Phenomics

Genetic Parameters Estimation and Relationships Between Image-Derived Morphometric and Sensor-Based Behavioral Traits in Holstein Cows

---

Abstract

Linear type traits influence production, reproduction, and longevity in dairy cattle. These traits are manually assessed through subjective visual scores or labor-intensive measurements that require animal restraint, limiting recording frequency. Digital phenotyping based on computer vision can complement traditional linear type evaluations by enabling non-invasive, continuous, and remote measurement of body conformation dimensions without disrupting management routines. Despite this potential, knowledge of genetic parameters for image-derived traits and their relationships with productivity indicators such as rumination remains limited. Estimating these parameters is essential for evaluating selection potential and guiding breeding decisions. This study aimed to estimate genetic parameters for image-derived morphometric traits and sensor-based rumination and activity traits. Using an Intel RealSense D435 camera, 276,642 top-down images were collected from 918 Holstein cows between September and November 2024 at a commercial dairy in California. A YOLOv8 keypoint detection model was trained on images from September and applied to subsequent images to infer body landmarks on the hips, rump, shoulders, and midsagittal plane. Pixel distances between predicted keypoints were used to derive six digital morphometric traits: rump width (RW), body length (BL), hip width (HW), shoulder width (SW), and right and left rump lengths (RLR and RLL). Daily rumination time (DRT) and activity time (DAT) were collected using wearable sensors. The final dataset contained 22,193 daily averages for image-derived traits and 71,887 observations for sensor-based traits. Animals were genotyped using four Zoetis SNP-chip densities, and lower-density genotypes were imputed using FImpute v2. Variance components were estimated with multi-trait GBLUP repeatability models implemented in BLUPF90+. Heritability (h2) estimates for digital morphometric traits ranged from 0.32 ± 0.04 (BL) to 0.53 ± 0.07 (HW), while repeatability (t2) values ranged from 0.51 ± 0.01 to 0.90 ± 0.01. Except for BL, t2 estimates were greater than 0.74 for all morphometric traits, indicating strong within-animal consistency in image-derived measurements across days. Although no manual morphometric measurements were collected, these genetic parameter estimates support the biological relevance of the digital traits. Genetic correlations among image-based traits ranged from 0.77 to 0.99, indicating that selection for any individual morphology trait would result in indirect genetic gains for the others. Genetic associations between morphometric and sensor-based traits were weak to moderate, ranging from 0.17 (RW) to 0.32 (RLR) for DRT, and between -0.16 (RW) to -0.09 (SW) for DAT. Rumination time had low heritability (0.09 ± 0.02) and a moderate negative genetic correlation with DAT (-0.36 ± 0.01), indicating a genetic trade-off between these behaviors. Overall, image-derived morphometric traits are heritable and repeatable, supporting their use as complementary conformation traits in genetic selection programs. The modest genetic associations with rumination and activity suggest minimal correlated behavioral responses to selection on digital morphometric traits.

Keywords: 2026

How to Cite:

Abdulrahman, T., Saha, C., Costa, R., Medeiros, B., Hooker, J., Bussiman, F. & Alves, A., (2026) “Genetic Parameters Estimation and Relationships Between Image-Derived Morphometric and Sensor-Based Behavioral Traits in Holstein Cows”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2286926. doi: https://doi.org/10.31274/wcgalp.24169

Rights: 1