Phenomics

Development of a User-Friendly Image Analysis Pipeline for Extraction of Meat Quality Phenotypes

---

Abstract

As global demand for premium beef continues to rise, breeding programs are placing greater emphasis on improving meat quality through genetic selection. Identifying animals with superior carcass traits such as marbling, ribeye area (REA), and fat thickness is critical for enhancing consumer satisfaction and market value. Despite their importance, carcass and meat quality traits are often expensive phenotypes to obtain. Unlike simple traits such as body weight, which can be collected repeatedly on live animals, most carcass traits require post-slaughter evaluation involving specialized equipment, trained personnel, and steak sampling, contributing to high costs. In addition, because these traits are typically measured only once per animal, large sample sizes are required to build robust datasets for genetic analyses. In this context, computer vision and image analysis offer a promising opportunity to address these challenges. The aims of this study were to analyze image data collected during technical slaughter, to quantify REA, ribeye length (REA-Length), ribeye depth (REA-Depth), and subcutaneous fat thickness (SFT) of ribeye steaks using an image processing pipeline, and to evaluate correlations between technician-scored and image-derived measurements. The dataset consisted of approximately 1,000 records from the Longissimus dorsi muscle, collected between the 12th and 13th ribs of Angus × Nellore F1 cross animals (ABS XBlack). Technicians recorded REA measurements, while images of each sample were collected. A Python-based script was developed to process and analyze steak images, resulting in a user-friendly program called MeatImager that requires standardized, user-guided annotation of anatomical features and converts pixel-based measurements into real-world values. Each image generated three labeled outputs (reference object, REA, and SFT) for subsequent analysis. The script successfully quantified REA, REA-Depth, REA-Length, and SFT. Correlations with technician measurements were 0.85 (±0.016) for REA, 0.58 (±0.031) for SFT, 0.37 (±0.052) for REA-Length, and 0.38 (±0.058) for REA-Depth. Mean absolute differences between script-estimated values and technician scores were 0.76 cm² (±5.173) for REA, 1.43 cm (±0.562) for SFT, 1.87 cm (±1.122) for REA-Length, and 1.05 cm (±1.068) for REA-Depth. The higher agreement observed for REA likely reflects the fact that area-based measures integrate information across the entire muscle boundary and are therefore less sensitive to minor annotation differences and steak orientation. In contrast, length- and depth-based traits rely on one-dimensional geometric definitions, making them more sensitive to image positioning and anatomical landmark selection. Variability in image capture conditions further contributed to discrepancies among traits, highlighting the importance of standardized imaging protocols. Overall, these findings demonstrate that image-based pipelines can complement traditional evaluations by providing scalable and cost-efficient measures of carcass traits. Incorporating images into routine data collection offers a cost-effective alternative for carcass phenotyping while supporting genomic selection and accelerating genetic progress toward a more efficient beef industry.

Keywords: 2026

How to Cite:

Amorim, S., Grigoletto, L. & Weng, Z., (2026) “Development of a User-Friendly Image Analysis Pipeline for Extraction of Meat Quality Phenotypes”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2286842. doi: https://doi.org/10.31274/wcgalp.24142

Rights: 1