Validation of Dual-Energy X-Ray Absorptiometry to Estimate the Carcass Tissue Composition of Nellore Cull Cows

Introduction

In recent decades, the meat industry has evolved from subjective methods, such as visual estimation of carcass fat, to objective technologies capable of accurately estimating the composition and quality characteristics of carcasses and meat. This transition has enabled more efficient classification of cuts based on criteria defined by the market, which in turn favors differentiated remuneration for producers who meet the quality standards demanded by both the industry and consumers (Allen, 2021; Calnan et al., 2021).

Carcass composition can be determined using reference methods such as dissection, which allows the separation and quantification of individual tissues. Although highly accurate, dissection requires the physical separation or grinding of carcass components, demanding time, resources, and skilled labor. In addition, because this method is destructive, it is impossible to sell the cuts, restricting its use to research environments (Lerch et al., 2023). The need to make this process more agile and applicable to the industry has led to the development and adoption of advanced technologies for the objective assessment of carcass composition.

Among these technologies, dual-energy X-ray absorptiometry (DXA) stands out as a noninvasive method for measuring body composition. DXA software estimates the amounts of bone mineral content (BMC) and total lean and fat tissues by analyzing the attenuation of dual X-ray beams as they pass through materials with different absorption properties (Schallier et al., 2019). This technology has been widely used to develop equation models to predict tissue composition of carcasses and primal cuts in pigs and lamb (Kipper et al., 2019; Gardner et al., 2021). In cattle, DXA has been validated against carcass dissection for the accurate prediction of lean, fat, and bone tissues in whole-carcass sides and primal cuts of steers (López-Campos et al., 2018). Despite these advances, prediction equations specifically developed for cull Nellore cow carcasses remain scarce in the literature, particularly under tropical production systems characterized by high biological variability.

In Brazil, cull cows represent a substantial proportion of the animals slaughtered and contribute significantly to beef production. In the 4th quarter of 2024, females accounted for approximately 39.9% of total cattle slaughter in Brazil. Most of these female animals were cows. In the Brazilian production context, these cows are predominantly derived from culling processes (IBGE, 2025). Furthermore, production systems are largely pasture-based (most animals are raised under extensive grazing conditions), while only 18.2% of slaughtered cattle are finished in feedlots (ABIEC, 2023). These characteristics contribute to wide variability in carcass composition, reinforcing the need for robust and objective methods capable of accurately predicting lean, fat, and bone tissues in cull cow carcasses.

From an industrial perspective, beef carcass evaluation and pricing in Brazil are based on traits such as conformation, fat cover, carcass weight, age, and sex. However, the absence of a standardized national carcass grading system results in heterogeneous carcass classification across slaughterhouses, leading to price inconsistencies and reduced market transparency (Nunes et al., 2024). Additionally, conventional methods, including visual scoring and linear measurements, are widely used but are subjective, have low repeatability, and provide only localized estimates. In this context, DXA provides a rapid, nondestructive, and objective method for accurately determining carcass composition—specifically lean, fat, and bone tissues—thereby enabling robust, reproducible measurements for experimental datasets and strengthening the biological interpretation and comparability of research outcomes. Although DXA quantifies total fat rather than subcutaneous fat specifically, its potential to predict subcutaneous fat in cull Nellore cow carcasses warrants further evaluation, given the importance of this tissue for carcass yield, trimming losses, and meat quality (Boito et al., 2018; Moreira et al., 2018).

Considering the large size of bovine half carcasses and the limitations of full-body scanning in industrial settings, identifying representative anatomical regions for DXA scanning is essential to provide accurate estimates of carcass composition. Thus, we hypothesized that DXA measurements obtained from specific sections accurately reflect the overall composition of the half carcass, allowing reliable estimation of lean, fat, and bone without scanning the entire carcass. Consequently, the objective of this study was to develop prediction equations for half carcasses of cull Nellore cows based on DXA scans of different standardized carcass sections.

Materials and Methods

The experiment was conducted at the Frigorífico Escola and the Body Composition and Densitometry Laboratory of the Department of Animal Science, Universidade Federal de Viçosa (UFV), in Viçosa, Minas Gerais, Brazil.

Dual-energy X-ray absorptiometry scan and dissection

The right carcass sides were then cut into 5 standardized sections (Figure 1), as described: S1 section delimited by a cross-cut immediately after the 2nd rib; S2 cross-cut after the 8th rib; S3 cross-cut after the 13th rib; S4 cross-section immediately after the sacrum, corresponding to the rump region; and for S5, the residual portion of the carcass remained after removal of the previous sections (S1–S4), adapted from Silva et al. (2023).

Figure 1.

Schematic representation of the 5 standardized sections (S1–S5) of the right half of the carcasses of cull cows used for analysis. The anatomical boundaries were as follows: S1, cross-section immediately after the 2nd rib; S2, cross-section after the 8th rib; S3, cross-section immediately after the thirteenth rib; S4, cross-section after the sacrum, corresponding to the rump region (alcatra); and S5, the remaining part of the carcass, corresponding to the hindquarter (round, coxão).

The sections were individually packaged in food-grade polyethylene bags (50 × 70 cm, 0.20-mm thickness), identified, and transported by car to the DXA facility located approximately 1.6 km from the slaughterhouse (∼3 min). During transportation, the samples were placed in insulated thermal boxes to maintain refrigeration. Upon arrival, the sections were scanned using medical DXA equipment (GE HealthCare, Lunar Prodigy Advance, Madison, WI, USA). The scanning room was climate-controlled at approximately 16°C, and the sections remained chilled during the scanning procedure. To avoid prolonged exposure to room temperature, sections from only 1 carcass were taken to the scanning room at a time. Immediately after scanning, they were returned to the refrigerated chamber at the slaughterhouse.

Before scanning, the DXA device was calibrated according to the manufacturer’s recommendations. The analysis was performed using the GE HealthCare enCORE software (version 18) in the “Small Animal” configuration mode. Prior to scanning, the average sample weight and dimensions (width × length × height) were entered into the software to ensure proper identification of the regions of interest. During scanning, the sections were positioned laterally on the DXA table, with the lateral carcass surface facing downward. The samples were scanned inside the polyethylene bags to prevent moisture loss and contamination. Under these conditions, the system estimated lean tissue (kg), fat tissue (kg), and BMC (kg) for each section.

Subsequently, each section (S1–S5) was manually dissected into lean, bone, subcutaneous fat, and intermuscular fat. Subcutaneous fat was defined as adipose tissue located beneath the skin, while intermuscular fat corresponded to adipose tissue located between adjacent muscles within each section. These 2 deposits, when added together, were considered total fat. Intramuscular fat (marbling) was not dissected individually and remained in the lean tissue. Dissections were performed by trained professional butchers following standard commercial procedures. Connective tissue, membranes, and cartilage were kept with the lean tissue. Bones were cleaned of adherent soft tissues before weighing, and internal fat deposits (renal, pelvic, and cardiac fat) were removed during carcass preparation and not included in the sections. Tissue weights were recorded separately to determine the physical composition of each section and the half carcass.

For analyses and comparisons between the weight of dissected tissue and the weight of DXA tissue, the weight of DXA lean tissue corresponded to dissected lean, total fat measured by DXA was considered to correspond to the sum of dissected intramuscular and subcutaneous fat, and DXA BMC was considered to correspond to dissected bones.

Results and Discussion

The cull cows used in the study had a BW of 482.2 kg plus or minus 65.4 kg (range: 327–595 kg). After chilling, the right half-carcass weight averaged 129.8 kg plus or minus 18.5 kg (range: 86.2–155.0 kg), with an average fat cover score of 2.1 (range: 1.0–3.5), REA of 78.85 cm² plus or minus 13.11 cm² (range: 53.63–99.75 cm²), and subcutaneous fat thickness of 5.40 mm plus or minus 4.60 mm (range 1.02–19.42 mm). The substantial variation in subcutaneous fat deposition among carcasses is characteristic of cull cows and reflects differences in age and physiological status.

Dissection of the cold carcass indicated that, on average, it consisted of 85.19 kg plus or minus 10.43 kg of meat, 22.93 kg plus or minus 7.89 kg of fat, and 21.31 kg plus or minus 2.75 kg of bone (Table 1). Regarding bone, the BMC values estimated by DXA were consistently lower than those obtained by dissection. Concerning this underestimation of BMC values compared to dissected bone mass, Nielsen et al. (1973) reported that while dissection evaluates the total bone mass (including organic components and water), DXA quantifies only the BMC, which is composed of calcium and phosphorus. Thus, it is expected that the BMC values obtained by DXA are lower than those observed in dissection.

Moreover, it is assumed that the BMC estimated by DXA more closely represents the mineral fraction of the bones, that is, the ash content obtained through chemical analysis. This equivalence is supported by the findings of Xavier et al. (2023), who reported an R²CV of 0.98 when comparing the BMC values estimated by DXA in the half carcass with the mineral mass determined by chemical analysis in bovine carcasses.

To assess the strength of association between methods, Pearson correlation analyses were performed between traditional carcass measurements, DXA-derived estimates, and dissection data (Table 2). Dissected lean tissue was strongly correlated with DXA lean estimates (r = 0.91, P < .001), whereas REA showed no significant association with dissected lean (r = 0.16, P > .05). Likewise, dissected fat was highly correlated with DXA fat estimates (r = 0.98, P < .001), while BFT presented a strong but lower correlation with dissected subcutaneous fat (r = 0.82, P < 0.001). These results indicate that DXA-derived estimates are more closely associated with whole-carcass tissue composition than traditional single-point carcass measurements.

Based on these strong associations, regression analyses were subsequently performed to quantify the predictive ability of DXA for carcass tissues. Figure 2 shows the linear regression equations used to contrast the dissection values of lean, fat, and bone in the carcass of cull cows based on the estimates obtained by DXA. The predicted fat tissue by DXA was more accurate and precise due to the highest coefficient of determination (R²CV = 0.96) and a low prediction error (RMSECV = 1.53 kg) than the other contrasted tissues.

Figure 2.

Linear regression equations for predicting the total carcass weight of lean, fat, and bone (kg) from dual-energy X-ray absorptiometry estimates in cull cows. Panels show the relationships between the weights of dissected lean (A), fat (B), and bone (C) and their respective estimates obtained by DXA. DXA lean tissue, DXA adipose tissue, and DXA BMC correspond to estimates of lean, fat, and bone mineral content, respectively. BMC, bone marrow content; DXA, dual-energy X-ray absorptiometry; R²CV, coefficient of determination of cross-validation; RMSECV, root mean square error of cross-validation.

The model for predicting lean mass indicates that 84% of the observed variability was explained. However, it also presented the highest RMSECV (4.31 kg), evidencing greater prediction error. These results suggest that, although the model captures a substantial portion of the variability, its predictive performance is weaker when compared with the fat.

For the bone variable, the model exhibited an R²CV of 0.73, indicating a substantial proportion of the variability in bone mass was captured by the DXA-based prediction. The RMSECV of 1.46 kg reflects a reasonably low prediction error, especially considering the complexity of accurately estimating total bone mass. This performance is consistent with expectations, since DXA quantifies only BMC, while dissection encompasses both the mineral and organic components as well as water content (Scholz et al., 2015). Despite these inherent limitations, the model demonstrates satisfactory predictive capability for bone mass in cull cows.

The results obtained in this study were consistent with those of Calnan et al. (2021), who validated a high-speed DXA prototype using computed tomography as a reference method. These authors reported R²CV values of 0.95 for fat and 0.89 for lean in whole beef cattle carcasses, results very similar to those observed here, which reinforces the reliability of DXA in predicting the physical composition of the carcass, even in different scenarios and with different categories of animals. Moreover, López-Campos et al. (2018) reported higher R²CV values for finished steers (0.99 for lean, 0.98 for fat, and 0.94 for bone) based on DXA scans of carcass sides and their corresponding primal cut. These results were obtained from young steers finished in feedlots and raised under controlled nutritional conditions. Although the values in this study are slightly lower, the results obtained with cull cows remain solid (R²CV = 0.96 for fat, 0.84 for lean, and 0.73 for bone).

The prediction equations by sections (Table 3) reveal variability in the prediction of lean mass by DXA across the different carcass regions. According to the findings, sections S4 (R²CV = 0.78) and S5 (R²CV = 0.76) demonstrated the highest ability to predict total lean mass in the carcass and also had the lowest AICc values, suggesting these regions provide the most accurate reflection of overall lean content. In contrast, section S3 showed the lowest performance (R²CV = 0.31), while sections S1 (R²CV = 0.68) and S2 (R²CV = 0.70) presented intermediate predictive ability. These results highlight that the accuracy of DXA predictions can vary substantially depending on the specific carcass section analyzed. Clarke et al. (1999), working with sheep carcasses, reported a high R²CV (0.96) for the prediction of lean tissue in the leg region when DXA estimates were compared directly with dissection. However, their analysis was limited to a single anatomical region (the leg), which is structurally well-defined and relatively uniform in tissue composition (Luitingh et al., 1962). In contrast, the present study evaluated 5 distinct anatomical sections to predict the entire carcass. The highest predictive capacity for lean mass was observed in sections S4 (R²CV = 0.78) and S5 (R²CV = 0.76), as demonstrated by linear regressions between each sectional lean measurement and total carcass lean mass. These results indicate that these regions are the most informative anatomical predictors of whole-carcass lean composition in cull Nellore cows and can be reliably used for indirect estimation when full-carcass scanning is not feasible.

DXA fat estimation performed well in all the sections analyzed, especially in section S2 (between the 3rd and 7th ribs), which had the best statistical fit (R²CV = 0.88, RMSECV = 2.75, and AICc = 77.73). The other sections also showed high accuracy (R²CV between 0.85 and 0.87), with prediction errors ranging from 2.94 kg to 3.10 kg. These results indicate that fat measurements obtained by DXA in each section consistently explain a substantial portion of the variation in total carcass fat, reflecting the differential deposition of subcutaneous and intermuscular fat depots across the carcass, with some regions accumulating more than others. In particular, S2 encompasses a region where subcutaneous and intermuscular fat tend to accumulate more prominently, which likely contributes to its superior predictive performance. Overall, the study demonstrates the strong predictive capacity of DXA for fat in cull cow carcasses, with consistent performance across sections and a robust explanation of variability, even in the presence of marked individual differences.

The estimation of bone tissue based on BMC derived from DXA showed heterogeneous performance among the carcass sections. Section S1 recorded the best performance (R²CV = 0.80, RMSECV = 1.25 kg), indicating strong explanatory power and low prediction error. Sections S2 (R²CV = 0.66, RMSECV = 1.64 kg) and S4 (R²CV = 0.55, RMSECV = 1.88 kg) showed moderate predictive ability, while S5 had similar, slightly lower performance (R²CV = 0.51, RMSECV = 1.96 kg). Section S3 had the lowest coefficient of determination (R²CV = 0.38) and the highest RMSECV (2.22 kg), although the model remained statistically significant (P = .0015). These results reflect the intrinsic limitations of DXA, which quantifies only BMC, while dissection assesses total bone mass, including organic matter and water. Consequently, predictive performance depends on the regional balance between the mineral and nonmineral components of bones.

Compared to the results of Mercier et al. (2006), who evaluated the predictive capacity of DXA by comparing BMC values with dissected bone mass in individual cuts of sheep (e.g., R²CV = 0.47 in the shoulder and R²CV = 0.43 in the leg), the present study achieved higher coefficients of determination, with and R²CV of 0.80 in the shoulder region (S1) and an R²CV of 0.51 in the leg region (S5). This indicates that the DXA-based BMC provided superior prediction of bone mass in these regions of cull Nellore cows. Regarding total bone mass prediction, Segura et al. (2023) reported an R²CV of 0.82 in adult cow cuts, while the present study obtained an overall R²CV of 0.73, showing that our results are comparable and in close agreement with previously reported values. Moreover, in regions with lower bone quantity and variability, such as the flank, Segura et al. observed reduced predictive performance (R²CV = 0.31), which is in line with the intermediate value found in section S4 (R²CV = 0.55) in the present study. These comparisons demonstrate that DXA can reliably predict bone mass across different anatomical regions, with performance influenced by regional bone distribution and variability.

Also, DXA estimates of fat from the whole half carcass and individual sections were compared to predict total dissected subcutaneous fat of the half carcass (Table 4). The DXA estimate of total fat for the entire carcass showed excellent predictive ability for dissected subcutaneous fat (R²CV = 0.97, RMSECV = 0.94 kg, AICc = 25.94). Among the predictors by section, performance was robust, with R²CV ranging from 0.79 (S1) to 0.92 (S4), RMSECV between 1.44 and 2.26 kg, and AICc between 46.50 and 68.08. Sections S4 and S3 stood out as the best predictions by section (S4: R²CV = 0.92, RMSECV = 1.44 kg; S3: R²CV = 0.90, RMSECV = 1.57 kg), while S1 had the weakest fit (R²CV = 0.79, RMSECV = 2.26 kg). These findings indicate that although DXA estimates total fat (including intramuscular fat), both the total carcass fat estimate and the section estimates on DXA reliably predict the subcutaneous fat obtained by dissection. When whole-carcass scanning is not available, section S4 (and, to a lesser extent, S3) is the best single-section alternative.

These results reinforce the classic use of subcutaneous fat as a low-cost method that can be quickly applied in both industry and nutritional management research (Tait, 2016). At the same time, they demonstrate that DXA not only replicates the reliability of this measurement across many carcass regions but also outperforms it, offering additional advantages in speed, objectivity, and data richness. Altogether, these findings support the use of DXA as a robust tool for experimental determination of carcass composition, offering greater precision, repeatability, and data integrity than traditional destructive methods—particularly in studies requiring accurate phenotyping for genetic selection, nutritional intervention assessment, and methodological validation.

Conclusions

The results of this study demonstrate that the DXA technique can predict the tissue composition of cull Nellore cow carcasses, with particular relevance for lean tissue estimation. DXA-derived lean mass predictions showed satisfactory performance, explaining a substantial proportion of the variability in dissected muscle mass, although with moderate accuracy compared to fat. This moderate performance likely reflects the inherent biological variability of cull cows and the heterogeneous distribution of muscle across the carcass.

Sectional analysis demonstrated that specific carcass regions can reliably represent whole-carcass composition, thereby increasing the practical feasibility of DXA application under industrial and experimental constraints. Section S2 showed slightly higher accuracy for fat estimation, whereas sections S4 and S5 provided the most accurate predictions of lean mass, and section S1 performed best for bone content. Furthermore, the strong agreement between DXA-derived estimates and dissected subcutaneous fat measurements supports DXA as a nondestructive, efficient, and scientifically robust alternative for carcass evaluation.

Conflict of Interest

The authors declare no conflicts of interest.

Author Contribution

Wenderson M. de Carvalho: conceptualization, methodology, investigation, formal analysis, and writing—original draft; Guilherme M. Rodrigues: methodology and investigation; Ana C. A. Luiz: investigation and data curation; Melissa A. B. Gonçalves: investigation and data curation; Vitória M. de S. Ferreira: methodology and investigation; Cris L. de C. Nunes: investigation, data curation, formal analysis, and writing—review and editing; Simone E. F. Guimarães: conceptualization, writing—review and editing, supervision, and funding acquisition; and Mario L. Chizzotti: conceptualization, writing—review and editing, supervision, and funding acquisition.