Overview

Artificial intelligence (AI) has infiltrated many areas of our daily lives, and medicine is no exception. As in other fields, the introduction of AI algorithms into medical practice has been met with concern, scepticism, and a lack of understanding of how AI works. Questions arise about the role of physicians and patients in the decision-making process, the reliability of data and outcomes, ethics, transparency, accountability, and responsibility.

At the same time, the application of AI can be beneficial in various areas of medicine, including drug discovery, medical diagnosis, patient education, medical training, healthcare administration, and decision-making.^1,2

In this article, we will discuss the history of AI development in medicine, its algorithms and applications, AI value in a decision-making process, as well as the risks and challenges.

From artificial to augmented intelligence

According to the American Medical Association (AMA), AI in medicine is not a new phenomenon. First appearing in the mid-20s, AI tools in healthcare at that time were rule-based algorithms that relied on human logic. Fifty years later, the use of more sophisticated predictive models has become common in medicine, thanks to more affordable computing power and greater access to data. Recently, generative AI models have emerged. These algorithms are capable of processing huge and multimodal datasets (text, images, and video) that are considered the next generation of AI in medicine and society.

Such models have the potential to support decision-making processes in medicine. Increasingly, the term 'augmented intelligence' has come to be used in the medical literature and at an organisational level instead of 'artificial intelligence'. This trend reflects the current vision of AI in healthcare, in which AI tools support, rather than replace, human decision-making.

AMA highlights that usually "artificial intelligence" refers to a computer's capacity to carry out operations that are normally associated with a rational human being. The alternate definition of "augmented intelligence", emphasises the helpful role of AI and highlights how AI design complements human intelligence rather than replaces it. 

Algorithms and types of AI systems in medicine

How AI works

AI algorithms can operate in a variety of ways, such as rules-based learning, machine learning, or deep learning, which uses multi-layered neural networks modelled after the human brain to extract complex patterns from input data.

Systems that learn from data without explicit programming are referred to as machine learning. As a subset of machine learning, deep learning describes systems that learn by analysing information and data in a way that mimics the neural connections seen in the human brain.

What AI does

Some AI applications include computer vision, which extracts information from images or videos; natural language processing, which extracts information from text; and generative AI, which creates new and original content.

AMA distinguishes five AI capabilities:

• Identifying objects, trends, and characteristics within data (including images)
• Translating data into another data format or type
• Summarising data inputs into shorter and more accessible outputs
• Predicting future events based on historical data and trends
• Providing recommendations, guidance, or advice

All of the above functions can be useful in the decision-making process, although the last one can be considered the most significant.

AI in medical decision-making

Healthcare professionals have many opportunities to influence patient well-being through clinical decision-making. Patients may be vulnerable in this process because of their health status and lack of medical knowledge. Doctors assess the condition of their patients and suggest treatments based on the results of this assessment.

Healthcare professionals receive specialised training and are continuously and carefully checked to maintain standards of practice. However, even highly qualified medical professionals are not immune to making mistakes in the decision-making process.³

Clinical decision-making

According to research papers, the topic of clinical decision-making may contain various aspects where AI can be applied. These include remote monitoring, computerised chart interpretation, and facilitating diagnosis and prognosis.⁴

Remote monitoring 

Remote monitoring using AI can help classify physical activity and detect abnormal events or behaviours using wearable sensors and cameras. This can help remotely monitor patient mobility, safety, and adherence to treatment plans. AI also has the potential to work in emergency situations, assisting in monitoring cardiovascular patients in intensive care units.⁴

Computerised chart interpretation

One possible application of AI in computerised chart interpretation in medicine is the use of machine learning algorithms to analyse electrocardiogram signals. A deep learning model called vision transformers has gained popularity in medical imaging, particularly in the case of lung cancer.

They analyse image classification as a sequence prediction task, capturing long-term dependencies in image regions. These results demonstrate the significant potential of using AI in analysing and interpreting medical charts.⁴

Facilitating diagnosis and prognosis

Using deep learning models, AI can assist in the prevention and diagnosis of a wide range of diseases, including mental health disorders, cancer, diabetes, Alzheimer's disease, etc. This can reduce human errors and help deliver a quicker and more accurate diagnosis. In addition, AI has demonstrated the ability to optimise the process of prescribing medication. Moreover, one article reported that AI can help triage patients by offering guidance and assessing the severity and prioritisation of cases.⁴

Recently, software development companies have created AI systems to assist in clinical decision-making. Currently, these systems do not need to get close to the patient; they simply process input data about the patient and generate output data that advises the attending physician. The system examines the patient's medical record, processes the information, calculates the patient's treatment options, and then presents the doctor with a list of treatments in rank order.³

The perceived advantages of this kind of system are as follows:

• it is able to access and process complex medical data much quicker
• stays up to date with the latest data (which is published in such volume and at such a speed that it is difficult for clinicians to keep up with it completely)
• uses all this data to create individualised plans for the patient³

Such a system creates the potential for physicians to improve information management by providing access to up-to-date knowledge (e.g. clinical trials), freeing up time, leading to increased capacity for patient care and perhaps even the potential to reduce physician burnout. Any abandonment of AI is therefore subject to consideration of opportunity costs.³

Organisational decision-making

This area includes two aspects: predicting administrative and quality indicators and providing cost-effective solutions for resource management.⁴

Predicting administrative and quality indicators

Using machine learning algorithms, AI can assist in predicting the results of healthcare processes, including patient satisfaction, safety, quality of care, and efficiency. It can help identify best practices, track progress, and provide feedback and suggestions for improvement.⁴

Providing cost-effective solutions for resource management

AI can help automate and optimise administrative tasks such as scheduling, billing, reporting and customer service using natural language processing and machine learning. This saves time and improves efficiency.⁴

Shared decision-making

Shared decision-making involves three aspects: providing personalised information, enabling patient self-management and improving patient adherence to treatment.⁴

Providing personalised information

AI can help create personalised educational software that can tailor learning content, pacing, feedback, and assessment for each individual learner. This technology can help provide personalised education and guidance to patients and carers through natural language processing and chatbots.⁴

Enabling patient self-management

One study investigated the use of learning algorithms to improve patient self-management; the system reduced the time and number of visits required by physicians, and patients expressed satisfaction with this. Moreover, no loss of monitoring was noted. AI can help patients with musculoskeletal pain, such as neck or low back pain, improve their physical activity, posture, and coping skills through a smartphone app that adapts to their individual needs and preferences. Similarly, AI can help diabetic patients manage their blood glucose levels, diet and exercise with a chatbot that provides training and coaching based on natural language processing and machine learning.⁴

Improving patient adherence to treatment

AI can help patients with chronic diseases such as diabetes, hypertension, or cancer adhere to their medication regimens by providing reminders, alerts, education, and coaching through mobile phone apps, chatbots, or wearable devices. In addition, AI can prevent medication non-compliance by using machine learning algorithms and clinical data such as patient characteristics, disease severity, or psychosocial factors.⁴

Challenges and risks

As AI tools evolve, the medical community must address several key challenges:

• Bias. They can be caused by historical policies, sampling, and algorithms. AI tools trained on biassed datasets can perpetuate existing inequalities in healthcare. Cases where AI exacerbates social inequalities, such as by disproportionately targeting certain patient groups or inaccurately assessing needs based on biassed data, highlight the need for careful assessment and mitigation of bias
• Transparency. Access to information about the training data and details of the AI model is essential to determining its suitability and reliability in practice. Transparency also involves documenting AI use in healthcare to inform patients and ensure ethical use
• Reliability. AI models can produce results that are factually inaccurate. Understanding and minimising such errors is a priority for developers
• Responsibility. This issue is complex and unclear. Legal frameworks such as product liability law, medical malpractice law, and ordinary negligence may apply, but the specifics will depend on the context and impact of AI tools
• Privacy and security. Protecting patient data used in the development and training of AI is critical. Existing regulations need to be updated to address AI-specific concerns such as data reuse and potential security issues
• Regulation. Effective regulation requires collaboration among health care professionals, policymakers, AI experts, and other stakeholders

FAQs

How does artificial intelligence (AI) work?

Artificial intelligence (AI) is computer technology that allows programmes to think and suggest as humans do. AI uses algorithms and datasets to learn and make decisions from this data.

How is AI being applied to healthcare?

Using AI, healthcare professionals can receive updates, analyses, and automatically generated reports, thus saving time and drawing more attention to the need for preventive procedures during patient appointments.

How is AI changing the future of medicine?

AI has the potential to fundamentally change the entire world of medicine, facilitating the development of new drugs, transforming the diagnostic system, improving the quality of medical services, and reducing costs. 

Summary

There is no doubt that AI will change the world of medicine, including the decision-making process. In order to ensure the harmonious development of AI, it is necessary to pay attention not only to its exciting opportunities but also to the risks and challenges that humanity still has to deal with.