AI in Medicine: Examples, Pros and Cons, Ethics, and the Future

AI in medicine, or artificial intelligence in medicine, refers to programs that learn from medical data and assist physicians: they detect abnormalities on scans, read ECGs, process documentation, and search for novel drug molecules. However, the final decision always remains with the physician. Below are the application areas with real-world examples, pros and cons, ethical issues, the current landscape in Russia, and a ready-made report outline.

What is artificial intelligence in medicine

AI in medicine is commonly understood to mean machine learning methods, most often neural networks. Therefore, "AI" and "neural networks in medicine" are practically interchangeable in conversation. The main AI technologies in medicine include computer vision (scans, photos, videos), text and speech processing, and predictive models based on electronic health record data. The model receives thousands of labeled examples (scans with confirmed diagnoses, ECG tracings, medical history texts) and learns to identify the same patterns in new data. A separate class is represented by large language models like ChatGPT: they work with text, answer questions, and write summaries, but can confidently produce errors.

It is important to distinguish between two operational modes. Most often, AI acts as a "second opinion": it highlights a suspicious area, while the final conclusion is signed off by a physician. Autonomous systems that make decisions without a physician are extremely rare, and each undergoes separate registration.

AI applications in medicine: 11 areas with examples

1. Diagnostic radiology: X-ray, CT, mammography

The most mature area. Algorithms search for lung nodules on CT scans, signs of pneumonia on radiographs, and suspicious masses on mammograms. In a study by McKinney et al. (Nature, 2020), a mammography analysis system using retrospective data from the UK and the US yielded fewer false-positive and false-negative results than physicians.

2. ECG

Automated electrocardiograph interpretations have existed for a long time; neural networks expand their capabilities by recognizing atrial fibrillation and other rhythm disturbances, and helping to triage tracings by urgency. The interpretation is still verified by a physician, and the skill of reading ECGs remains essential, which is convenient to practice in the ECG simulator.

3. Dermatoscopy

In a study by Esteva et al. (Nature, 2017), a neural network was trained on 129,450 clinical skin images and compared with 21 certified dermatologists: in the task of distinguishing melanoma and carcinoma from benign lesions, it demonstrated a level comparable to specialists.

4. Diabetic retinopathy

In 2018, the US FDA cleared IDx-DR, the first autonomous AI diagnostic system: it analyzes fundus photographs in diabetic patients and provides a result without an ophthalmologist, enabling screening at the primary care level.

5. Clinical decision support

Clinical decision support systems (CDSS) analyze electronic health record data and prompt physicians regarding complication risks, potential drug-drug interactions, or missed examinations according to clinical guidelines. These are suggestions, not prescriptions.

6. Medical texts and documentation

Speech recognition turns dictation into medical records, language models generate discharge summaries, structure complaints and anamnesis, and search for required information in lengthy documents. For physicians, this saves time on routine tasks; for students, it is a way to quickly find the necessary section in clinical guidelines.

7. Drug discovery

AlphaFold (Google DeepMind) predicts the 3D structure of proteins from their amino acid sequences. A paper by Jumper et al. was published in Nature in 2021, and in 2024 Demis Hassabis and John Jumper were awarded half of the Nobel Prize in Chemistry for this work. Another example is halicin, an antibiotic described in Cell in 2020, which was discovered by a neural network among existing molecules and showed activity against drug-resistant bacteria in mouse experiments.

8. Personalized medicine

Models combine genetics, laboratory tests, and medical history to assess individual risk and select tailored therapy, such as predicting tumor response to treatment or pharmacogenetic aspects of drug metabolism.

9. Wearable devices

In 2018, the FDA cleared the ECG feature on the Apple Watch: the watch records a single-lead ECG and notifies the user of signs of atrial fibrillation. The diagnosis is made by a physician; the watch only provides a reason to seek medical attention.

10. Telemedicine and chatbots

Chatbots collect patient complaints prior to visits, remind patients to take medications, and assist with navigation. The main risk is that patients may mistake a chatbot's response for a diagnosis.

11. Healthcare organization and education

AI predicts department workloads, helps schedule staff, and checks the quality of documentation. In education, it explains topics, creates study notes and tests, and conducts mock exams, as discussed below.

AI in medicine in Russia: implementation and regulation

• Moscow computer vision experiment. Since 2020, Moscow has been running an experiment on the use of AI in diagnostic radiology: algorithms analyze radiographs, CT scans, and mammograms in the routine workflow of municipal polyclinics and hospitals. The experiment operator is the Center for Diagnostics and Telemedicine, via the mosmed.ai platform.
• Roszdravnadzor registration. Software utilizing AI intended for diagnosis or treatment is classified as a medical device and must be registered with Roszdravnadzor before clinical use.
• Standards. The GOST R 59921 series of standards, "Artificial Intelligence Systems in Clinical Medicine," is in effect, covering clinical evaluation, technical testing, and datasets for training and testing.
• Strategy. By Presidential Decree No. 490 dated October 10, 2019, the National AI Development Strategy for the period up to 2030 was approved (updated by Decree No. 124 dated February 15, 2024). The federal project "Artificial Intelligence" is included in the national project "Data Economy.".

Pros and cons of AI in medicine

Key risks in detail

Errors and hallucinations. A model that performs well in a published study may perform worse on different equipment or in a different patient population. Furthermore, language models are prone to fabricating non-existent sources. Therefore, any response must be cross-checked with the primary source, as discussed in detail in the article ChatGPT for Physicians.

Data bias. A notable case was described by Obermeyer et al. (Science, 2019): an algorithm in the US assessed the need for additional care based on past healthcare expenditures, and due to unequal access to medical care, underestimated health risks for Black patients.

Privacy. Health status and diagnostic data constitute medical confidentiality (Art. 13 of Federal Law No. 323-FZ) and belong to special categories of personal data (Art. 10 of Federal Law No. 152-FZ). Entering a patient's full name, date of birth, address, or insurance policy number into a public chatbot is prohibited.

Ethical issues of AI in medicine

• Autonomy and consent. Should a patient know if an algorithm reviewed their scan, and can they refuse?
• Fairness. The model must not produce worse outcomes due to gender, age, region, or social status.
• Transparency. The physician must understand what the recommendation is based on, otherwise they cannot critically evaluate it.
• Accountability. Who is liable for an error: the physician, the clinic, or the developer?
• Confidentiality. Training models requires large arrays of patient data.
• Dehumanization and loss of skills. The risk that communication with the patient will decrease and the physician will forget how to think without a prompt.

In 2021, the WHO formulated six principles for ethical AI in healthcare: protecting human autonomy, promoting human well-being and safety, ensuring transparency and explainability, fostering responsibility and accountability, ensuring inclusiveness and equity, and promoting AI that is responsive and sustainable. In 2024, the WHO issued separate guidance on large multimodal models. In Russia, the Code of Ethics in AI (2021) was adopted, and for healthcare, the interagency working group under the Russian Ministry of Health approved the Code of Ethics for the Application of AI in Healthcare.

Will AI replace the physician?

In the foreseeable future, no. AI is good at solving narrow tasks: finding a nodule on a CT scan, recognizing a rhythm on an ECG. Meanwhile, the physician takes the history, examines the patient, reconciles contradictory data, explains the diagnosis to the patient, and bears legal liability. Even in diagnostic radiology, which is most often cited as a candidate for replacement, AI currently operates as a second opinion, while the report is signed by the radiologist.

A realistic scenario is different: the nature of work changes. There is less routine, and the ability to verify a machine prompt and understand where it made a mistake becomes part of one's qualifications. A comparison of models that physicians are already using can be found in our review of neural networks for physicians.

AI in medical education

For a student, AI is useful primarily as a tutor and a routine assistant. The main requirement: answers must be based on verifiable sources, rather than the model's "memory". This is how the neural network for medical professionals by WhiteCoat Pro is designed:

• The AI tutor answers based on the platform's study guides (31 subjects, 3,400+ video lessons) and clinical guidelines, provides footnotes to study guide pages, and shows which video review to watch;
• study notes are compiled from a lecture recording, a video via link, a photo of a notebook, a PDF or Word file, complete with tests and flashcards for review;
• presentation: first the slide outline, then generation and export of PPTX and PDF;
• oral exam via voice: the examiner asks questions on the topic, followed by a checklist-based debrief after the call.

Works in Russia without a VPN. The restriction is the same as for any AI in medicine: it is a study tool, it does not prescribe a diagnosis or treatment for a specific individual, and patient personal data must not be entered. Read more about use cases in the article neural network for medical students.

Ready-made outline for a report and presentation on "AI in medicine"

1. Title slide: "Artificial Intelligence in Medicine", Full Name, group, department.
2. What are AI, machine learning, and neural networks: definitions and model training flowchart.
3. Brief history: from expert systems to large language models.
4. Image-based diagnostics: X-ray, CT, mammography, dermatoscopy, fundus imaging.
5. ECG and wearable devices.
6. Clinical decision support and medical documentation workflow.
7. Drug discovery: AlphaFold and antibiotic search.
8. AI in medicine in Russia: the Moscow experiment, medical device registration, GOST R 59921, national strategy.
9. Pros and cons: a table of benefits and risks.
10. Ethical and legal issues: WHO principles, medical confidentiality, personal data.
11. Will AI replace the physician: arguments and conclusion.
12. Conclusions and references.

Tip: keep one thesis and one example per slide, and take numerical figures only from primary sources. A step-by-step guide on how to put together such a presentation is covered in the article generating a medical presentation with a neural network.

References and sources for the report

1. Decree of the President of the Russian Federation No. 490 dated October 10, 2019, "On the Development of Artificial Intelligence in the Russian Federation" (as amended on February 15, 2024).
2. Federal Law No. 323-FZ dated November 21, 2011, "On the Fundamentals of Protecting the Health of Citizens in the Russian Federation", Article 13.
3. Federal Law No. 270-FZ dated July 27, 2006, "On Personal Data", Article 10.
4. GOST R 59921.1-2022 "Artificial intelligence systems in clinical medicine. Part 1. Clinical evaluation".
5. Gusev A.V., Vladzimirsky A.V., Sharova D.E., et al. Evolution of research and development in the field of artificial intelligence technologies for healthcare in the Russian Federation: results of 2021 // Digital Diagnostics. 2022. Vol. 3, No. 3. pp. 178-194. DOI: 10.17816/DD107367.
6. World Health Organization. Ethics and governance of artificial intelligence for health: WHO guidance. Geneva, 2021.
7. Topol E.J. High-performance medicine: the convergence of human and artificial intelligence // Nature Medicine. 2019. Vol. 25. P. 44-56.
8. Esteva A. et al. Dermatologist-level classification of skin cancer with deep neural networks // Nature. 2017. Vol. 542. P. 115-118.
9. McKinney S.M. et al. International evaluation of an AI system for breast cancer screening // Nature. 2020. Vol. 577. P. 89-94.
10. Stokes J.M. et al. A Deep Learning Approach to Antibiotic Discovery // Cell. 2020. Vol. 180, № 4. P. 688-702.

Frequently asked questions

Where is artificial intelligence currently used in medicine?

Mostly in diagnostic radiology (X-ray, CT, mammography), ECG analysis, dermatoscopy, and diabetic retinopathy screening. Its application in documentation, drug discovery, wearable devices, and education is growing.

What are the main problems of using AI in medicine?

Errors and hallucinations, training data bias, lack of decision transparency, unclear liability for errors, and the protection of medical confidentiality and personal data.

What ethical issues of AI in medicine should be mentioned in the report?

Autonomy and informed patient consent, fairness, transparency, accountability, confidentiality, and the risk of dehumanization. Reference document: the 2021 WHO guidance with its six principles.

Will AI replace the radiologist?

No. AI accelerates finding abnormalities and sorts studies, but the physician signs the report and bears the ultimate liability.

Is it permissible to insert patient data into a neural network?

Personal data must not be used: full name, date of birth, address, insurance policy number. Health information is protected by Article 13 of Federal Law 323-FZ and Federal Law 152-FZ. For educational tasks, data must be anonymized.

Will this article help prepare for the CME test on AI in medicine?

There are no ready-made test answers here, but the article covers the main topics: areas of application, regulation in Russia, risks, and ethics. Unclear questions can be reviewed with the AI tutor using source links.