Dataset Information

Impact Evaluation of Score Classes and Annotation Regions in Deep Learning-Based Dairy Cow Body Condition Prediction.

ABSTRACT: Body condition scoring is a simple method to estimate the energy supply of dairy cattle. Our study aims to investigate the accuracy with which supervised machine learning, specifically a deep convolutional neural network (CNN), can be used to retrieve body condition score (BCS) classes estimated by an expert. We recorded images of animals' rumps in three large-scale farms using a simple action camera. The images were annotated with classes and three different-sized bounding boxes by an expert. A CNN pretrained model was fine-tuned on 12 and 3 BCS classes. Training in 12 classes with a 0 error range, the Cohen's kappa value yielded minimal agreement between the model predictions and ground truth. Allowing an error range of 0.25, we obtained minimum or weak agreement. With an error range of 0.5, we had strong or almost perfect agreement. The kappa values for the approach trained on three classes show that we can classify all animals into BCS categories with at least moderate agreement. Furthermore, CNNs trained on 3 BCS classes showed a remarkably higher proportion of strong agreement than those trained in 12 classes. The prediction precision when training with various annotation region sizes showed no meaningful differences. The weights of our trained CNNs are freely available, supporting similar works.

SUBMITTER: Nagy SA

PROVIDER: S-EPMC9855016 | biostudies-literature | 2023 Jan

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Impact Evaluation of Score Classes and Annotation Regions in Deep Learning-Based Dairy Cow Body Condition Prediction.

Nagy Sára Ágnes SÁ Kilim Oz O Csabai István I Gábor György G Solymosi Norbert N

Animals : an open access journal from MDPI 20230104 2

Body condition scoring is a simple method to estimate the energy supply of dairy cattle. Our study aims to investigate the accuracy with which supervised machine learning, specifically a deep convolutional neural network (CNN), can be used to retrieve body condition score (BCS) classes estimated by an expert. We recorded images of animals' rumps in three large-scale farms using a simple action camera. The images were annotated with classes and three different-sized bounding boxes by an expert. A ...[more]

PMID: 36670733

Similar Datasets

Project description:A transcriptomic approach was used to evaluate potential interactions between prepartum body condition score (BCS) and feeding management in the weeks before calving on hepatic metabolism during the transition period. Thirty-two mid-lactation grazing dairy cows of mixed age and breed were randomly allocated to one of four treatment groups in a 2 × 2 factorial arrangement: two prepartum BCS categories [4.0 (thin, BCS4) and 5.0 (optimal, BCS5); based on a 10-point scale], and two levels of energy intake during the 3 wk preceding calving (75 and 125% of estimated requirements). Liver samples were obtained at -7, 7, and 28 d relative to parturition and subsequent RNA was hybridized to the Agilent 44K Bovine (V2) Microarray chip. The Dynamic Impact Approach was used for pathway analysis, and Ingenuity Pathway Analysis was used for gene network analysis. The greater number of differentially expressed genes in BCS4 cows in response to prepartum feed allowance indicated that these animals were more responsive to prepartum nutrition management than optimally-conditioned cows. However, independent of prepartum BCS, pathway analysis revealed that prepartal feeding level had a marked effect on carbohydrate, amino acid, lipid, and glycan metabolism. Altered carbohydrate and amino acid metabolism indicated a greater and more prolonged negative energy balance post-calving in BCS5 cows overfed prepartum. This was indicated by opposite effects of pre-calving feeding in in BCS4 compared with BCS5 cows in amino acid, vitamin, and co-factor metabolism. The prepartum feed restriction ameliorated the metabolic adaptation to the new lactation in BCS5 cows, while detrimentally affecting BCS4 cows, which seemed to adapt better when overfed. Moreover, alterations in the glycosaminoglycans synthesis pathway support this idea, indicating better hepatic health status in feed-restricted BCS5 cows and overfed BCS4 animals. IPA network analysis indicated liver damage in feed-restricted thin cows, likely due to metabolic overload. Overall, the data indicate that overfeeding in late-pregnancy should be limited to underconditioned animals, while cow with optimal degree of body condition should be maintained on an energy restricted diet.

Project description:The relationship between ewe body condition score (BCS) and liveweight (LW) has been exploited previously to predict the former from LW, LW-change, and previous BCS records. It was hypothesized that if fleece weight and conceptus-free liveweight and LW-change, and in addition, height at withers were used, the accuracy of current approaches to predicting BCS would be enhanced. Ewes born in 2017 (n = 429) were followed from 8 mo to approximately 42 mo of age in New Zealand. Individual ewe data were collected on LW and BCS at different stages of the annual production cycle (i.e., prebreeding, at pregnancy diagnosis, prelambing, and weaning). Additionally, individual lambing dates, ewe fleece weight, and height at withers data were collected. Linear regression models were fitted to predict current BCS at each ewe age and stage of the annual production cycle using two LW-based models, namely, unadjusted for conceptus weight and fleece weight (LW alone1) and adjusted (LW alone2) models. Furthermore, another two models based on a combination of LW, LW-change, previous BCS, and height at withers (combined models), namely, unadjusted (combined1) and adjusted for conceptus and fleece weight (combined2), were fitted. Combined models gave more accurate (with lower root mean square error: RMSE) BCS predictions than models based on LW records alone. However, applying adjusted models did not improve BCS prediction accuracy (or reduce RMSE) or improve model goodness of fit (R 2) (P > 0.05). Furthermore, in all models, both LW-alone and combined models, a great proportion of variability in BCS, could not be accounted for (0.25 ≥ R 2 ≥ 0.83) and there was substantial prediction error (0.33 BCS ≥ RMSE ≥ 0.49 BCS) across age groups and stages of the annual production cycle and over time (years). Therefore, using additional ewe data which allowed for the correction of LW for fleece and conceptus weight and using height at withers as an additional predictor did not improve model accuracy. In fact, the findings suggest that adjusting LW data for conceptus and fleece weight offer no additional value to the BCS prediction models based on LW. Therefore, additional research to identify alternative methodologies to account for individual animal variability is still needed.

Project description:The relationship between energy status and fertility in dairy cattle was retrospectively analyzed by comparing fertility with body condition score (BCS) near artificial insemination (AI; experiment 1), early postpartum changes in BCS (experiment 2), and postpartum changes in body weight (BW; experiment 3). To reduce the effect of cyclicity status, all cows were synchronized with Double-Ovsynch protocol before timed AI. In experiment 1, BCS of lactating dairy cows (n = 1,103) was evaluated near AI. Most cows (93%) were cycling at initiation of the breeding Ovsynch protocol (first GnRH injection). A lower percentage pregnant to AI (P/AI) was found in cows with lower (≤ 2.50) versus higher (≥ 2.75) BCS (40.4 vs. 49.2%). In experiment 2, lactating dairy cows on 2 commercial dairies (n = 1,887) were divided by BCS change from calving until the third week postpartum. Overall, P/AI at 70-d pregnancy diagnosis differed dramatically by BCS change and was least for cows that lost BCS, intermediate for cows that maintained BCS, and greatest for cows that gained BCS [22.8% (180/789), 36.0% (243/675), and 78.3% (331/423), respectively]. Surprisingly, a difference existed between farms with BCS change dramatically affecting P/AI on one farm and no effect on the other farm. In experiment 3, lactating dairy cows (n = 71) had BW measured weekly from the first to ninth week postpartum and then had superovulation induced using a modified Double-Ovsynch protocol. Cows were divided into quartiles (Q) by percentage of BW change (Q1 = least change; Q4 = most change) from calving until the third week postpartum. No effect was detected of quartile on number of ovulations, total embryos collected, or percentage of oocytes that were fertilized; however, the percentage of fertilized oocytes that were transferable embryos was greater for cows in Q1, Q2, and Q3 than Q4 (83.8, 75.2, 82.6, and 53.2%, respectively). In addition, percentage of degenerated embryos was least for cows in Q1, Q2, and Q3 and greatest for Q4 (9.6, 14.5, 12.6, and 35.2% respectively). In conclusion, for cows synchronized with a Double-Ovsynch protocol, an effect of low BCS (≤ 2.50) near AI on fertility was detected, but change in BCS during the first 3 wk postpartum had a more profound effect on P/AI to first timed AI. This effect could be partially explained by the reduction in embryo quality and increase in degenerate embryos byd 7 after AI in cows that lost more BW from the first to third week postpartum.

Project description:Animal robustness may be defined as a complex trait characterizing the ability of an individual to be adapted, productive and healthy under contrasted and fluctuating environmental situations. Such a trait is now considered an essential criterion in order to meet the more ambitious goals of farming sustainability. In ruminants, one of the key mechanisms via which robustness is expressed is the capacity to mobilize or restore body reserves (BR) to cope with the challenges of negative energy balances. The objectives of this work were as follows: 1) to estimate the genetic parameters related to BR dynamics in ewes over successive production cycles and 2) to investigate BR management relationships between different physiological stages. For this, historical individual BW and BCS data from 2,920 phenotyped ewes were used for genetic analysis. The changes in BW (∆BW) and BCS (∆BCS) over time were analyzed. Eight physiological stages were considered to investigate these changes over time: mating, early pregnancy, mid-pregnancy, lambing, early suckling, mid-suckling, weaning, and postweaning. The estimated heritability were low for both ∆BW (h2 = 0.13 to 0.18) and ∆BCS (h2 = 0.04 to 0.16). Moderate to high positive genetic correlations (from 0.48 to 0.91) were obtained between BR mobilization phases and between BR accretion phases. Similarly, moderate to high negative genetic correlations (from -0.36 to -0.75) were estimated between the BR mobilization and accretion periods, suggesting that mechanisms driving BR mobilization and accretion processes were genetically correlated. This is the first study in ruminants that demonstrate that the extent and temporal changes in profiles of BR mobilization and accretion are heritable and genetically linked, indicating that such traits could be considered in genetic programs aimed at improving robustness. Nevertheless, further research is needed for a more comprehensive understanding of BR dynamics, notably by including other physiological parameters (i.e., metabolites and hormones) and additional information on the productive and reproductive life of the ewe.

Project description:Application of manure from antibiotic-treated animals to crops facilitates the dissemination of antibiotic resistance determinants into the environment. However, our knowledge of the identity, diversity, and patterns of distribution of these antibiotic resistance determinants remains limited. We used a new combination of methods to examine the resistome of dairy cow manure, a common soil amendment. Metagenomic libraries constructed with DNA extracted from manure were screened for resistance to beta-lactams, phenicols, aminoglycosides, and tetracyclines. Functional screening of fosmid and small-insert libraries identified 80 different antibiotic resistance genes whose deduced protein sequences were on average 50 to 60% identical to sequences deposited in GenBank. The resistance genes were frequently found in clusters and originated from a taxonomically diverse set of species, suggesting that some microorganisms in manure harbor multiple resistance genes. Furthermore, amid the great genetic diversity in manure, we discovered a novel clade of chloramphenicol acetyltransferases. Our study combined functional metagenomics with third-generation PacBio sequencing to significantly extend the roster of functional antibiotic resistance genes found in animal gut bacteria, providing a particularly broad resource for understanding the origins and dispersal of antibiotic resistance genes in agriculture and clinical settings. IMPORTANCE The increasing prevalence of antibiotic resistance among bacteria is one of the most intractable challenges in 21st-century public health. The origins of resistance are complex, and a better understanding of the impacts of antibiotics used on farms would produce a more robust platform for public policy. Microbiomes of farm animals are reservoirs of antibiotic resistance genes, which may affect distribution of antibiotic resistance genes in human pathogens. Previous studies have focused on antibiotic resistance genes in manures of animals subjected to intensive antibiotic use, such as pigs and chickens. Cow manure has received less attention, although it is commonly used in crop production. Here, we report the discovery of novel and diverse antibiotic resistance genes in the cow microbiome, demonstrating that it is a significant reservoir of antibiotic resistance genes. The genomic resource presented here lays the groundwork for understanding the dispersal of antibiotic resistance from the agroecosystem to other settings.

Dataset Information

Impact Evaluation of Score Classes and Annotation Regions in Deep Learning-Based Dairy Cow Body Condition Prediction.

Publications

Impact Evaluation of Score Classes and Annotation Regions in Deep Learning-Based Dairy Cow Body Condition Prediction.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets