In vivo estimation of chicken breast and thigh muscle weights using multi-atlas-based elastic registration on computed tomography images

1. This study employed an automated estimation method for quantitatively assessing valuable meat parts in broiler chickens. This involved the segmentation of computed tomography (CT) images through elastic registration, utilising feature and model selection.2. Sixty Tetra HB colour broiler chickens (30 males and 30 females) were randomly selected and examined by CT at 10 weeks of age (live weight: 2560 ± 400 g). The animals were slaughtered, and their breast and thigh muscles were dissected and weighed (thigh and breast weights were 90 ± 19 g and 337 ± 58 g). Multi-atlas registration was used for segmentation, followed by feature extraction (256 features/individual) from the CT images.3. Four different regression analysis techniques (linear, PLS, lasso and ridge) with and without feature selection were applied to the collected data with k-fold cross-validation for estimating the thigh and breast muscle weights. The feature selection produced significantly better results in all cases.4. Among the analysis techniques, lasso and ridge regression performed the best for both muscle groups (thigh and breast muscles). These were as follows: lasso for breast: r2 = 0.993, RMSE = 4.87 g; ridge for breast: r2 = 0.995, RMSE = 4.03 g; lasso for thigh: r2 = 0.976, RMSE = 2.94 g; and ridge for thigh: r2 = 0.965, RMSE = 3.53 g.5. The results demonstrated the effectiveness of the automated method, initially tested on rabbits, in accurately estimating valuable meat parts of broiler chickens. The robust performance of the selected regression models underscores the potential for widespread application in poultry production, offering a reliable and efficient means of quantitative assessment.