Automated pose estimation reveals walking characteristics associated with lameness in broilers

Area of Science:

• Animal Science
• Biotechnology
• Agricultural Engineering

Background:

• Selective breeding for improved broiler walking ability requires large-scale, objective data.
• Current expert-based gait scoring is time-consuming and subjective.
• Precision phenotyping tools offer a high-throughput alternative for assessing broiler locomotion.

Purpose of the Study:

• To investigate the link between specific walking characteristics derived from pose estimation and broiler gait.
• To evaluate the potential of deep learning-based pose estimation for objective gait phenotyping in broilers.

Main Methods:

• Filmed male broilers walking in a corridor at 14, 21, and 33 days of age.
• Utilized a DeepLabCut model to track 8 key body points for pose estimation.
• Quantified 7 pose features during double support and maximum leg lift phases of walking.
• Gait scored by experts (0-5 scale) and broilers classified as good or suboptimal gait.

Main Results:

• Broilers with suboptimal gait exhibited sharper hock joint lateral angles and lower hock-feet distance ratios during double support.
• Suboptimal gait was associated with lower relative step height during the stepping phase.
• Step height and hock-feet distance ratio showed the most significant deviations in broilers with suboptimal gait.

Conclusions:

• Pose estimation is a viable tool for assessing broiler walking characteristics throughout their productive life.
• This technology can objectively phenotype and monitor broiler gait, supporting selective breeding programs.
• Insights gained can enhance understanding of lameness and improve gait prediction models in broilers.