Development of an in vivo pig atlas based on Computed Tomography (CT) for use in pig breeding

During the final part of the project period, we have focused on implementing PigAtlas in a commercial pig breeding program. Our model now works in operation and is put into production in a Google cloud solution. Data is generated continuously, and preliminary results show that variance and genetic parameters are as expected and consistent with previous publications. We are also working on new knowledge in AI, machine learning and annotation as we help us to obtain new and improved phenotypes.

H1: Develop an anatomical atlas for segmenting the carcass from the non-carcass part in a live pig using Computed Tomography (CT). We successfully developed an in vivo anatomical atlas for the carcass part of the pig with the corresponding automatic pipeline. H2: Develop an anatomical atlas for segmenting the major cuts (ham, belly, loin, shoulder) in the carcass part of the live pig. We successfully developed an in vivo anatomical atlas for primal cuts of the pig with the corresponding automatic pipeline. H3: Validate the atlases in #1 and #2 against a reference dissection method. Validation with a reference dissection method was performed. The results were satisfactory, and there were some challenges with the accuracy in the dissection method which has been discussed during the project period. H4: Estimation of genetic parameters for the atlas traits. and test the ability to implement in the Norsvin breeding program for pigs. The traits has been successfully implemented.

Measuring body composition in breeding animals is important for improvement of feed efficiency, growth, health, meat quality and market value of farmed animals. Since 2008, Norsvin has been using Computed Tomography (CT) as a tool for measuring body composition in breeding pigs (boars). CT is an in vivo technique, which measures the body composition of a live pig. In order to move towards the next level of precision, not only focusing on the whole body composition of fat, muscle and bone, we need to get anatomical data from the animal. Yield of cuts (ham, belly, loin and shoulder) and internal organs size and volume, are important traits with respect to efficiency, health and market value of the animals. We need to build automatic algorithms in order to analyze, process and gather anatomical data in a large scale breeding system. Today, we are not able to accurately cut the animal into its anatomical pieces in a large scale and automated way. To be able to distinguish between the different meat cuts anatomical labeled atlases from CT have to be developed and new samples to these labeled atlases have to be recorded automatically by image registration The planned innovation is to implement the use of anatomical atlases from CT in a commercial breeding program for pigs. As far as we know, this is the first attempt to use anatomical atlases from CT in a commercial breeding program, and represents an entirely new way of analyzing images and data from CT scans in a livestock breeding environment. This project will present a novel approach to its kind to study the development and use of anatomical atlases from CT scans of breeding pigs in animal genetics.