Estimating backfat depth, loin depth, and intramuscular fat percentage from ultrasound images in swine

Fast, accurate, and reliable estimates of backfat depth, loin depth, and intramuscular fat percentage in swine breeding stock are used to increase genetic improvement and farm profitability. The objective of this study was to develop an equation-based model for the estimation of swine backfat depth, loin depth, and intramuscular fat percentage estimates obtained from longitudinal ultrasound images. Images were collected from purebred Duroc (n = 230), purebred Large White (n = 154), and commercial (n = 190) pigs born in January 2021 at three farms located in North Carolina. An Exapad ultrasound machine captured longitudinal images across the 10th to 13th ribs at 182 (±12.8 SD) days of pig age. The total number of images processed for Duroc, Large White, and commercial pigs was 1 385, 928, and 1 168 images, respectively. To establish a standard measurement for model comparison, trained personnel following standard company procedures using the BioSoft Toolbox (v4.0.1.2; Biotronics Inc., Ames, IA) obtained backfat and loin depth measurements from the images. Longissimus muscle intramuscular fat percentage was predicted using near-infrared spectroscopy at approximately 22 h postmortem. Backfat and loin depth estimation were conducted only for commercial pigs (n = 190) while intramuscular fat estimation was conducted on all pigs (n = 574). Average backfat depth, loin depth, and intramuscular fat percentage were 14.6 (±2.6 SD) mm, 63.7 (±5.5 SD) mm, and 2.21 (±0.82 SD) %. Image analysis and estimation model development were conducted in MATLAB R2021a. Edge detection via the image gradient was applied to segment ultrasound images into backfat, loin, and rib regions. Segmented images were used to estimate backfat depth, loin depth, and loin intramuscular fat percentage. After image quality control and filtering, the image inclusion rate for each breed-trait combination ranged from 76 to 97%. All Duroc and commercial pigs and 97% of Large White pigs were represented by at least one image for trait estimation. Coefficient of determination of models for the estimation of backfat depth, loin depth, and intramuscular fat percentage were 0.58, 0.57, and 0.56, respectively. Root mean square error of backfat depth, loin depth, and intramuscular fat estimation were 1.65 mm, 3.58 mm, and 0.54%, respectively. Results demonstrate the feasibility of using ultrasound image gradient and an equation-based approach to estimate swine backfat and loin depth, and intramuscular fat percentage. This equation-based approach to estimate carcass traits in live swine can enhance genetic improvement.