Integrating computed tomography and deep learning to assess tibiotarsus cortical bone area for leg health evaluation in purebred male turkeys.
Authors
Affiliations (6)
Affiliations (6)
- Department of Animal Biosciences, University of Guelph, Guelph, Ontario N1G 2W1, Canada.
- Hybrid Turkeys, Kitchener, Ontario N2K 3S2, Canada.
- Department of Food Science, University of Guelph, Guelph, Ontario N1G 2W1, Canada.
- Department of Animal Biosciences, University of Guelph, Guelph, Ontario N1G 2W1, Canada; Lactanet Canada, Guelph, Ontario N1K 1E5, Canada.
- Department of Animal Biosciences, University of Guelph, Guelph, Ontario N1G 2W1, Canada; Department of Animal and Veterinary Sciences, Aarhus University, Tjele 8830, Denmark.
- Department of Animal Biosciences, University of Guelph, Guelph, Ontario N1G 2W1, Canada. Electronic address: [email protected].
Abstract
Compromised leg health is a leading contributor to preslaughter mortality in fast-growing poultry, raising animal welfare concerns and contributing to economic losses. Thus, inclusion of leg health parameters in breeding programs is crucial to ensure a simultaneous improvement of livability and performance. Routine gait scoring and leg defect assessments are laborious and offer limited insight into bone health, making non-invasive imaging technologies like computed tomography (CT), a supplementary tool for phenotyping and genetic improvement. A CT-derived indicator to study bone morphology is tibiotarsal cortical bone area (Ct.Ar), defined as the number of white pixels (cortical bone) multiplied by the size of the image pixel. The objectives of this study were to derive Ct.Ar from CT images using deep learning algorithms, estimate genetic parameters for Ct.Ar, and evaluate correlations between Ct.Ar and other production traits. The dataset included 2 605 purebred turkey toms from line A and 3 337 from line B. At 20 weeks of age, manual measurements of BW and walking score were recorded. The birds were also CT scanned between 16 and 22 weeks of age. A Detectron2 panoptic segmentation deep learning model was developed and implemented in Python 3.11 and trained using transfer learning techniques. Heritability estimates for right and left leg Ct.Ar were low to moderate, ranging from 0.18 ± 0.04 to 0.33 ± 0.04 across both lines. Favorable correlations between Ct.Ar and BW, ranging from 0.14 ± 0.06 to 0.34 ± 0.06, support the hypothesis that bone adapts to increased mass, while unfavorable correlations between Ct.Ar and breast meat yield, ranging from -0.07 ± 0.11 to -0.31 ± 0.09, highlight a trade-off between skeletal robustness and production traits. These findings highlight the complexity of selecting for leg health in turkeys and suggest that Ct.Ar, while informative, does not fully capture bone integrity. Future research should explore additional CT-derived leg health traits, alongside continued model development, to ensure animal welfare, livability, and sustainability, while increasing genetic gain in commercial turkey populations.