AI in veterinary medicine: for students to learn current state

AI in veterinary medicine

1. Using AI-Enabled Real-Time Media Monitoring to Assess Veterinary Issues : Insights from recent events Andrzej Jarynowski thanks to Vitaly Belik

2. Cooperation All (Vitaly Belik - FU Berlin) Genetics (Alisa Sergeeva - FU Berlin) Social (Stanisław Maksymowicz - UMW) Infodemic (Maja Romanowska - WHO) Internet Scraping (Alexander Semenov – UF) Nature Conservation (Magdalena Lenda – PAN) Sanitary Inspection (Ireneusz Skawina - PPIS) Veterinary Inspection (Jakub Kubacki - GIW) Polish Army (Łukasz Krzowski - WAT) VetApp (Tom Brownlie – Ingenum ) The views and interpretations expressed by me in this presentation are solely my personal thoughts. They do not represent the official position or opinion of the Veterinary or Sanitary Inspection.

3. Plan AI and LLMs in epidemiology/epizootiology (One Health) Vetapp Haff disease (May 2025) retrospectively Oder river (tools available since half of 2024) Iberian Blackout (April 2025) Central European Flood (2024) FMD outbreak in Germany and Slovakia (January – April 2025 ) Borna Virus in Germany ( june 2025)

4. AI’s (LLMs) in One Health: AI has revolutionized disease diagnosis, drug discovery, and personalized medicine. Accelerated Threat Identification: LLMs could quickly detect and identify new issues ( such as pathogens ) , aiding in rapid response.

5. Create your own model to support classroom instruction Create sets of questions based on given material (RAG) Students can easily access products such as maps and statistical test results without any knowledge of math or programming. Applying GPT in academic research and organization of work There is no triage or symptom checker for animals (only some decision support tool for farms and clinics) Support for group work, simulation exercises Belief that AI will revolutionize veterinary medicine soon after human medicine Young breeders and vets are digital natives and MUST use it in the future

6. Loss of intuition in statistics Weakness of veterinary science compared to human medicine Create bias in thinking (i.e., models do not "feel" concepts such as geographic distribution of prevalence) Errors and mistakes (i.e. hallucinations) Induced changes in assessment types, scoring, or steps in PhD program (literature review) Reference statements for some journals that pay to be referenced by tools powered by ChatGPT Not for DVM, MD, but for supporting stuff (i.e. nurses, zootechnitians ) Cost of license for university? Some queries (i.e. pain, aggression or bioterrorism) may be blocked

7. Advances in computing power and modelling techniques, together with the amount of data obtained from disease surveillance, registries digital traces enable machine learning, complex system analytics standard statistical tools as well as computer simulation to be applied to the field of Veterinary Epidemiology.

8. Introduction to ChatGPT What is ChatGPT? An AI developed by OpenAI, capable of understanding and generating human-like text. Large Language Model/ Generative AI Capabilities: Natural language processing, information retrieval, language understanding.

9. ChatGPT in the Veterinary Field Potential Uses: Assisting with diagnoses, client education, research updates. Example Scenarios: Answering common pet health queries, providing care tips.

10. Accessing and Interacting with ChatGPT Access Methods: Via web platforms, mobile apps, or API integration. Interaction: Typing in queries and analyzing ChatGPT's text responses.

11. Assisting with Diagnosis Preliminary Diagnosis: ChatGPT can suggest possible conditions based on symptoms. Limitations: Must be used alongside professional veterinary judgment.

12. SYMPTOMATE?

13. Objectives Is there any value in veterinary observations? What mechanisms exist to unlock any value? What are the applications? Potential bias and other hidden limitations. In every epidemiologist's toolbox?

14. Commercial pretrained disease detection neural network WorkMate app Problem solving LLM for animal health professionals and while delivering alerts to the frontline Dedicated neural network, pretrained to detect biological traits in animal health and agricultural data Foresight API Delivers INSIGHTS and ALERTS D ATA and TRAINING Automatic invoice creation AI-enhanced clinical notes PMS and lab integrations Early disease detection (Foresight) Data secure Practice-level insights Early disease detection Endemic disease reporting Disease Forecasting Live industry health scores

15. Video/Stills of WorkMate

16. GPTs are designed for this purpose The dog chewed the bone because it was hungry The dog chewed the bone because it was delicious Self attention Vets do not record observations in tidy numerical codes Llama 2 7b Mechanisms to find value

17. Retrieval Augmented Generation (RAG) augments prompts and content generation with individual and external data Self attention RAG GPT’s provide non-specific (beige) responses

18. Self attention RAG Ontology https://bioportal.bioontology.org/ontologies/AHSO/?p=summary https://obofoundry.org/ontology/hso.html https://www.atol-ontology.com/en/a-ahol/ https://venomcoding.org/venom-codes/ List of labelled terms with classifications and relationships Veterinary observations are data sparse

19. Cow Lame Graphed data structures powerful representations Bursitis Carpus Horse Vertex (node) Clinical sign diagnosis. treatment Edge (link) Relationship between each node Global (master) Knowledge graphs Self attention RAG Ontology Graphs The value in veterinary observations is multidimensional

20. Prompt to LLM prompt = f""" I want to add to a networkx graph that I have created. The graph variable is G. I will provide a list of observations from vets. For each observation, I want you to assign the observation to a category by connecting the observation node to the category node. The categories are { categories } with the type "{ cat_type } ". These categories { exclusive } mutually exclusive. If any category is missing, please add it in the form of a node, using the ` G.add_node ()` with the `type="{ cat_type }"` argument. Return python code to add networkx edges between the observations and the categories. Return only the python code, no commentary at all. Do not provide example data or reiterate my data. Assume that all nodes exist. Just add the edges, and any new category nodes required. The observations follow, separated by the £ symbol: {observations} """ Anthropic Claude 3 - 5 Sonnet

21. Existing veterinary observations 180,534 clinical visits recorded with LLM augmentation 5 Categories: [Animal; Clinical Sign; Diagnostics; Treatment; Herd management] 2 further layers of subcategories (children) Veterinary observations

22. Self attention RAG Ontology Graphs RLHI Reinforced learning through human input (RLHI) Feedback from users fine tunes the model using a reward-based modelling system Vets don’t yet see the full value of their observations

23. Hybrid graph structure Hierarchical bipartite graph structure Turns static, tree-based ontologies into dynamic graph structures Piecemeal approach to progressively improve the graph Combines the bounded context approach with the computable ontology for the LLM Designed to accommodate growing datasets and evolving ontologies as veterinary services evolve. Horse Musculoskeletal Cow Subgraph for prompt: [‘carpal bursitis’] yields 1007 associated nodes

24. https://chatgpt.com/ Login or create account

25. 1) Cat Patient Information: Name: Whiskers Breed: Domestic Shorthair Age: 5 years Sex: Male Weight: 8 lbs Presenting Complaint: Whiskers was brought to the veterinary clinic by his owner with a history of persistent weight loss, lethargy, and recurrent respiratory infections. The owner also reported that Whiskers had recently been involved in more fights with neighborhood cats. Clinical History: Whiskers had a routine wellness exam six months prior, during which he tested negative for Feline Leukemia Virus ( FeLV ) and Feline Immunodeficiency Virus (FIV). His vaccinations were up-to-date, and he was an indoor-outdoor cat. The owner reported that Whiskers was generally healthy until the last few months. Clinical Examination: Upon examination, Whiskers appeared thin with a body condition score of 3/9. He had pale mucous membranes and enlarged submandibular lymph nodes. His coat was unkempt, and he exhibited signs of dental disease. The veterinarian noted a mild respiratory wheeze during auscultation. Diagnostic Tests: Complete Blood Count (CBC): Severe leukopenia (low white blood cell count) Anemia (low red blood cell count) Biochemical Profile: Elevated liver enzymes Hyperglobulinemia (increased globulin levels)

26. 1) Cat Symptoms and Signs: - Recurrent minor illnesses - Poor coat condition - Persistent fever - Loss of appetite - Inflammation of gums and mouth - Weight loss - Slow, progressive deterioration of health

27. 2) laying hens farm Farm Information: Location: Smith Poultry Farm, Anytown, USA Type: Commercial egg-laying facility Number of Hens: 50,000 Biosecurity Measures: Strict biosecurity protocols in place, including controlled access, sanitation, and monitoring. Presenting Complaint: The farm manager at Smith Poultry Farm reported a sudden increase in mortality among the laying hens. Many birds were observed with respiratory distress, drop in egg production, and neurological signs. A veterinary consultation was urgently requested. Clinical History: The farm had no previous history of major disease outbreaks. Routine vaccinations against common poultry diseases were up-to-date. The farm sourced feed from reputable suppliers, and water sanitation protocols were meticulously followed. However, wild birds were occasionally seen in the vicinity. Clinical Examination: Upon arrival, the veterinarian noted a significant increase in mortality, with dead birds exhibiting cyanosis (blue discoloration) of combs and wattles. Live birds displayed signs of depression, coughing, and swollen sinuses. Egg production had dropped sharply.

28. 2) laying hens farm Symptoms and Signs: - Sudden death without any signs - Loss of energy and appetite - Decreased egg production - Soft-shelled or misshapen eggs - Swelling of the head, eyelids, comb, wattles, and hocks - Purple discoloration of the wattles, combs, and legs - Nasal discharge - Coughing, sneezing

29. 3) Pig farm Location: Happy Pig Swine Farm, Rural County, Brandenburg Type: Finishing pig facility Number of Pigs: 2,000 Biosecurity Measures: Moderate biosecurity measures in place, including limited access, truck disinfection, and employee training. Presenting Complaint: Happy Pig Swine Farm reported an increased number of sick pigs characterized by high fever, lethargy, and sudden deaths. The farm owner noticed a decline in feed consumption and weight gain. A veterinary investigation was requested to determine the cause of the illness. Clinical History: The finishing pig farm had a clean health record, with no recent introductions of pigs. The farm sourced feed from a reputable supplier, and the transportation of pigs followed biosecurity protocols. The farm had not experienced any major disease outbreaks in the past. Clinical Examination: Upon arrival, the veterinarian observed pigs with high fever, reluctance to move, and in some cases, hemorrhagic skin lesions. The mortality rate had significantly increased, and the surviving pigs showed signs of depression and anorexia.

30. 3) Pig farm – a problem Symptoms and Signs: High Fever (40.5°C to 42°C) Loss of Appetite Lethargy and Weakness Respiratory Signs (coughing, difficulty breathing) Skin Hemorrhages and Cyanosis Vomiting and Diarrhea (often with blood) Abortions in Pregnant Sows Swollen Lymph Nodes Neurological Signs (convulsions, tremors) Sudden Deaths

31. https://chatgpt.com/share/68093f6b-9000-8008-97c2-a45903e32d9a

32. Client Education and Communication Education Material: Generating easy-to-understand guides for pet owners. Communication: Offering automated responses for common queries.

33. Continued Learning and Research Staying Updated: Using ChatGPT to find the latest studies and veterinary news. Professional Development: Assisting with learning new information and concepts.

34. Testing by image recognition

35. Integrating ChatGPT in Veterinary Practice Management Administrative Tasks: Automating scheduling, reminders, and record-keeping. Data Management: Streamlining the organization of practice information.

36. Ethical Considerations and Limitations Ethics: Responsible use of AI in veterinary settings. Limitations: Understanding AI biases and the importance of human oversight.

37. Conclusion and Future Prospects Benefits: Summarizing how ChatGPT can enhance veterinary practices. Future: Discussing potential future advancements in AI for veterinary medicine.

38. Epidemiology Epidemiology Tutor: https://chat.openai.com/g/g-t6JtO1Tmc-epidemiology-tutor

39. Epidemiology 2 BioStats Tutor: https://chat.openai.com/g/g-l6FAPg6p0-biostats-tutor

40. Mathematics https://chat.openai.com/g/g-0S5FXLyFN-wolfram

41. Education and academic tools https://chatgpt.com/g/g-bo0FiWLY7-consensus

42. Scenario 1 Mammals A/H5N1 laboratory positive values in USA 202 4 https://www.aphis.usda.gov/livestock-poultry-disease/avian/avian-influenza/hpai-detections/mammals  epidemiological curve  

43. Scenario 2 Mammals A/H5N1 laboratory positive values in USA 2024 https://www.aphis.usda.gov/livestock-poultry-disease/avian/avian-influenza/hpai-detections/mammals Maps and spatial information

44. INFOVEILLANCE Epidemiological surveillance deals with the analysis of web content to predict bio medical phenomena. Its most important advantage is the possibility of early warning (e.g. participatory reporting), or forecasting or improving estimators of incidence, prevalence or complications. Moving syndromic surveillance to the internet has great relevance (estimating the scale of health problems, early warning of events).

45. INFODEMIOLOGY Infodemiology is concerned with the study of the demand (e.g. search engine queries) and supply (social media content creation or commenting) trajectory of information, which was strongly articulated during the COVID- 19 pandemic. An infodemic is an overwhelming and uncontrolled information surge where reliable and unreliable sources coexist. Accurate and credible information, misinformation, disinformation, rumors, current data, and outdated content flood communication channels, overwhelming the recipient. Monitoring actual (real- time) and declarative attitudes should, in the WHO's view, be a priority for local decision- makers.

46. SOCIAL LISTENING What is Social Listening? The process of longitudinal surveying population and monitoring online conversations at social media platforms to gain insights into public sentiment, trends, and emerging issues. In ONE and public health, social listening involves tracking discussions related to animal health, diseases, symptoms, and outbreaks. Why is Social Listening Important in ONE healh Epidemiology? Early Detection: Identify potential outbreaks or disease clusters before they are officially reported. Real- time Surveillance: Monitor public sentiment and concerns regarding health issues. Risk Communication: Tailor public health messages based on real-time insights from social media. Outbreak Response: Evaluate the effectiveness of interventions and identify misinformation or rumors.

47. Standard Surveillance Disease and other Registries, Covariates, Surveys Sensors Bio-surveillance Contact patterns Models Statistical, ML, SD or ABM Infoveillence Internet data collection INFECTIONS DISEASES Zoonoses and Only animal hosts Digital veterinary epidemiology

48. SURVEILLANCE DATA

49. Standard NLP vs Prompt engineering keywords search vs LLM ( topic ) Lemmatization Words variants languages Good Prompts (i.e. i dentify texts that are about mercury (the substance).

50. Methods (Data & Analytics) Brand24 data collection; keyword-based capture across languages NLP & LLMs for topic and sentiment classification Metrics: mentions vs. reach; negative/positive sentiment shares

51. Gemini, Llama etc : open source offline model (no Low resource languages till now)

52. FMD outbreak in Germany and Slovakia (January – April 2025 )

53. FMD Food-and mouth disease in Europe 2025

54. Germany phase Main Topics  Main emotions Main actors 10.01 Declaration Just reporting Neutral reporting tone with note of Sadness  Various Traditional media 11-12.01 Risk assessment Local restriction Mixed Small local social media 13.01 Measures implementation International restriction Neutral with note of Disgust Professional Traditional media 14-15.01 Public awareness Meaning for human health and economy Negative with fear High reach Social media  16-17.01 Fake news Fake news Very negative, anger Professionals 18-19.01 Restriction release Release of measures Neutral with highest fraction of positive Various Traditional and social media After 20.01 Post factum discussion Preparedness , pork price Neutral with neg emotions Professionals ( mainly farming )

55. Germany Fake news propagation “#MORE A second outbreak of foot-and-mouth disease (FMD) has been reported in eastern Germany, days after an outbreak led to widespread culling and trade restrictions.  https://nordot.app/1252721126551928958 ”

56. FMD as a Lab-Made or Released Virus Suspicion that the virus may have come from a laboratory (US and foreign labs), possibly deliberately released. "Nem egy laborból szabadult ki? (szabadították ránk?)"   👥 Political Cover-Up and Misinformation Authorities are accused of hiding the truth or minimizing the outbreak’s scale. "nem minden úgy van, ahogy lennie kellene"   💰  Hidden Agenda & Meat Market Distortion Allegations that the outbreak is used to control meat supply and import milk and beef for Mercosur or Ukraine. Animal cull includes healthy livestock and serves unknown goals. "a 'vraj' slintačka és krívačka és a megmagyarázhatatlan egészséges állatok leölése" Slovakia , Hungary Conspiracy Theories

57. retrospectively Oder river (tools available since half of 2024)

58. FMD: Conspiracy Mentions in Traditional Media (by Language) Language Conspiracy Rate (%) Total Mentions slk 2.365 3890 pol 0.941 4675 deu 1.426 5259 hun 4.207 5467 eng 4.322 8468

59. A fish kill of unknown origin has been spreading along the Oder and its tributaries since the last days of July 2022. Oder river disaster 2022

60. Central European Flood (2024)

61. Iberian Blackout (April 2025)

62. Apagao , Apagon

63. Poultry and Pig Farms Ventilation, feeding, and incubators were affected: Minor fatalities in poultry Some meat had to be destroyed due to loss of refrigeration Hatcheries generally had backup generators.

64. Dairy Farms The blackout disrupted milking processes: - Cows were milked manually - Up to 24 million liters of milk spoiled due to lack of cooling - Increased risk of mastitis and animal stress.

65. Companion Animals Pets experienced stress mainly due to: - Inactive elevators - Disrupted feeding routines Animal welfare groups advised on safe practices during blackouts.

66. Government Response Fuel for generators was NOT distributed to key farms and vet centers. Communication delays hindered planning. Veterinary organizations issued animal care guidelines during outages.

67. Haff disease (May 2025)

68. Haff disease (Mai 2025)

69. LMS (LARGE LANGUAGE MODELS) USED Disruption of settlement and academic structures in 1945 (Albertus-Universität Königsberg -> Immanuel Kant University; Medizinische Akademie Danzig-> Medical University of Gdańsk ) Digital Library of Annales Academiae Medicae Gedanensis , and general search in Polish scientific and medical literature: No recorded cases in Polish literature Digital Libraries of Russian Continuation of documentation in the Kaliningrad Oblast: In Russian sources, the disease is therefore often called “ Юксовская болезнь ” - “ Juksovo disease”

70. News about HMPV in China FMD In Gemany FMD In Hungary / Slovakia USA stopped publishing AI V reports USA has canceled AI and Mpox vaccine projects

71. Applications Veterinary observations LLM Publications and veterinary text Borne Virus

72. Play with known data, pictures , info and chat gpt Teamwork in outbreak investigation involves a multidisciplinary group working collaboratively towards a common goal : to identify the source , mode of transmission , and ways to control and prevent the spread of the disease All groups – veterinary public health officers in Bavaria

73. Scenario group 1 Animal host ( Reservoir and Transmission )

74. Create infograhic for forestry workers Scenario group 2

75. Scenario group 3 a medical overview of the disease (symptoms, progression, treatment options) in human

76. The Path Forward : Teaching future DVMs and MDs about LLMs and AI — how to use them in early warning and triage systems Our Collective Responsibility: We must take proactive measures (or at least be aware of) to ensure LLM is used ethically (i.e. dis-/mis-information) and responsibly, safeguarding global health security

77. Using AI‑Enabled Real-Time Media Monitoring to Assess Animal Welfare During Crises: Comparative Insights from the Iberian Blackout (April 2025) and Central European Flood (2024) — Jarynowski , A. — ASzwoj FMD in Germany — Jarynowski , A. — IBI
FMD in Slovakia (Panic) — Jarynowski , A. — IBI
The 2024 Flood in Poland from the Perspective of Social Aspects of One Health — Jarynowski , A. — ( IBI )
Lameness in Cattle – New Opportunities for Detection and Assessment Using Smart Technologies in Health Management — Błaszkiewicz , O., et al. — Życie Weterynaryjne Antibiotic Overuse and Antimicrobial Resistance: A Growing Problem in Poland and Worldwide — Jarynowski , A. — Nowa Konfederacja Avian influenza: The looming threat of disease X and lessons from Poland and Europe — Jarynowski , A., et al. — European Journal of Translational and Clinical Medicine
One Health multimodal surveillance in time of change — Яриновський , А., et al. — One Health Journal
Excess mortality during the COVID-19 pandemic in low-and lower-middle-income countries — Gmanyami , J. M., et al. — BMC Public Health
Animal health discourse during ecological crises in the media — Meletis , E., et al. — Veterinary Sciences

78. Assessing large language models in the context of bioterrorism — Jarynowski , A., et al. — War Science Academy
Opportunities for Using AI in Monitoring the Health of Cattle and Pigs — Jarynowski , A. — (n/a)
AI/ML in poultry, cattle and pig farming — Jarynowski , A. — Interdisciplinary Research
Can Avian Influenza Trigger the Next Human Pandemic? — Jarynowski , A. — Nowa Konfederacja Information flow during culmination of online public discourse based on the Oder disaster — Grelowska , M., et al. — E-methodology
The Tragedy of the Oder. How Poles and Germans Talk Past Each Other — Jarynowski , A., & Maksymowicz , S. — Forum Dialog
Farmers' Protests and Animal Husbandry — Jarynowski , A. — Nowa Konfederacja Digital traces of protests — Jarynowski , A., et al. — Motra -K
Multiplex network approach for modeling African swine fever — Jarynowski , A., et al. — Computational Data and Social Networks
The curious case of the lion from Berlin — Oelke , J., Jarynowski , A., & Belik , V. — E-methodology
Agroterrorism involving biological agents — Jarynowski , A. — Terroryzm

79. Triangulated phylodynamic- spatio -temporal analysis — Jarynowski , A., & Belik , V. — Zenodo Biological mis(dis)-information as Kremlin warfare — Jarynowski , A., et al. — Zenodo Analysis of perception of infectious diseases online — Jarynowski , A., et al. — SVEPM
The Impact of Online Information on Health Decisions — Jarynowski , A., & Wójta-Kempa , M. — E-met Grain and food security as a tool of biopolitics — Jarynowski , A., et al. — E-methodology
African swine fever – Epizootiology, economics and crisis management — Jarynowski , A., et al. — Studia Administracji i Bezpieczeństwa African swine fever – Potential biological warfare threat — Jarynowski , A., et al. — EasyChair Preprints
Medical data processing and analysis for remote health — Vitabile , S., et al. — High-Performance Modelling and Simulation
Cost analysis of the spread of African Swine Fever in Poland — Jarynowski , A., & Belik , V. — Public Health Forum
African swine fever awareness in the Internet media — Jarynowski , A., et al. — E-methodology