Lars Erik Gangsei

Automatic segmentation of Computed Tomography (CT) images of domestic pig skeleton using a 3D expansion of Dijkstra's algorithm

Display Omitted The major bones in CT scans of pigs were automatically segmented and identified.Semi-automatic methods might be implemented for reducing errors.The method will constitute an important basis for pig breeding and management. A 3D expansion of Dijkstras algorithm used for automatic segmentation and identification of the bones in CT images of live pigs was developed and validated. The major bones in the skeletons of 208 out of 485 live pigs (43%) were segmented and identified from the images without major errors. The segmentation and identification is executed through 8 main operations: (1) identify the full bone structure by a threshold of Hounsfield units, (2) identify forelimbs by voxel connectivity and set landmarks, (3-8) segment out and identify the individual bones in different main parts of the bone structure by the 3D expansion of Dijkstras algorithm. The algorithms described will constitute an important basis for further work applying CT in pig breeding and management.

Building an in vivo anatomical atlas to close the phenomic gap in animal breeding

Display Omitted A pig atlas, i.e. a 3D model of a pig, was constructed based on CT scans.The major commercial cuts and major organs were identified.The atlas is useful for automatic virtual segmentation of living pigs.The atlas is expected to constitute an important tool for pig breeding. Currently, a growing gap is observed between the enormous amount of genomic information generated from genotyping and sequencing and the scale and quality of phenotypes in animal breeding. In order to fill this gap, new technologies and automated large-scale measurements are needed. Body composition is an important trait in animal breeding related to growth, feed efficiency, health, meat quality and market value of farmed animals. In vivo anatomical atlases from CT will aid large-scale and high-throughput phenotyping in order to reduce some of the gap between genotyping and phenotyping in animal breeding. We demonstrated that atlas segmentation was able to predict major parts and organs of the pig with a numerical test applied to the primal commercial cuts.

Prediction precision for lean meat percentage in Norwegian pig carcasses using ‘Hennessy grading probe 7’: Evaluation of methods emphasized at exploiting additional information from computed tomography

ABSTRACT The present study aims at improving the prediction of lean meat percentage (LMP) for pig carcasses based on on-line measurements from the slaughterhouses using the ‘Hennessy Grading Probe 7’ (HGP7) and auxiliary information such as gender and breed. The prediction performance is evaluated using an empirical Bayes method capable of utilizing information from a surrogate variable, that is, LMP from computed tomography. HGP7 measures thicknesses of fat and meat layers. The HGP7 measurements of subcutaneous fat, sirloin height and interior fat layer should be included as predictor variables together with gender. For efficiency at the slaughter-line gender might be omitted. The empirical Bayes method improved prediction precision only marginally compared with the standard ordinary least-squares method when applied to the full set of data. However, simulations show that the empirical Bayes method enables a considerable reduction of the data sample size without appreciable loss of prediction precision.

Genetic parameters of in vivo primal cuts and body composition (PigAtlas) in pigs measured by computed tomography (CT)1

Abstract Genetic parameters of in vivo primal cuts in breeding pigs using computed tomography were estimated. A total of 2,439 Duroc and 1998 Landrace boars from the Topigs Norsvin boar testing station in Norway were CT scanned as part of the genetic program. In vivo primal cuts were derived from the CT images using atlas segmentation; the method called the Pig Atlas. The (co)variance estimates were obtained from univariate (heritabilities) and multivariate (correlations) animal genetic models using DMU software. The heritabilities for all primal cuts proportions (%) were intermediate to large for both breeds, h2 ranging from 0.15 to 0.50. Negative genetic correlations were found between most of the other primal cuts, and the strongest correlation was between belly and ham. Carcass lean meat percentage showed a positive correlation to shoulder and ham, but was negatively correlated to belly. In this study, in vivo primal cuts from atlas segmentation are used for genetic parameter calculations for the first time. Computed Tomography (CT) makes it possible to measure in vivo body or carcass composition. This will aid the selection response by measuring on the candidates themselves instead of using relatives. Primal cut proportion and composition measured in vivo by computed tomography and atlas segmentation show heritable variation comparable to previous post mortem studies.

Effects of dietary beef, pork, chicken and salmon on intestinal carcinogenesis in A/J Min/+ mice

The International Agency for Research on Cancer has classified red meat as “probably carcinogenic to humans” (Group 2A). In mechanistic studies exploring the link between intake of red meat and CRC, heme iron, the pigment of red meat, is proposed to play a central role as a catalyzer of luminal lipid peroxidation and cytotoxicity. In the present work, the novel A/J Min/+ mouse was used to investigate the effects of dietary beef, pork, chicken, or salmon (40% muscle food (dry weight) and 60% powder diet) on Apc-driven intestinal carcinogenesis, from week 3–13 of age. Muscle food diets did not differentially affect carcinogenesis in the colon (flat ACF and tumors). In the small intestine, salmon intake resulted in a lower tumor size and load than did meat from terrestrial animals (beef, pork or chicken), while no differences were observed between the effects of white meat (chicken) and red meat (pork and beef). Additional results indicated that intestinal carcinogenesis was not related to dietary n-6 polyunsaturated fatty acids, intestinal formation of lipid peroxidation products (thiobarbituric acid reactive substances, TBARS), or cytotoxic effects of fecal water on Apc-/+ cells. Notably, the amount of heme reaching the colon appeared to be relatively low in this study. The greatest tumor load was induced by the reference diet RM1, underlining the importance of the basic diets in experimental CRC. The present study in A/J Min/+ mice does not support the hypothesis of a role of red meat in intestinal carcinogenesis.

Theoretical evaluation of prediction error in linear regression with a bivariate response variable containing missing data

ABSTRACT Methods for linear regression with multivariate response variables are well described in statistical literature. In this study we conduct a theoretical evaluation of the expected squared prediction error in bivariate linear regression where one of the response variables contains missing data. We make the assumption of known covariance structure for the error terms. On this basis, we evaluate three well-known estimators: standard ordinary least squares, generalized least squares, and a James–Stein inspired estimator. Theoretical risk functions are worked out for all three estimators to evaluate under which circumstances it is advantageous to take the error covariance structure into account.

A CT-image based pig atlas model and its potential applications in the meat industry

In this communication we present a novel pig atlas model which is represented by a parametric linear Lagrange or cubic Hermite mesh. The model is developed from data points digitized from a 3D pig CT image. In total 84 muscles and 121 bones are included in the atlas, representing the tissue structures most relevant to the industry. We discuss its potential applications in virtual meat cuts and statistical shape analysis for pig breeding and genetics companies.

The use of deep learning to automate the segmentation of the skeleton from CT volumes of pigs1

Abstract Computed tomography (CT) scanning of pigs has been shown to produce detailed phenotypes useful in pig breeding. Due to the large number of individuals scanned and corresponding large data sets, there is a need for automatic tools for analysis of these data sets. In this paper, the feasibility of deep learning for fully automatic segmentation of the skeleton of pigs from CT volumes is explored. To maximize performance, given the training data available, a series of problem simplifications are applied. The deep-learning approach can replace our currently used semiautomatic solution, with increased robustness and little or no need for manual control. Accuracy was highly affected by training data, and expanding the training set can further increase performance making this approach especially promising.

Monitoring lean meat percentage predictions from optical grading probes by a commercial cutting pattern

The lean meat percentage (LMP) classification in Norwegian slaughterhouses is obtained by Hennessy Grading Probe 7 (HGP7), an optical tool. Even though the HGP7 method is validated frequently, there is industrial and legislative demand to reconsider the applied LMP equation, typically due to the introduction of new breeds. A deboning pilot plant generates precise yield data using cutting and deboning stratified pork carcasses by a specific commercial cutting pattern (CCP) at an annual rate of approximately 250 slaughter pigs. This paper shows how results obtained by CCP can be used to measure LMP in pork and how these results can be used for monitoring the quality of LMP predicted by HGP7. The effect of gender, maternal- and paternal lines on validity of HGP7 predictions was evaluated. The effect of introducing a new maternal line (TN70) seems to be substantial, whereas the effects of the tested paternal lines are small to negligible.