ARTIFICIAL INTELLIGENCE IN MEDICINE: PROS AND CONS

© 2021Marina Vladimirovna Bakanova

2021 – № 2 (22)

DOI: https://doi.org/10.33876/2224-9680/2021-2-22/05

Citation link:

Bakanova M. V. (2021). Iskusstvennyj intellekt v medicine: za i protiv [Artificial Intelligence in Medicine: Pros and Cons]. Medicinskaja antropologija i biojetika [Medical Anthropology and Bioethics], 1 (22).

Author info:

Marina Vladimirovna Bakanova is the head of the international medical center Dua Hospital.

Keywords: artificial intelligence, health care, medical anthropology, conventional and non-conventional medicine

Abstract. Since 2017, the presence of AI in medicine has been growing exponentially – at the same time, the COVID-19 pandemic has only accelerated the processes of its implementation in the health care system. To a certain extent, its presence has brought about many positive changes: simplification of medical documentation, acceleration of analysis of massive databases, improvement of patient access to medical services, etc. Despite these achievements, the strengthening of AI presence in medicine causes several problems. Both doctors and patients raise questions and express certain concerns. This article offers a review of the main positions in the ongoing discussion.

Artificial intelligence in health care is algorithms and programs designed to simulate human cognition in the analysis, presentation, and understanding of complex medical and health care data.

In the mid-1930s, Alan Turing’s concept envisaged the creation of machines capable of independently solving complex problems. The so-called Baby Machine concept, which involves teaching artificial intelligence in the manner of a small child, rather than creating an “intelligent adult” robot at once, was the prototype of what we now call machine learning (Taylor 2015: 20-21). At the University of Dartmouth in the United States, the first working conference was held in the summer of 1956 with the participation of McCarthy, Minsky, Shannon, Turing, and other scientists who were later named the founders of the field of artificial intelligence. It was at the conference that the very term “Artificial Intelligence” was formed. The beginning of research into AI in health care was laid in the 1960s, which resulted in the creation of the MYCIN, INTERNIST-1, and CASNET systems, none of them were ever widely used. The capabilities of computers of that time did not allow for quick calculations in isolation from a person, which froze AI programs for a long time (Markoff 2017: 221-235).

A renewed interest in AI and in particular in AI in health care has been observed since the mid-90s in connection with the creation of high-power computers and the Basic programming system. In particular, interest was greatly fueled in 1997, when the IBM Deep Blue supercomputer managed to defeat the reigning world chess champion, Garry Kasparov. A new leap in the development of AI took place in 2005-2008 when scientific mathematics developed new theories and learning models for multilayer neural networks, which laid the foundation for the theory of deep machine learning, and in the field of IT, the production of inexpensive high-performance computers was established (Koomey et al 2011: 46-54). As a result, after 10 years, it became possible to talk about a historical computer breakthrough, and AI came to various aspects of social life, including medicine.

At present, the term “Artificial Intelligence” denotes software systems as well as methods and algorithms used in them, with their main feature being the ability to solve intellectual problems in the same way a person thinking about a solution would do. Thus, modern AI to a certain extent copies the human nervous system. The main successful implementations of AI, including in medicine, are currently: deep neural networks and deep learning technologies (Pirzada 2014). The main neural networks that are currently in use: “classical” single-layer, multilayer neural (convolutional network as its most complex form), neural networks with feedback. Machine learning differs in approaches: teaching with a teacher (the quality of the final information is significantly influenced by: the subjectivity of a human expert, representativeness of the sample, the level of development of science and technology, population and geographical aspects, etc.), reinforcement learning, self-learning (AI focuses on the independent search for hidden dependencies, requires a huge amount of data, is barely capable of translating from a machine language into a human one, there is no “inverse function” – that is, the machine cannot explain to the expert why it made a decision) (Bahl et al 2018: 810-818 ). Information systems (digital platforms) have appeared that are designated as “IT + DT + AI + IOT”: “IT” – universal digitalization of processes and computerization of workplaces, “DT” – the accumulation of data and the use of powerful information processing technologies, “AI” – the creation of robotic AI algorithms that operate both in partnership with a person and independently, “IOT” – “Internet of Things” or a computing network consisting of physical objects (“things”) equipped with built-in technologies to interact with each other or with the external environment (Bhattad, Jain 2020: 1-5).

The AI ​​sphere is divided into three types: weak, medium, and strong. The tasks of “weak AI” include – solving issues as a cybernetic apparatus working according to the laws and rules prescribed by a person. The “average AI” is actively developing, which has elements of adaptive self-learning, improving as data accumulates, issuing new classifications, analysis using new algorithms, etc. The sphere of “strong AI” is considered to be the simulation of the elements of human higher nervous activity, the experience of emotions – its development is only projected for 2030-2050.

The explosive development of AI since 2017, fueled by the COVID-19 pandemic, has led to the active adoption of AI systems in health care. The introduction of artificial intelligence systems in medicine is one of the most important trends in modern health care, including in Russia. Experts identify both optimistic and pessimistic forecasts in regards to the new phenomenon, focusing their attention on overcoming obvious disadvantages.

In April 2017, scientists from the University of Nottingham presented artificial intelligence technology that can predict the onset of myocardial infarction. The system combined four computer programs (with machine learning algorithms), data from 378 thousand patients, and 22 assessment criteria (including age, nationality, presence of arthritis and kidney disease, blood cholesterol level, etc.). The data were reconciled with the results for 2015, and the accuracy of the computer program was higher than the accuracy of cardiologists (working on the recommendations of the American College of Cardiology and the American Heart Association): from 74.5% to 76.4% of accuracy versus 72.8%; those using this technology could potentially save 355 more lives (Weng, Reps 2017).

Also in 2017, scientists from the Massachusetts Institute of Technology (MIT), with the participation of specialists from the Massachusetts Central Hospital, developed an AI system that can control a person’s sleep using radio waves. It remotely analyzes radio signals around a person and determines the stages of sleep – light, deep or fast – by eye movement. Potentially, according to her work, scientists plan to study his disorders, and then begin an experimental treatment that can be carried out remotely (MIT 2017)

In 2018, an Algorithm known as VarQuest (described in the journal Nature Microbiology) was developed. In the words of Hossein Mahimani, a professor at Carnegie Mellon University, “VarQuest has completed a search that would have taken hundreds of years using traditional computations.” The article deals with the search for new antibiotics to which microorganisms have no resistance. But in fact, further improvement of the program or the creation of its analogs may be capable of searching for new drugs and other groups, moreover, shortening the time of launching new drugs on the market (it currently takes several years, so that the drug even before the mass production start is already becoming obsolete) (Gurevich et al 2018: 319-327).

Also in 2018, a pilot project on AI diagnostics of fetal developmental disorders was launched in the UK. Ultrasound diagnostics featuring artificial intelligence was named ScanNav and initially used a database of 35 thousand reference images. In 2021, the ScanNav Anatomy PNB enhances real-time ultrasound support program for regional anesthesia was developed, which was officially introduced to medical institutions in the UK and is being tested in the United States (Intelligent ultrasound 2021).

From 2018 to 2021, several programs have been developed that allow for the successful detection of oncological diseases: Second Read (Ibex at the Institute of Pathological Anatomy Maccabi Health care Services Israel), IBM WatsonHealth for liver cancer diagnostics, skin cancer recognition programs (FotoFinger, Doctor Smart) and cancer mammary glands (Deep Mind Al, Celsus), Konica Minolta (Japan) has begun research on an oncogenic panel test, SRL Diagnostics, together with Microsoft, is testing cervical cancer recognition systems, Botkin.AI platform successfully recognizes signs of lung cancer, development of a program to search for cancer metastases has begun in lymph nodes (Sensei) and remote (virtual) biopsy programs (including automated histoanalysis). In general, working with oncology is one of the most effective in medical AI.

AI technologies for assessing the quality of embryos in assisted reproductive technologies were proposed at the end of 2018 by experts from Cornwall University in the USA and Imperial College London. Research has shown that when using AI, the quality of IVF increases by 20%.

In 2019, an AI facial recognition program was tested in China at the Shanghai Children’s Medical Center (SCMC) to detect genetic abnormalities in newborns.

At the end of 2019, an AI program for predicting epileptic seizures was presented. Scientists Hisham Daoud and Magdy Bayoumi at Louisiana State University developed it with an efficiency of 99.6%. However, now it is used to a limited extent, due to the difficulties with the manufacture of a special chip.

Shortly before the COVID-19 pandemic, the British health care system, under the active patronage of the minister, was engaged in the analysis of AI indications for the hospitalization of patients. Unfortunately, the future has shown that although the reduction of beds is cheaper for the budget, it negatively affects the course of the pandemic. At the same time, the project promises to be very effective in the absence of emergencies.

The developer of computer data processing technologies Xilinx and the Spline.ai Company, already in October 2020, presented a draft AI program for analyzing X-rays of the lungs, which greatly simplified the work of doctors in the red zones. The model was built using over 30,000 confirmed X-ray images of pneumonia and 500 X-rays images of COVID-19 complications that have been provided by government medical and research institutions and trained using the SageMaker AI service (AuntMinnie.com 2021). A little later, the services of several other companies using AI began to work similarly.

The Israeli company MedyMatch Technology has developed a program for recognizing the type of stroke based on AI and big data. To do this, in real-time, the MedyMatch system compares the image of the patient’s brain with hundreds of thousands of other images that are in its “cloud”. Since a stroke by its etiology can be hemorrhagic (with cerebral hemorrhage) or ischemic (due to a narrowing of a vessel or blockage of it with a thrombus), the etiotropic treatment should be different, moreover, it should be prescribed very quickly within 24 hours, otherwise, the changes may become irreversible. At the same time, the error of doctors in the analysis of CT is about 30%, which leads to disability. AI errors in this direction are minimal.

The mobile application of the British company Your.MD (the development of which was started back in 2015) uses a large database of symptoms and generates recommendations based on the patient’s responses. There are several modern developments in the form of bots, however, their general level of diagnostics remains very low and is mainly limited to general recommendations.

In 2020, according to ResearchAndMarkets, the market for medical AI solutions reached $ 4.2 billion. Most countries in the world have made a serious bet on the development of these technologies and services.

In Russia, the main development and testing of AI projects are on the level of the Ministry of Health Care, the health care system of large cities (Moscow, St. Petersburg, and some others), or individual initiative groups. Currently, the following projects are being developed: The Unified State Information System in Health Care (Unified State Health Information System) and a single digital circuit based on it, the priority project “Electronic Health care”, a unified digital system for diagnosing cancer, electronic medical records, and sick leaves, electronic prescription, Medical Information System (MIS), telemedicine – it is noticeable that AI is mainly used in the health care organization system, working with documents and databases, processing medical records and, in general, does not participate much in real clinical practice.

On November 17, 2021, Russian President Vladimir Putin announced the creation of the Center for Genetic Information and, in general, set the direction for conducting genetic research and creating bioresource collections. Such work with large databases will definitely require the participation of AI programs. On December 1, 2021, based on the Federal Center for Brain and Neurotechnologies, the Scientific and Production Complex of Personalized Medicine was opened – a unique platform that actively uses, among other things, AI technologies. Thus, in contrast to Western and Chinese medicine, Russian medicine in terms of the development of AI relies more on government regulation.

The most effective AI directions in medicine at the moment:

  1. Document management: Nuance systems, Camden Group, GE Health care, etc. Pros: Reducing the likelihood of human error and increasing the effectiveness of treatment. Cons: the complexity of servicing software and technological systems, their protection from failures and cyber attacks, as well as ensuring the confidentiality of patient data
  2. Using virtual assistants instead of nurses, which allows you to keep patients in touch with health workers and at the same time reduce the number of visits to hospitals – the Sensely system
  3. Conducting surgical operations with full or partial use of robots: Da Vinci complex, miniature HeartLander robot.
  4. There is great potential in projects for the care of the elderly – systems for automatic control of pressure, pulse, blood sugar, and other parameters with periodic consultations of specialists
  5. Research of the genome – not only of a person but also of agents threatening him, the creation of vaccines.
  6. Discovery of new drugs, reduction of terms of their introduction into medicine
  7. Projects of medical imaging and automated methods for diagnosing various diseases by video, audio, photo, and other materials.
  8. Systems for analyzing big data and predicting events – it is especially important to predict the next outbreaks of infections and epidemics
  9. Support systems for making medical decisions – for an additional guarantee of patient safety.

The main drivers of growth in the field of AI in medicine:

  • A growing volume of medical data that is difficult for a doctor to use in daily clinical practice;
  • Increasing the complexity of data sets;
  • The needs of health care organizations to reduce costs, including for equipment, and increase computing power;
  • Increase in the number of cross-sectoral partnership projects;
  • The growing imbalance between the number of doctors and patients, which pushes the development of demand for impromptu medical services (shortage of doctors in developing countries, projects of AI booths in China)
  • Imbalance of medicine between developed and backward countries
  • Implementation of such technologies by numerous pharmaceutical and biotechnology companies around the world. They, in particular, will use such developments to create vaccines and drugs for the treatment of COVID-19.

Negative factors:

  • The reluctance of practicing doctors to introduce artificial intelligence into their work,
  • A high possibility of discrediting AI and technologies not only in the eyes of doctors – representatives of the practical level of health care but also in the eyes of patients
  • Creation of a rigid system for analyzing the work of a doctor (with a possible transition to totalitarian control): good for health care, but not for humans
  • The shortage of specialists capable of working with such technologies – the complexity of retraining to new systems of work of older doctors, the complexity of retraining the professors of medical universities and changes in training programs, as well as the transition of students to a new work system
  • Ambiguous legislative regulation of the medical software market.
  • Lack of carefully selected medical data, the complexity of an informed request to patients for permission to use from personal data, the complexity of parsing data depending on race, gender, age, isolated ethnic groups, etc. Difficulties in interpreting similar data from machines of different power levels and different production (for example, ultrasound images from machines of different generations)
  • Concerns about the confidentiality of such information – hackers, interest in health information from insurance funds and employers
  • Problems of integration between AI solutions from different manufacturers, incompatibility of programs.
  • All AI systems proved to be as effective as possible where they were created. When the system was transferred even to the base of another hospital (not to mention the change of the country or the nationality of the patients), most of the programs failed, since all medical centers are qualitatively different in the people who visit them. Those now in clinical practice, AI systems are not universal.
  • Legal and ethical problems of using patient data, non-elaboration of legislation, and the possibility of creating depersonalized databases.
  • Automation and disappearance of a significant part of jobs

Like all new technologies, artificial intelligence can be misused and harm patients, according to WHO Director-General Tedros Adhanom Ghebreyesus. To regulate and monitor the use of AI in medicine, WHO has published new guidelines that outline six principles for limiting risks and maximizing the benefits of AI for health (UN news 2021)

The WHO report, The Ethical Principles and the Application of AI for Health Care, indicates that AI can be used to:

  • Acceleration and improvement of the accuracy of diagnosis and screening of diseases,
  • Assisting in difficult clinical situations,
  • Accelerating health research and drug development,
  • Supporting a variety of public health interventions, including outbreak response and health systems management,
  • Artificial intelligence allows patients to independently monitor their health
  • Enables countries with limited resources to remotely access health services.

However, WHO experts are simultaneously warning doctors and patients against overly enthusiastic assessments of the health benefits of AI. WHO points out the possibility of unethical collection and use of health data, the manifestation of various biases in human communities in AI algorithms, and recalls the risks related to patient safety, cybersecurity, and the environment. In addition, WHO notes that systems trained on data from high-income countries may be ineffective in low- and middle-income countries (UN news 2021)

However, traditional Chinese, Ayurvedic, and other types of medical care can use modern artificial intelligence programs. Opportunities of AI in unconventional medical practices and in bringing them closer to conventional medicine:

  1. Extended analysis of the constituent parts of these practices with the formation of large amounts of information, identification on their basis of patterns in treatment and effectiveness in treatment and prevention
  2. Comparison of the effectiveness of treatment with conventional and unconventional practices for certain diseases
  3. Compilation and analysis of traditional herbal collections, their effectiveness, and applicability; identification of patterns for the formation of new phytotherapeutic kits with desired properties
  4. Drawing up detailed maps and research of points of reflexology for various options of impact
  5. Carrying out studies and comparisons with the available data of little-known variants of unconventional practices: tribes of Africa, the Amazon, the Pacific Islands, or local groups of peoples.
  6. Digitization and restoration of ancient medical treatises, comparison of the knowledge contained in them with modern medicine
  7. Identification of patterns and effects of treatment with the use of conventional and non-conventional approaches in medical practice, as well as patterns and effects of treatment only by medical professionals or with the involvement of additional specialists: clinical psychologists, medical anthropologists, or doulas.

Health care is traditionally a high-risk area where any mistake can have significant consequences for human life. Accordingly, public safety, ethical issues and all other potentially negative aspects associated with the use of AI in health care should become the central focus of their implementation in practice. It is also necessary to achieve trust in AI systems, both among medical professionals and patients.

References

AuntMinnie.com URL: https://www.auntminnie.com/index.aspx?sec=log&URL=https%3a%2f%2fwww.auntminnie.com%2findex.aspx%3fsec%3dsup%26sub%3daic%26pag%3ddis%26ItemID%3d130473 (30.11.2021)

Bahl M, Barzilay R, Yedidia AB, Locascio NJ, Yu L, Lehman CD. (2018) High-Risk Breast Lesions: A Machine Learning Model to Predict Pathologic Upgrade and Reduce Unnecessary Surgical Excision. Radiology. Vol. 286 (3) pp 810–818.

Bhattad Pradnya Brijmohan, Jain Vinay (2020) Artificial Intelligence in Modern Medicine – The Evolving Necessity of the Present and Role in Transforming the Future of Medical Care USA East Tennessee State University Cureus, Vol. 12 (5) pp 1-5

Gurevich Alexey, Mikheenko Alla, Shlemov Alexander, Korobeynikov Anton, Hosein Mohimani & Pavel A. Pevzner (2018) Increased diversity of peptidic natural products revealed by modification-tolerant database search of mass spectre Nature Microbiology. Vol. 3, pages 319–327

Intelligent ultrasound URL: https://www.intelligentultrasound.com/wp-content/uploads/2021/09/PNB-brochure-220921_digi.pdf (30.11.2021)

Koomey Jonathan G, Berard Stephen, Sanches Marla (2011) Implications of Historical Trends in the Electrical Efficiency of Computing IEEE. Annals of the History of Computing, Vol. 33(3) pp 46 – 54

Markoff John (2015) Machines of Loving Grace: The Quest for Common Ground Between Humans and Robots, Ecco, 400p

MIT – Massachusetts Institute of Technology URL: https://web.mit.edu/innovation/ (30.11.2021)

Pirzada Usman (2014) Did Nvidia Just Demo SkyNet on GTC 2014? – Neural Net Based “Machine Learning” Intelligence Explore URL: https://wccftech.com/nvidia-demo-skynet-gtc-2014-neural-net-based-machine-learning-intelligence/ (30.11.2021)

Taylor Ian (2015) Alan M. Turing: The Applications of Probability to Cryptography the National Archives in the UK at www.nationalarchives.gov.uk using reference number HW 25/37, 33p

UN news URL: https://news.un.org/en/story/2021/06/1094902 (30.11.2021)

Weng SF, Reps J, Kai J, Garibaldi JM, Qureshi N (2017) Can machine-learning improve cardiovascular risk prediction using routine clinical data? Harbin Institute of Technology Shenzhen Graduate School, CHINA PLoS ONE Vol. 12(4) URL: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0174944 (30.11.2021)