102. Software Development to Measure Ultrasonic Images in Beef Cattle

Abstract Ultrasound imaging has been a critical tool in beef cattle genetic selection for over 30 years, enabling producers to evaluate carcass traits in live animals. The goal of this study was to develop and test software capable of measuring ribeye area (REA) and predicting intramuscular fat (IMF) from ultrasound images using artificial intelligence (AI), reducing variability among laboratories and technicians. Data were collected annually from 2021 to 2024 (n = 267) during the Ultrasound Guidelines Council (UGC) Field Certification events hosted by West Texas A&M University. Diverse groups of cattle, differing in breed, age, and finishing stage, were scanned using approved ultrasound devices. Mean REA and live weight were 15.253 ± 2.148 in2 and 1,407.333 ± 173.876 lb., respectively. A reference technician captured images of the ribeye, IMF, and rump fat according to UGC protocol. Carcass data were collected at a commercial beef processor facility in the Texas Panhandle, and tissue samples were analyzed for chemical IMF at a third-party laboratory. The images were labeled using polygon segmentation in Label Studio to produce coordinate data for AI model training. Initial software development focused on computer vision and supervised deep learning models for feature extraction. Model development focused on image segmentation and feature extraction to identify correlations between pixel intensity and marbling content. For the regression equation, live weight and AI REA were included to evaluate the prediction of final REA. The AI model was a significant predictor (P value < 0.001) of actual REA. Pearson’s correlation between actual and AI-predicted values was 0.638, while Spearman’s rank correlation was 0.622, indicating a moderately strong and consistent association between the two variables. AI-derived software is considered to be an accurate predictor of final carcass ribeye size.