The costliest1 and most frequent2 disease affecting the North American beef cattle industry is bovine respiratory disease (BRD). In the dairy industry, BRD has a similar effect as a common cause of morbidity in calves.3 The etiological components of BRD are multifactorial, including viral and bacterial pathogens, host genetics and immune patterns, and environmental factors and stressors,4,5 making the disease difficult to prevent and subjective to diagnose both retrospectively and antemortem. In an evaluation of clinical diagnosis of BRD in feedlot cattle, lay health workers showed a low diagnostic sensitivity when visually assessing for disease.6 In a study7 observing calves’ clinical illness scores, researchers found only slight agreement among the scores assigned by trained veterinarians. This subjectivity creates further difficulty in understanding clinical symptoms, ensuring judicious use of antimicrobials, and stopping the spread of disease.
Machine learning (ML) algorithms, AI sensors, and precision technologies have great utility for the detection and classification of disease states in various stages of both beef and dairy production. Artificial intelligence, a field of science in which computing tasks are designed to mimic human-like cognitive thinking,8 encompasses aspects from sensors to ML algorithms built upon this idea of cognition. Machine learning techniques can compile a broad variety of data, making them ideal for the analysis of real-life data that possess spatial dimensionality.9 The ability to objectively evaluate large datasets, as well as the efficiency and labor-saving nature of utilizing technology to complete tasks, makes ML an excellent solution candidate for the diagnostic problems traditionally associated with BRD. In this review, we address specific studies of applications of ML technologies and techniques in BRD diagnostics in the past 10 years, highlighting both ML algorithms utilized to perform data analysis, including disease classification and predictive epidemiological modeling, and technologies that have built-in utilization of ML algorithms, including computer-aided veterinary equipment, automated feeders, and wearable sensors.
Machine learning modeling in BRD diagnosis
Disease classification and pathogen identification models
Because of its multiple factors, classifying BRD by pathogenic cause is difficult, if not impossible, through a gross visual examination. Traditionally, diagnostic laboratories utilize PCR testing to identify specific pathogenic causes of BRD, like Mannheimia haemolytica.10 More recently, metagenomic sequencing has become an option to identify a multitude of bacterial pathogens in a bovine sample. Using graph representations and deep learning modeling, one study10 identified important features, which were assigned to ML nodes and graphs, then used deep learning classification to detect complex metagenomic signatures for a BRD-causing pathogen. The maximum classification accuracy in the study, 89.74%, was realized by the deep learning classification task with 10,000 samples; however, this accuracy is much higher than many of the study's10 other results. Using ML approaches to detect BRD-associated pathogens in metagenomic sequencing can assist in locating the pathogen associated with a specific BRD complex and relate that information to other information ascertained in host sequencing, like the host's response to and interaction with a specific pathogen. Additionally, with a baseline of signatures developed by the common BRD pathogens, researchers can identify signatures that do not fit those known pathogens as well, perhaps identifying newly associated components of BRD.
Similarly, identifying the genomic response of a host to specific pathogenesis of BRD can aid in the diagnosis of disease and selection of treatment protocol. In one study11 of sample collections from cattle experimentally challenged with 1 of 5 BRD pathogens (bovine respiratory syncytial virus, bovine viral diarrhea virus, infectious bovine rhinotracheitis, M haemolytica, or M bovis) or sham controls, researchers used supervised ML to evaluate and differentiate the host response to each viral and bacterial pathogen used in the experimental challenges. This study11 used 6 ML models, sparse Poisson linear discriminant analyses (with and without a power transformation),12 negative binomial linear discriminant analysis,13 sparse variance modeling at the observational level–based nearest shrunken centroids,14 support vector machine, and nearest shrunken centroids, to classify cattle by the pathogen with which they were challenged and extract features within the host transcriptome that differentiated infections by each pathogen. The support vector machine modeling performed best in terms of balanced accuracy when classifying by challenge for all challenges but bovine viral diarrhea virus, and the authors noted that viral classification datasets yielded greater accuracies across models than those including bacterial challenges.11 Understanding the host response to the viral and bacterial components of BRD, along with classifying the host response by a specific pathogen, specifically in combination with ML that can identify signatures of those same pathogens, results in improved capabilities for identifying the associated agents of a specific BRD case.
While the previous 2 analytical applications of ML were applied to datasets with information on DNA and RNA molecules, ML has also recently been utilized on a gross pathology level to differentiate syndromes diagnosed as BRD. Bronchopneumonia, interstitial pneumonia, and bronchopneumonia with an interstitial pattern are all respiratory syndromes affecting cattle that are clinically diagnosed with BRD. Immediately upon necropsy, these syndromes are often diagnosed grossly, but these diagnoses are later confirmed or reclassified via histopathology. Classification ML using MobileNet (light-weighted model for mobile applications), SE-ResNeXt (squeeze and excitation networks), ResNet50 (residual networks), and ViTl16r224 (vision transformer networks), 4 models available through the Microsoft Azure ML Studio, was completed in one study15 to diagnose gross lung images taken at necropsy of deceased feedlot cattle, training models with either a gross or histopathological diagnosis. The best image classification models varied in accuracy, ranging from 38% to 86%, which the authors attributed to a lack of uniqueness in the gross features of each syndrome and the potential of camera and image variation.15 However, with method and model refinement, an ML-aided field diagnosis can lead producers and veterinarians to better understand the specific respiratory syndrome causing mortality within their herds.
Epidemiological models
Epidemiologists have developed dynamic models to prevent and control disease and track the progression and spread of ongoing disease outbreaks. In the case of BRD, timely on-farm decision making by the affected producer in collaboration with their veterinarian leads to efficient diagnosis and prevention of further spread. Leveraging an existing BRD mechanistic model,16 a study17 used a combination of AI techniques to develop a user-friendly platform for farmers and veterinarians. This platform, when deployed on-farm, could provide customized scenarios, potential economic impacts, and pathogenic effects in a user-friendly decision-support tool, following the framework of domain-specific knowledge and a multilevel agent-based simulation architecture ML model.17 By utilizing AI to give farmers and veterinarians understandable access to a veterinary epidemiological model, those individuals controlling a BRD outbreak can make more informed decisions to better manage their herds’ health status.
Prediction models
Producers in many cattle industry sectors need to predict BRD occurrences with high accuracy. Especially in the feedlot phase of beef cattle production, the early prediction of cohorts of cattle that will experience high respiratory morbidity could allow for a more precise management strategy. By training ML models with arrival and 15-day feed delivery data, a study18 classified cohorts of cattle into high- or low-morbidity categories using advanced perceptron,19 decision forest,20 logistic regression, neural network,21 and boosted decision tree predictions.22 The decision forest was the most specific for low-morbidity lots but showed low sensitivity (33%), whereas the logistic regression and neural network models resulted in the best identification of high-morbidity cattle lots.18 With the inclusion of arrival characteristics and feed consumption data, which have been shown to traditionally associate with morbidity level, ML models developed morbidity predictions by compiling inputs and outcomes of the training data. While the results of this specific study leave room for improvement in the identification of morbidity level, understanding the morbidity level early in the feeding period could provide insights for preventative treatment and management strategies.
The prediction of prognosis (final outcome) is equally important for the management of animals affected with BRD; after treatment, cattle can recover and return to their cohort for finishing or be removed from the feedlot due to chronicity or mortality. While the previous study used ML to predict morbidity, another study23 trained the same ML models to predict the final outcome of cattle after the first treatment for BRD. All of the models included performed similarly, with higher specificity, which signified cattle finishing the feeding period, than sensitivity, which authors attributed to the infrequency of cattle not finishing the feeding phase, a bias toward the majority class.23 The results of the study displayed wide, and often low, ranges of evaluation metrics, with the authors noting that this specific strategy for prediction modeling was not ready for implementation.23 However, with the ability to input individual calf information and cohort descriptive characteristics into ML models, researchers, and eventually feedlots, could make management decisions based on predictions for full recovery and finishing the feeding phase.
Machine learning–assisted technology for BRD diagnosis
Veterinarian-operated technology
Recent advancements in veterinarian-operated technology with an ML-based algorithm allow veterinarians to make better informed decisions when combined with traditional diagnostic tools, such as thermometers. One such technology is the computer-aided lung auscultation (CALA) stethoscope device, which uses an ML algorithm to calculate lung scores based on the lung sounds recorded.24–26 Early field results of this device26 suggested that the ML aid, in combination with a traditional detection of fever, produced a more refined case diagnosis and, therefore, a better prognosis prediction. Additionally in feedlot cattle, another study25 found that the CALA device agreed with experienced veterinarians in the diagnosis of BRD, with a κ value of 0.77, and increased the sensitivity of diagnosis as compared to pen riders, with sensitivities of 92.9% and 62%, respectively. Computer-aided lung auscultation devices could be utilized by a veterinarian or trained caretaker to better diagnose animals as well as provide a prognosis to assist in treatment protocol development.
While the CALA device was originally used at treatment to score BRD presence and severity, one study27 utilized a proprietary ML algorithm to account for CALA scores, temperature, weight, and other important risk factors of BRD to develop on-arrival thresholds (both conservative and aggressive) for the implementation of preventative treatment protocols. For multiple sites, these technology-guided thresholds for preventative treatment resulted in a statistically significant reduction in BRD control antimicrobial use, decreased morbidity, and increased average daily gain in cohorts of cattle that were screened with at-arrival CALA algorithms as compared to control cohorts.27 The at-arrival potential for the utilization of the CALA device would allow for more informed management decisions and a potential disease reduction throughout the operation in which it is used; however, the use of risk factors for BRD prediction is difficult due to the multifactorial nature of the disease complex.
Sensor technology
Automated feeders and spatial sensors that rely on radio frequency identification tags on animals; motion-based sensors, such as pedometers and accelerometers; and on-animal biosensors utilize ML algorithms to prevent and diagnose respiratory disease in beef and dairy operations. While the specific modality of ML implemented in each of these technologies differs, the main goal of each is to detect changes in feeding or behavior patterns to develop an early diagnosis of respiratory disease.
Tag transmitters and other central hardware that track location and movement in pens, behavioral patterns, and assumed social behaviors use software as an application of ML for remote early disease identification.28,29 This ML uses a combination of multiple proprietary algorithms to make decisions and recommend pulling cattle to treat based on their behaviors and the deviation of these behaviors from normal.28 A Bayesian analysis29 comparing visual observation, which has an assumed low sensitivity due to its subjective nature, and the remote monitoring concluded that the ML algorithms had a greater sensitivity, positive predictive value, and negative predictive value in BRD prediction in beef calves when compared to traditional visual observation. The use of sensors with built-in ML algorithms allows commercial operations to leverage the utility of ML technologies without having to develop, train, and evaluate the models themselves.
In contrast with those systems using tag transmitters and central hardware with built-in ML capabilities, other studies have implemented raw accelerometer and automated feeder data into an ML algorithm. One study30 utilized collar-mounted sensors, similar to the tag system of previously described remote monitoring, to detect calf location, walked distances, and a variety of social and contact behaviors, and automated feeder data, to identify factors like number of meals, time spent at the feeder, and feeding speed. Researchers incorporated these features into multiple ML algorithms to test their performance in BRD detection, with a gradient-boosting machine classification algorithm producing the highest detection accuracy of 77% using a majority of movement features for prediction training.30 Other studies31–34 have applied ML to detecting BRD using features collected by automated feeders and accelerometers applied to the legs of calves, implementing convolutional neural network, K-nearest neighbor, and polynomial cost-aware learning feature models for analysis. All 4 of these studies31–34 using automated feeders and accelerometers produced accuracies of about 80% for detecting BRD with their respective models. Another study35 used the same precision technologies and research design in preweaned calves but used linear mixed models to determine at which point before BRD diagnosis changes in data from the technologies occurred. Another study36 utilized, again, the same study design with ML analyses, but rather than the accelerometer data, they included activity behaviors manually collected and timed from video footage. This study used a random forest ML algorithm and a moving average calculation, as well as their combination, to develop BRD predictions with high specificity values, all over 95%, but relatively low balanced accuracies, under 75%.36 While these studies require more scientific and mathematical background for implementation than the previously discussed remote identification systems, they are often completed with a future application in mind, focusing on prediction or early detection with technologies or algorithms that will eventually be available commercially.
Finally, a study37 of sensor-based precision technologies and supervised ML analysis used a wireless sensor applied to the base of the tail of Japanese black calves to detect surface temperature. Using random forest modeling, the study developed associations between the temperature values given by the sensors and rectal temperatures to better define and detect fever periods,37 which are arguably the most objective clinical sign of BRD. The wearable sensors were 98.8% accurate in properly detecting cattle with a fever.37 The detection of fever by a wearable sensor allows for faster diagnosis in animals that may not show outward clinical signs as well as decreases labor requirements for caretakers and veterinarians.
Conclusions
The automation of analysis, scalability to operations, and complex capabilities of ML and AI could make them excellent tools in the hands of those in the cattle industry when faced with the long-standing task of predicting, detecting, and diagnosing BRD. The innovations of ML are outlined in this review, displaying the many ways researchers and private companies alike have implemented ML-guided precision technologies and ML-driven algorithms to tackle BRD from a variety of angles. However, many of the algorithms and technologies addressed in this review lack the performance capabilities for application to commercial production at this time. While the performance metrics of some of the models are lacking, many do show promise for improvements that would allow for implementation in the future. It is important to note that ML models can be improved and altered, especially as technology capabilities continue to improve, and thus the strategies proposed throughout this review can serve as a framework for future innovations. Data management, including computational size requirements, various data formats, and the ability to upload data in a timely manner, is a challenge that must be overcome before the widespread application of ML in BRD diagnostics. Nonetheless, the efficiency that can be gained by faster diagnosis times, reduced animal management, and increased computational analysis speed that may be ascertained by the implementation of ML could move cattle industries toward a cost-effective and labor-saving BRD management strategy. The insights that the technologies can provide, especially in combination with traditional methods, can better define the way we predict, detect, and describe BRD, along with the way we come to these conclusions.
Acknowledgments
None reported.
Disclosures
The authors have nothing to disclose. No AI-assisted technologies were used in the composition of this manuscript.
Funding
The authors have nothing to disclose.
References
- 1.↑
Griffin D. Economic impact associated with respiratory disease in beef cattle. Vet Clin North Am Food Anim Pract. 1997;13(3):367–377. doi:10.1016/S0749-0720(15)30302-9
- 2.↑
USDA. Part IV: Health and Health Management on U.S. Feedlots with a Capacity of 1,000 or More Head. 2013. Accessed January 30, 2025. https://www.aphis.usda.gov/sites/default/files/feed11_dr_partiv.pdf
- 3.↑
USDA. Morbidity and mortality in U.S. preweaned dairy heifer calves NAHMS dairy 2014 study calf component. 2021. Accessed January 30, 2025. https://www.aphis.usda.gov/sites/default/files/morb-mort-usprewean-dairy-heifer-nahms-2014.pdf
- 4.↑
Earley B, Buckham Sporer K, Gupta S. Invited review: relationship between cattle transport, immunity and respiratory disease. Animal. 2017;11(3):486–492. doi:10.1017/S1751731116001622
- 5.↑
Grissett GP, White BJ, Larson RL. Structured literature review of responses of cattle to viral and bacterial pathogens causing bovine respiratory disease complex. J Vet Intern Med. 2015;29(3):770–780. doi:10.1111/jvim.12597
- 6.↑
Timsit E, Dendukuri N, Schiller I, Buczinski S. Diagnostic accuracy of clinical illness for bovine respiratory disease (BRD) diagnosis in beef cattle placed in feedlots: a systematic literature review and hierarchical Bayesian latent-class meta-analysis. Prev Vet Med. 2016;135:67–73. doi:10.1016/j.prevetmed.2016.11.006
- 7.↑
Amrine DE, White BJ, Larson R, Anderson DE, Mosier DA, Cernicchiaro N. Precision and accuracy of clinical illness scores, compared with pulmonary consolidation scores, in Holstein calves with experimentally induced Mycoplasma bovis pneumonia. Am J Vet Res. 2013;74(2):310–315. doi:10.2460/ajvr.74.2.310
- 8.↑
Holzinger A, Langs G, Denk H, Zatloukal K, Müller H. Causability and explainability of artificial intelligence in medicine. WIREs Data Min Knowl. 2019;9(4):e1312. doi:10.1002/widm.1312
- 9.↑
Arel I, Rose DC, Karnowski TP. Deep machine learning - a new frontier in artificial intelligence research [research frontier]. IEEE Comput Intell Mag. 2010;5(4):13–18. doi:10.1109/MCI.2010.938364
- 10.↑
Narayanan S, Ramachandran A, Aakur SN, Bagavathi A. GraDL: a framework for animal genome sequence classification with graph representations and deep learning. In: 2020 19th IEEE International Conference on Machine Learning and Applications (ICMLA). IEEE; 2020:1297–1303. doi:10.1109/ICMLA51294.2020.00203
- 11.↑
Scott MA, Woolums AR, Swiderski CE, Perkins AD, Nanduri B. Genes and regulatory mechanisms associated with experimentally-induced bovine respiratory disease identified using supervised machine learning methodology. Sci Rep. 2021;11(1):22916. doi:10.1038/s41598-021-02343-7
- 12.↑
Buttenschoen K, Kornmann M, Berger D, Leder G, Beger HG, Vasilescu C. Endotoxemia and endotoxin tolerance in patients with ARDS. Langenbecks Arch Surg. 2008;393(4):473–478. doi:10.1007/s00423-008-0317-3
- 13.↑
Dong K, Zhao H, Tong T, Wan X. NBLDA: negative binomial linear discriminant analysis for RNA-seq data. BMC Bioinformatics. 2016;17(1):369. doi:10.1186/s12859-016-1208-1
- 14.↑
Zararsiz G, Goksuluk D, Klaus B, et al. VoomDDA: discovery of diagnostic biomarkers and classification of RNA-seq data. PeerJ. 2017;5:e3890. doi:10.7717/peerj.3890
- 15.↑
Bortoluzzi E, Schmidt P, Brown R, et al. Image classification and automated machine learning to classify lung pathologies in deceased feedlot cattle. Vet Sci. 2023;10(2):113. doi:10.3390/vetsci10020113
- 16.↑
Picault S, Huang YL, Sicard V, Arnoux S, Beaunée G, Ezanno P. EMULSION: transparent and flexible multiscale stochastic models in human, animal and plant epidemiology. PLoS Comput Biol. 2019;15(9):e1007342. doi:10.1371/journal.pcbi.1007342
- 17.↑
Picault S, Niang G, Sicard V, Sorin-Dupont B, Assié S, Ezanno P. Leveraging artificial intelligence and software engineering methods in epidemiology for the co-creation of decision-support tools based on mechanistic models. Prev Vet Med. 2024;228:106233. doi:10.1016/j.prevetmed.2024.106233
- 18.↑
Heinen L, Lancaster PA, White BJ, Zwiefel E. Evaluation of predictive models to determine total morbidity outcome of feedlot cattle based on cohort-level feed delivery data during the first 15 days on feed. Transl Anim Sci. 2022;6(3):txac121. doi:10.1093/tas/txac121
- 19.↑
Park YS, Lek S. Artificial neural networks. In: Jørgensen SE. ed. Developments in Environmental Modelling. Vol 28. Elsevier; 2016:123–140. doi:10.1016/B978-0-444-63623-2.00007-4
- 20.↑
Ho TK. Random decision forests. In: Proceedings of 3rd International Conference on Document Analysis and Recognition. IEEE; 1995:278–282.
- 21.↑
Chan KY, Abu-Salih B, Qaddoura R, et al. Deep neural networks in the cloud: review, applications, challenges and research directions. Neurocomputing. 2023;545:126327. doi:10.1016/j.neucom.2023.126327
- 22.↑
Coadou Y. Boosted decision trees. In: Calafiura P, Rousseau D, and Terao K, eds. Artificial Intelligence for High Energy Physics. World Scientific; 2022:9–58.
- 23.↑
Heinen L, White BJ, Amrine DE, Larson RL. Evaluation of predictive models to determine final outcome for feedlot cattle based on information available at first treatment for bovine respiratory disease. Am J Vet Res. 2023;84(10):1–8. doi:10.2460/ajvr.23.05.0094
- 24.↑
Booker CW, Kim GK, Grimson TM, Hill KT, Wildman BK, Nickell JN. Association between computer-aided lung auscultation and treatment failure risk in calves treated for respiratory disease. Can Vet J. 2021;62(5):511–514.
- 25.↑
Mang AV, Buczinski S, Booker CW, Timsit E. Evaluation of a computer-aided lung auscultation system for diagnosis of bovine respiratory disease in feedlot cattle. J Vet Intern Med. 2015;29(4):1112–1116. doi:10.1111/jvim.12657
- 26.↑
Noffsinger T, Brattain K, Quakenbush G, Taylor G. Field results from Whisper stethoscope studies. Anim Health Res Rev. 2014;15(2):142–144. doi:10.1017/S1466252314000218
- 27.↑
Nickell JS, Hutcheson JP, Renter DG, Amrine DA. Comparison of a traditional bovine respiratory disease control regimen with a targeted program based upon individualized risk predictions generated by the Whisper On Arrival technology. Transl Anim Sci. 2021;5(2):txab081. doi:10.1093/tas/txab081
- 28.↑
White B, Goehl D, Amrine D. Comparison of a remote early disease identification (REDI) system to visual observations to identify cattle with bovine respiratory diseases. Intern J Appl Res Vet Med. 2015;13(1):23–30.
- 29.↑
White BJ, Goehl DR, Amrine DE, Booker C, Wildman B, Perrett T. Bayesian evaluation of clinical diagnostic test characteristics of visual observations and remote monitoring to diagnose bovine respiratory disease in beef calves. Prev Vet Med. 2016;126:74–80. doi:10.1016/j.prevetmed.2016.01.027
- 30.↑
Bushby EV, Thomas M, Vázquez-Diosdado JA, Occhiuto F, Kaler J. Early detection of bovine respiratory disease in pre-weaned dairy calves using sensor based feeding, movement, and social behavioural data. Sci Rep. 2024;14(1):9737. doi:10.1038/s41598-024-58206-4
- 31.↑
Ghaffari MH, Monneret A, Hammon HM, et al. Deep convolutional neural networks for the detection of diarrhea and respiratory disease in preweaning dairy calves using data from automated milk feeders. J Dairy Sci. 2022;105(12):9882–9895. doi:10.3168/jds.2021-21547
- 32.
Cantor MC, Casella E, Silvestri S, Renaud DL, Costa JHC. Using machine learning and behavioral patterns observed by automated feeders and accelerometers for the early indication of clinical bovine respiratory disease status in preweaned dairy calves. Front Anim Sci. 2022;3:852359. doi:10.3389/fanim.2022.852359
- 33.
Casella E, Cantor MC, Silvestri S, Renaud DL, Costa JHC. Cost-aware inference of bovine respiratory disease in calves using precision livestock technology. In: 2022 18th International Conference on Distributed Computing in Sensor Systems (DCOSS). IEEE; 2022:109–116. doi:10.1109/DCOSS54816.2022.00031
- 34.↑
Casella E, Cantor MC, Setser MMW, Silvestri S, Costa JHC. A machine learning and optimization framework for the early diagnosis of bovine respiratory disease. IEEE Access. 2023;11:71164–71179. doi:10.1109/ACCESS.2023.3291348
- 35.↑
Cantor MC, Costa JHC. Daily behavioral measures recorded by precision technology devices may indicate bovine respiratory disease status in preweaned dairy calves. J Dairy Sci. 2022;105(7):6070–6082. doi:10.3168/jds.2021-20798
- 36.↑
Bowen JM, Haskell MJ, Miller GA, Mason CS, Bell DJ, Duthie CA. Early prediction of respiratory disease in preweaning dairy calves using feeding and activity behaviors. J Dairy Sci. 2021;104(11):12009–12018. doi:10.3168/jds.2021-20373
- 37.↑
Sasaki Y, Iki Y, Anan T, Hayashi J, Uematsu M. Assessment of ventral tail base surface temperature for the early detection of Japanese black calves with fever. Animals. 2023;13(3):469. doi:10.3390/ani13030469
