IVAN, a diagnostic aid system for bovine autopsy: the (not so) artificial intelligence that makes the dead talk

Context :

The veterinary autopsy is a major procedure, which makes it possible to detect an emerging or re-emerging disease, to monitor the overall health of the herd by analysing one of the animals in it and to implement corrective measures (zootechnical or therapeutic) if necessary. Veterinary autopsy requires a high level of expertise and skills that not all veterinarians necessarily master, especially in the context of rural desertification. The IVAN ("Innovative Veterinary Assisted Necropsy") diagnostic aid system aims to guide veterinarians in necropsy diagnosis and in taking appropriate samples to confirm or refute their hypotheses. IVAN relies on AI methods (Bayesian network in particular) to deduce relevant proposals at each stage of the diagnostic process. The process is totally transparent and the veterinarian is in control of the system through validation at each stage of the reasoning.

Results :

IVAN is an intuitive, ergonomic and powerful application. It provides relevant and rapid results, avoiding the "black box" effect. Each step is validated by the veterinarian, who is able to follow and influence the thinking process. Initial tests have confirmed the reliability and relevance of IVAN's proposals. The implementation of a self-learning system, by extending the data in the knowledge base on which the system is based, will enable IVAN to gain in performance as it is used.

Perspectives :

An experimental evaluation of IVAN's performance on a larger scale is planned in the near future. This phase will make it possible to determine precisely the system's performance (reliability of proposals) and to measure IVAN's assistance capacity by evaluating the impact of proposals on the veterinarian's choices. In addition, two development perspectives are envisaged:

1) extend the scope of application to other animal species (pigs, poultry, rabbits, etc.)

2) to adapt the IVAN processes (hypothetico-deductive inferences, AI, etc.) to live cattle, taking into account not the lesions but the clinical signs. The adaptation of NAVI to other animal species and the evolution to live animal diagnosis opens up innovative perspectives in terms of the application of AI in veterinary medicine.

Valorisation :

Software declaration filed (Software declaration DV4255 of 12/08/2020, with a view to filing with the APP). The conditions for making IVAN available to veterinarians are currently being studied (professional software platform to be offered as an e-service to field veterinarians). IVAN was also the subject of a presentation at the animal e-health section of the e-health summer university (Castres, 2019 : https://www.youtube.com/watch?v=8m1TReeUzIU)