Segmentation hierarchies and border features for automatic pregnancy detection in porcine ultrasound
In this paper we present an automatic method for ultrasound-based early detection of pregnancy in pigs, which is a crucial information for commercial pig farming. We employ a strategy of region-based classification within multiple segmentation hierarchies which is able to efficiently find target structures in a large set of possible segmentations. Furthermore, we present a novel set of border features, effectively capturing the border appearance under the noisy and diffuse conditions inherent to ultrasound images. Tested on 802 image series, our detection algorithm reaches a sensitivity/specificity of 84.2%/86.4%.