass weight, TRAIT2 | 0.93 | 0.86 | 1.89 | 2.7 |
| Model 5 | Overall | 2054 | Carcass weight | 0.87 | 0.76 | 2.46 | 2.1 |
| Model 6 | Gilts | 914 | Carcass weight, TRAIT2, TRAIT8, TRAIT11 | 0.95 | 0.90 | 1.63 | 3.2 |
| Model 7 | Barrows | 1140 | Carcass weight, TRAIT2, TRAIT8, TRAIT11 | 0.94 | 0.88 | 1.70 | 2.9 |
- Abbreviations: N, number of animals used for formula development; R, correlation coefficient; R2, determination coefficient; RMSE, root mean squared error; RPD, residual predictive deviation.
- a Names of the variables match those in Table 4. Definitions for each TRAIT are as follows: TRAIT2, fat above M. gluteus medius; TRAIT8, spinous process length of 13th thoracic vertebra; TRAIT11, length from 1st thoracic vertebra (top end) to backfat (including supraspinous ligament).
| Models | Sex | N | Chosen variablesa | R | R2 | RMSE | RPD |
|---|---|---|---|---|---|---|---|
| Model 1 | Overall | 2054 | TRAIT2, TRAIT8, TRAIT11 | 0.72 | 0.52 | 2.21 | 1.5 |
| Model 2 | Overall | 2054 | TRAIT2, TRAIT8 | 0.70 | 0.48 | 2.30 | 1.4 |
| Model 3 | Overall | 2054 | TRAIT2 | 0.62 | 0.38 | 2.52 | 1.3 |
| Model 4 | Gilts | 914 | TRAIT2, TRAIT8, TRAIT11 | 0.69 | 0.47 | 2.14 | 1.4 |
| Model 5 | Barrows | 1140 | TRAIT2, TRAIT8, TRAIT11 | 0.72 | 0.52 | 2.20 | 1.4 |
- Abbreviations: N, number of animals used for formula development; R, correlation coefficient; R2, determination coefficient; RMSE, root mean squared error; RPD, residual predictive deviation.
- a Names of the variables match those in Table 4. Definitions for each TRAIT are as follows: TRAIT2, fat above M. gluteus medius; TRAIT8, spinous process length of 13th thoracic vertebra; TRAIT11, length from 1st thoracic vertebra (top end) to backfat (including supraspinous ligament).
The addition of more explanatory variables to the prediction formula only marginally improved the accuracy and was not worth the increased measurement effort. The accuracy of the prediction formula developed for gilts plus barrows was comparable to the accuracy of the prediction formula developed for each sex, indicating that, as in the first test, there was no need to separate the prediction equation for gilts and barrows in the second test.
3.7 Validation of predictive formulas for primal weight and primal yield
- Primal weight (prediction value) kg = −9.806 + (0.735 × carcass weight) + (−2.327 × backfat thickness above M. glutes medius) + (1.758 × spinous process length of 13th thoracic vertebra) + (0.647 × length from 1st thoracic vertebra to backfat)
- Primal yield (prediction value)% = 60.743 + (−3.034 × backfat thickness above M. glutes medius) + (2.297 × spinous process length of 13th thoracic vertebra) + (0.838 × length of 1st thoracic vertebra to backfat)
The prediction formulas for primal weight and primal yield were subtracted by fat thickness and added by length of bone, which are thought to reflect meat volume; this is consistent with meat production theory.
The accuracy of the prediction formula for primal weight was high, given that primal weight was highly correlated with carcass weight (r = 0.87, Table 4) and the variability in the distribution of carcass weight was high. The prediction accuracy of primal yield was not as high as that of primal weight because the correlation between the “backfat thickness above M. glutes medius” and the weight of “ends and pieces” (fat and bone to be eliminated) was only moderate (r = 0.64, Table 4); the correlations between the explanatory variables reflecting the amount of meat (“spinous process length of 13th thoracic vertebra”; “length of 1st thoracic vertebra to backfat”) and primal weight were also moderate (r = 0.50 and 0.49, respectively, Table 4), and the variability in the distribution of the primal yield ratio of pigs in Japan was lower than that for carcass weight.
Although there are no universally accepted criteria for evaluating the accuracy of prediction formulas, the coefficient of determination and RPD have previously been used to evaluate the accuracy of prediction formulas (Chang et al., 2001; Prieto et al., 2017). The prediction formula for primal weight was rated as “Good, Quality control” by Prieto's criteria and “Successful” by Chang's criteria, indicating that this formula was highly accurate. The prediction formula for primal yield was evaluated as “Very poor, Not recommended” by Prieto's criteria and “Possibility” by Chang's criteria. Prieto's criteria were considered too strict for evaluating the accuracy of primal yield in this study because these criteria were developed for evaluating the accuracy of formulas derived from ingredient analysis using near-infrared spectroscopy. Given that it was rated as “Possibility” by Chang's criteria, the accuracy of the primal yield prediction formula was considered relatively good, but not high.
An ultrasound-based classification system initially launched in Europe has been used for the classification of pig carcasses in more than 14 countries (Choi et al., 2018). Highly accurate estimates of the carcass composition have been obtained in other countries using state-of-the-art equipment (Kim et al., 2021), and three-dimensional information is particularly useful. However, there are approximately 140 pig slaughterhouses in Japan. This is a large number of slaughterhouses relative to the number of pigs slaughtered, and this makes the cost of using expensive equipment on a large scale difficult to justify. The prediction formulas for primal weight and primal yield developed in this study were considered suitable for use in Japan because of their low cost and relatively high accuracy.
In conclusion, our results indicated that primal weight can be predicted with high accuracy and primal yield can be predicted with relatively good accuracy without accounting for the effect of sex using measurements of “carcass weight,” “backfat thickness above M. gluteus medius,” “spinous process length of 13th thoracic vertebra,” and “length from 1st thoracic vertebra to backfat.” Predicted primal weight and primal yield data could be used to improve breeding, feeding management, and the efficiency of meat processing. Information on intramuscular fat (Kohira et al., 2021) and fatty acid composition (Okumura et al., 2022), which affect the sensory characteristics of pork, is now available to producers and others using the measurement software that we developed for near-infrared instruments. In the future, the use of information on pork meat quality and yield could enhance pork production.
ACKNOWLEDGMENTS
This work was supported by the Livestock Promotional Subsidy from the Japan Racing Association and “Urgent Survey Project for Improving Domestic Pork Productivity and Quality” conducted in 2019 to 2021. We thank staff of the Operations Department and graders of the Japan Meat Grading Association. We thank Edanz (https://jp.edanz.com/ac) for editing a draft of this manuscript.
CONFLICT OF INTEREST STATEMENT
The authors declare no conflict of interest.
