4th Industrial Revolution, Artificial Intelligence and Companion Animal Medicine
According to Klaus Schwab, founder of the World Economic Forum, “We are on the brink of a technological revolution that will radically change the way we live, work and relate to each other. We do not yet know exactly how it will unfold, but one thing is clear: the response to it must be comprehensive and unified, involving all global stakeholders, from the public and private sectors to academia and civil society”.
The 1st Industrial Revolution used water and steam power to mechanize production. The 2nd used electricity to create mass production and the 3rd used electronics and information technology to automate it. The 4th Industrial Revolution refers to the current phase of industrial development characterized by the integration of advanced technologies into the digital, physical and biological world. It involves the fusion of technologies such as Artificial Intelligence (AI), Robotics, Internet of Things, 3D Printing, Nanotechnology and others. The 4th Industrial Revolution is characterized by the speed, scope and impact of the changes it is bringing about in all sectors of society, radically altering the way we live, work and interact.
AI, the branch of computing that deals with the creation of systems that can perform tasks that usually require human intelligence and a major component of the 4th Industrial Revolution, is not new. The term “artificial intelligence” was first used in 1955 by John McCarthy. In the last decade, AI has been growing at a geometric rate due to the increase in computing power, digitalization and the availability of large amounts of data.
Our everyday life is heavily influenced by AI. The movies and TV programs that are suggested to us and that we watch, the music we listen to, the text we type on our computer or mobile phone, the translation software, the management of our email, the photos we take, and numerous other activities are influenced to a small or large extent by AI. Our smart watch daily informs us about the quality of our sleep and suggests a personalized fitness program according to our physical condition and the goals we have set, combining information from the human body with the digital world.
A few definitions…
Medicine and Veterinary Medicine could not be absent as fields of development and use of AI. To understand how AI works and how it is applied in veterinary practice, we need to know some general concepts and terms.
There are different types of AI. AI systems that are designed for a very specific function (walking, speaking, translating, answering a specific clinical question) are “narrow” or “weak” AI systems. AI that has a human-like intelligence is known as “general” or “strong” AI, while that with greater than human intelligence is known as “superintelligence”.
Despite the increasing complexity and capability of computers, there are no computer systems that approach either general or superintelligence and, according to some researchers, may never exist.
“Machine Learning” (ML) is the subfield of AI where algorithms are trained to perform tasks by learning patterns from data rather than through programming. ML is developed and improved using a system that includes training, testing and validation processes. According to the types of learning it can be supervised, unsupervised and semi-supervised. In supervised learning, labeled data sets are used to train algorithms to sort data or predict numbers. The most common and the most frequently used ML method in diagnostic imaging is supervised, which requires the results of medical data to be known (labelled) before training the ML model. In unsupervised learning, the ML algorithm creates its own set of criteria by which it classifies unlabeled data or predicts outcomes, helping to understand them and considering whether there are potential clinical correlations. Semi-supervised learning uses a combinational approach and can be valuable for developing algorithms when some of the data are missing from the outcome. These ML methods are considered as “classical ML”. Modern ML includes “artificial neural networks” (ANNs) and “deep learning” (DL). ANNs, as their name suggests, are artificial computer network systems that stimulate the concept of human neurons and, like them, have many inputs and outputs and are connected to other nodes, with many inputs and outputs. The input data for the AI can be for example the medical image. The image is processed and filtered through a series of “layers” that help predict the outcome. This is where “deep learning” comes in, based on multiple layers of processing (usually more than 10) of artificial neural networks, and allows the algorithm to handle and process extremely complex data, such as medical imaging for example.
AI in companion animal medicine
The potential applications of AI in companion animal medicine are numerous and relate to almost every aspect of veterinary science. As long as digital data are available and manageable, AI technologies can be exploited. For example, in veterinary diagnostic imaging applications of AI focus on the detection of pathological findings, their description and their categorization. Radiomics is concerned with extracting large numbers of quantitative features from medical images using data characterization algorithms. The collected data are evaluated and used to create AI models to improve clinical decisions. Radiomics is applied to most imaging modalities. It can also be used in combination with clinical, biochemical and genetic data for greater diagnostic accuracy, assessment of prognosis and prediction of response to treatment. Similar algorithms can be applied by intensive care specialists in intensive care units, reducing the required diagnostic time. Biochemical and hematological analyzers include AI systems and, using specialized algorithms, provide clinical interpretation of results. AI applications analyze the movement and behavior of animals, facilitating their clinical and orthopedic examination. The use of AI applications may even be useful in the management of the social problem of stray animals.
In 2023 Bouchemla et al published the results of a systemic review of all AI-related international publications and scientific papers. Up to March 22nd, 2023, 812 papers were published/announced. In total, 192 studies were related to diagnostic imaging, 93 to veterinary education, 91 to animal production, 86 to epidemiology, 63 to health and welfare, 55 to internal medicine and 33 to microbiology. Fewer papers concerned the use of AI in toxicology, pharmacology, oncology, hematology, anatomy, nutrition, anesthesiology, statistics, environment and ecology, biochemistry, histology and embryology. It is obvious that the use of AI is relevant to almost the entire spectrum of veterinary science.
Reflections for today and for the future…
Concerns have already been raised about issues related to the adoption of AI technologies in veterinary medicine, which include jurisdiction, transparency, regulation, fairness and bias, privacy, ownership, liability and oversight.
There is no approval process for medical devices using AI for veterinary use. AI technologies for veterinary use should be approved by a regulatory authority prior to their use and then periodically monitored to ensure that they are up to date and remain valid. The characteristics of the training and test data, the type of algorithm used, and the performance data of the system should be known. If veterinarians do not know the characteristics of the data used to generate an AI product, they will not be able to evaluate its accuracy, its range of clinical applicability and its limitations, nor judge the potential benefits or potential risks of its use. Algorithms also require a management plan, which describes the oversight and quality assurance procedures necessary to ensure their validity. Some ML systems are evolving and as they continue to learn, it is likely that their results will change. Obviously, these systems cannot be used in everyday practice for commercial purposes, but they are ideal for research and development. Commercial systems need to be “locked-in” to provide repeatable results.
In 2023 James Bellamy raised certain questions which need immediate answers, and which are summarized below: Knowing that data in the AI era has enormous value, who is the owner of the data? How is the confidentiality and security of patient information ensured? Veterinary medical records are confidential and should not be disclosed unless consent is obtained from the patient’s guardian. How and by whom will the guardian be informed of the potential use of their animal’s medical data to develop an algorithm in a machine learning AI system? Will large datasets of medical records be able to be placed in the public domain so that everyone can benefit from their use? Should the results of an AI system be accepted in veterinary medicine without veterinary supervision? Who is responsible when an AI technology fails? If a veterinarian uses an AI product that produces and incorrect result, leading to death of an animal, how will liability be assessed/allocated? If the AI product fails, is liability different depending on whether the data set or algorithm is flawed? Should patient owner consent be required before implementing an AI system in veterinary practice? If a veterinarian ignores the correct result of an AI product and makes a different decision that is incorrect, is the veterinarian liable for ignoring the AI result?
All privacy and ownership issues need to be addressed immediately and the respective responsibilities and obligations of the manufacturersusers- costumers of AI applications need to be clarified before the use of AI systems in veterinary medicine becomes routine.
An important first step has recently been taken. On March 13th, 2024, the European Parliament adopted the first comprehensive legislative framework for the regulations of Artificial Intelligence (AI), which will enter into force by July 2024 and be implemented in summer 2026. However, it is necessary to continue efforts to develop a regulatory structure related to the use of AI systems in veterinary medicine and a regulatory framework for the use of veterinary medical data. The veterinary community may need to modify its ethical guidelines to address the many challenges surrounding the use of AI systems and adopt them in everyday practice.
A concerted effort involving all relevant stakeholders should perhaps be launched immediately…
George Mantziaras
DVM, PhD,
ECAR resident, Private practitioner
Athens, Greece
References
- Appleby and Basran (2022) Artificial intelligence in veterinary medicine. J Am Vet Med Assoc 30, 260(8): 819-824.
- Bellamy JEC (2023) Artificial intelligence in veterinary medicine requires regulation. Can Vet J 64(10), 968-970.
- Bouchemla F, Akchurin SV, Akchurina IV, Dyulger GP, Latynina ES, Grecheneva AV (2023) Artificial intelligence feasibility in vet- erinary medicine: A systematic review, Veterinary World, 16(10): 2143–2149.
- Candelon F, Charme di Carlo R, De Bondt M, Evgeniou T (2021) AI regulation is coming: How to prepare for the inevitable. Harvard Business Review, 99.
- Currie G. A muggles guide to deep learning wizardry. Radiography (Lond) 2022, 28(1): 240–248.
- Fjelland R (2020) Why general artificial intelligence will not be realized. Humanit Soc Sci Commun 7:10.
- Gillies RJ, Kinahan PE, Hricak H (2016) Radiomics: images are more than pictures, they are data. Radiology Radiology 278: 2, 563-577.
- Jiang F, Jiang Y, Zhi H, et al (2017) Artificial intelligence in healthcare: past, present and future. Stroke Vasc Neurol 2(4): 230–243.
- Kaul V, Enslin S, Gross SA (2020) History of artificial intelligence in medicine. Gastrointest Endosc 92(4): 807–812.
- Rizzo S, Botta F, Raimondi S, Origgi D, Fanciullo C, Morganti AG, Bellomi M (2018) Radiomics: the facts and the challenges of image analysis. Eur Radiol Exp 2:36.
- Schwab K (2016) The Fourth Industrial Revolution. Geneva, Switzerland: World Economic Forum.
- Shen YT, Chen L, Yue WW, Xu HX (2021) Artificial intelligence in ultrasound. Eur J Radiol: 139:109717.
- Waljee AK, Higgins PDR (2010) Machine learning in medi¬cine: a primer for physicians. Am J Gastroenterol 105(6): 1224–1226.
- Wischmeyer T, Rademacher T (2020) Regulating Artificial Intelligence. Cham, Switzerland: Springer Nature.
- Yoon (2017) What we need to prepare for the fourth industrial revolution, Healthc. Inform. Res, 23 (2) 75–76.
