Adaptive weight estimation and segmentation of beef carcass tissues from computed tomography images

Meat yield and quality are often measured through the weight and distribution of muscle and fat and determine the value of a carcass and hence play a crucial role in the meat industry. Traditional methods such as physical dissection and chemical analysis for carcass composition (muscle, fat, bone) are often destructive, time-consuming, and the accuracy is operator dependent. As a credible alternative, computed tomography (CT) scanning of the carcass enables automated weight estimation of different carcass tissues. A novel Adaptive Segmentation and Weight estimation (ASW) method was developed to infer HU thresholds in the absence of physical gold standard, allowing robust classification and weight estimation of fat, muscle, and bone tissues from CT images obtained under varying protocols, based on the frequency distribution of voxel HU values. Using spiral CT technology, the study scanned 175 beef carcass cuts from nine Angus, seven Brahman, and nine Charolais steer carcasses. The ASW method demonstrated high precision and accuracy with three different CT protocols compared to the fixed threshold weight estimation method. Lin's concordance correlation coefficient (Lin's CCC) against physical dissection was above 0.997 for total carcass weight using ASW estimation, and above 0.974 for fat and muscle tissue. Further, the present method achieved high efficiency in determining thresholds within 15 s and it is therefore recommended to process CT images for tissue segmentation. It is concluded that the AWS method developed in the present study offers advantages over traditional, fixed threshold methods.