Automated identification of bones and joints in whole-body CT-scan images of pigs.

Abstract

Computed tomography (CT) images provide fast and accurate non-invasive measurements of anatomy, which are crucial in pig breeding. In the commercial breeding program of Topigs Norsvin, the CT-images are mainly used for estimation of carcass value and for scoring the severity of osteochondrosis lesions in joints. This study presents the first major step towards automated detection of osteochondrosis, using CT images. This involves an anatomic segmentation model that can segment 29 classes of different tissues, like individual bones, muscles, and organs. The algorithm then identifies major joints in the fore- and hindlimbs by detecting the center points of the joints, and this method is validated against manually labeled data. Average distance between labeled and predicted center points was 29 mm (SD = 13mm). The next step will utilize these center points to create bounding boxes for local segmentation models to focus on relevant subsets of voxels, enhancing the detection of lesions in joints. This approach aims to improve the efficiency and accuracy of osteochondrosis quantification in CT scan images of pigs, ultimately -benefiting pig breeding programs by providing detailed and automated phenotypes.