Deep learning can automate chicken tibia-breaking strength quantification to improve animal welfare

Area of Science:

• Animal Science
• Biotechnology
• Veterinary Medicine

Background:

• Bone damage is a significant welfare concern in the poultry industry.
• Current methods for assessing chicken bone strength, such as manual X-ray annotation and post-mortem testing, are time-consuming and destructive.

Purpose of the Study:

• To develop an automated deep-learning pipeline for segmenting chicken tibiotarsus and predicting bone breaking strength from X-ray images.
• To evaluate the accuracy and heritability of image-derived bone strength predictions.

Main Methods:

• A deep-learning pipeline using U-Net was developed to automatically segment chicken tibiotarsus from lateral X-ray images.
• A predictor model was trained using pixel intensities from 916 curated bone images to estimate breaking strength.
• Performance was evaluated using Pearson's correlation and heritability estimates (h²).

Main Results:

• The U-Net model achieved a Dice score of 0.91 for tibiotarsus segmentation.
• The predictor model achieved a maximum Pearson's correlation of 0.74 with measured breaking strength, outperforming manual annotation.
• Image-derived predictions were moderately heritable (h² ≈ 0.16) with a high genetic correlation to physical bone strength.

Conclusions:

• The developed deep-learning workflow offers a rapid, non-invasive, and genetically informative method for assessing chicken bone strength.
• This approach can be integrated into commercial breeding programs to improve bone quality and enhance poultry welfare at scale.