Towards automatic interpretation of sheep ultrasound scans

Abstract

AbstractUltrasound imaging is widely used in animal breeding to provide in vivo estimates of the carcass composition of candidates for selection. Although the technique is less accurate than more recent medical imaging methods, such as X-ray computed tomography and magnetic resonance imaging, it is relatively cheap and mobile. Therefore large numbers of animals can be measured. Most current ultrasound scanners require some degree of manual interpretation of images, which is time consuming and liable to vary both between and within individuals. Hence, this study investigated the automatic interpretation of ultrasound scans from sheep. A computer algorithm is proposed for identifying tissue boundaries. Estimates of tissue depth are shown to be comparable with those obtained by manual interpretation, for images of 72 sheep scanned twice at the position of the 13th thoracic vertebra. The root-mean-square errors of subcutaneous fat depth and m. longissimus muscle depth ivere 0·7 mm and 1·7 mm, respectively.