AI in veterinary medicine Enhancing Medical/Veterinary Teaching with AI Technology

AI in veterinary medicine Enhancing Medical/Veterinary Teaching with AI Technology

1. Andrzej Jarynowski , Łukasz Krzowski , Łukasz Czekaj, Daniel Płatek, Vitaly Belik African swine fever: mathematical models of the epizootic processes 1) System Modeling Group, Institute of Veterinary Epidemiology and Biostatistics, Freie Universität Berlin, Germany 2) Interdisciplinary Research Institute, Wroclaw, Poland 3) Aidmed , Gdańsk

2. Advances in computing power, 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.

3. ASFV (African Swine Fever) and HPAI (Highly Pathogenic Avian Influenza) ASF In Poland since 2014, In Lower Silesia since 2019, In Germany since 2020 POLAND Rank 3→6 in pigs (up to 90% farms closed due to restrictions in some areas), HPAI In Poland since late 2019 (current wave), In Lower Silesia since. 3 →1 in poultry in EU

4. Why modelling? COVID-19 Nature Conservation: endangered species Economy: crops or livestock disaeses Climate change scenarios ONE Public health

5. ASF “is probably the most serious animal disease the world has had for a long time, if not ever” 2018 Dirk Pfeifer, world class veterinary epidemiologist The pork shortage in China caused by African swine fever could contributed to the spillover of SARS-CoV-2 from animals to human “For the first time in history, global consumption of poultry meat has exceeded the pork (. . . ) the main influence of this change in consumption is the crisis caused by the African Swine Fever” 2019 FAO

6. ASF 5% losses to Pig Industry Revenue DIRECT COSTS 100 M EUR -Farmers’ compensations (for killed animals for discontinuing production for closing production) - Diagnostics (active and passive surveillance), - Carcasses utilization (pig, WB); - Payment to carcass searchers and hunters - Labour and material costs of administration as veterinary inspection, border control. INDIRECT COSTS 300 M EUR - Lost Foreign pork/pig markets and lost self-sufficiency; - Hog’s price volatility and difficulty in selling pigs for slaughterhouses ; - Farmers’ capacity building costs for biosecurity preparation; - Operational costs of biosecurity procedures running in farms; - Social crises management (protests, mistrust).

7. Figure adapted from: Chain of infection; the chain of events that lead to infection. Genieieiop Purposes: Understand the roles Predict / Test hypothesis and scenarios Improve control and ultimately eradicate infection An epidemiological model is a simplified representation of a real-life system of disease transmission. “all models are wrong, but some are useful” Box GE. Science and statistics. J American Stat Assoc. (1976) 71:791–9. doi: 10.1080/01621459.1976.10480949 INFECTIOUS DISEASE MODELING

8. Differential equations Agent based models Machine learning Short -term prediction Medium-term prediction Long -term prediction

9. DYNAMIC CAUSAL MODELLING OF COVID-19 https://covid19forecasthub.eu/visualisation.html

10. Differential equations Agent based models Machine learning Short -term prediction Medium-term prediction Long -term prediction

11. ML (machine learning) Agent base d Model

12. We run set of simulations for selected subspace of simulation parameters: a - swine amount significance, b - wild boars significance, c - pork production chain significance. d- human mobility significance Where: i , j – poviats ; P – normalized amount of pigs; F – coverage of forests; p ij -probability of infection from a neighbor ; g ij -probability of infection from a whole networks; d ij -angular distance between centroids of poviats (counties).

13. http://interdisciplinaryresearch.eu/index.php/asf/ http://interdisciplinaryresearch.eu/index.php/asf/ http://interdisciplinaryresearch.eu/index.php/asf/ Jump to Western Poland Arrival times estimated before jump

14. We are comparing standard directly generalized linear regression with machine learning in boosted tree classification and indirectly phenomenological SIR - type model. We found that predictability of epizootic state (defined binary) one period in advance on the border between disease free and in affected regions reaches over 95% sensitivity and specificity with XBoost , which outperformed other methods. Machine learning seems to be the best predictor in a neighborhood of an already affected area. WB pigs

15. „Ideal” barrier for WB: - fence on the Eastern Border - blocking corridors on A1 motorway Objectives: - Arrival time to swine district (fencing, A1) - Border vs Interior dynamics

16. A1:There is an important difference in the arrival time to swine district (most likely to be affected in 39 months – on average half a year barrier ) . Fence: There is only a small difference in the arrival time to swine hot spot district (most likely to be affected in 34.2 months – on average less than one month barrier ) Base line (33 arrival time) months since December 2018 vs Border Fencing or corridors blocking on A1

17. Individual Sensor features Aidmed Recorder Functionality: RECORDS: respiration, chest movements, animal activity level&position, snoring and cough sound level, skin temperature TRANSMITS: data in real-time to AIDMED mobile app COLLECTS: all data without cables Aidmed.ai Telemedical System: COLLECTS: data from AIDMED MOBILE APP PROVIDES: visualization of data; signal processing and AI based analytics of ECG, reports for the patient and physician; ability to manage devices, patients, user, access rights Aidmed Mobile App: COLLECTS: data from AIDMED Recorder from any BT enabled device TRANSMITS : data in real-time to AIDMED Telemedical system PROVIDES: an interactive patient guide on how to use AIDMED , questionnaires

18. Hens sensors Outdoor range RFID tag RFID Antenna

19. to assess usability of commonly used surveillance technology as video cameras, microphones, sensors, Radio Frequency Identification (RFID) to detect characteristic movement and contact patterns of hens using key resources (feeders, nest boxes) to explore the relationship between events with mortality or infections , which might allow for prediction (ML/AI) Objective

20. Materials and methods

21. Networks

22. Materials and methods Outdoor range RFID tag RFID Antenna

23. Hour 22 Whole day

24. behaviour of hens i.e. can be grouped into “stayers”, “roamers”, “rangers” social contact patterns, does hen has friends? real-time location tracking systems can automatically predict some of the disease Can we then predict death before it happens? Prediction Welch et al, 2022 Response Accuracy AUC Cestodes 0.60 ± 0.01 0.58 ± 0.01 Ascaridia galli 0.80 ± 0.01 0.60 ± 0.02 Spotty Liver Disease 0.82 ± 0.01 0.61 ± 0.02   Belik et al, 2021 Sibanda et al, 2020

25. How? S I R E   gamma   delta   susceptible recovered infectious exposed

26. Wildlife Workers wil d _ agr = 0.1 11 work er _ agr = 0.1 05