MMOOZZGG.com

Reference

Types of medical AI: examples and rules for clinics in Russia

7 min read3 sources

Medical AI is software that helps a doctor diagnose and choose treatment, works with medical documents and data, or takes over booking and patient questions. For a clinic, the difference between the types of medical AI is practical: some need state registration as a medical device, others do not. According to the head of Roszdravnadzor, Russia’s healthcare regulator, Russia has already registered 59 Russian medical solutions with AI (August 2026).

On this page · 6 sections

What is called medical AI

Medical AI means software with artificial intelligence that works in healthcare. There are three groups inside, and the rules for them differ.

AI in medicine left the labs long ago: the reviews in this section collect cases from Russian regions, national centers, clinics and universities around the world.

1. Helps the doctor with diagnosis and treatment: analyzes scans, ECGs and test results, prompts the doctor during a visit, assesses risk. The doctor decides, but the software influences the decision.

2. Works with documents and data: prepares drafts of medical documents, transcribes visit recordings, gathers data for reports and research. The effect of such software is measured in doctor hours and doctor surveys.

3. Helps the clinic with patients: answers questions about booking, prices and preparation, books and moves visits, sends reminders. This is clinic service, not medicine.

The line between the groups runs not by technology but by the purpose of the software: what the maker intended it for. The purpose decides whether registration is needed.

Which medical AI needs registration

Under article 38 of 323-FZ, special software that the maker intended for prevention, diagnosis, treatment and medical rehabilitation also counts as a medical device. In Russia, medical devices may be used only after state registration.

On 18 August 2026, Alla Samoylova, head of Roszdravnadzor, said that Russia had registered 59 Russian medical solutions with AI. Software of the first group goes through registration: the tools that take part in diagnosis and treatment.

Software for booking, reminders and organizational questions does not fall under this purpose by the law’s definition. It works with the clinic’s schedule and price list, not with diagnosis and treatment.

For software of the second group, the answer depends on the purpose the maker declared. The criteria, risk class and procedure are covered on our page on registering AI medical devices.

For a clinic this gives a simple rule: ask a vendor of diagnostic AI for the registration certificate number and the risk class, and automate service tasks in parallel, without waiting for registration.

Where medical AI already works

The section covers 16 fields of medicine, each with its own review of cases from Russia and worldwide, with sources and dates. Here there is one line per field: what AI does there.

1. Diagnosis: clinical decision support systems, where a lot of evidence has built up and regulation is strict.

2. Radiology: X-ray, mammography, CT and MRI, where there are many similar images and a clear metric.

3. Oncology: breast screening, colonoscopy, finding a tumor on an ordinary CT, risk assessment by a general practitioner.

4. Cardiology: new information from an ordinary ECG, a standard heart CT and medical record entries.

5. Emergency care and hospitals: round-the-clock monitoring of patient vital signs and a warning to the doctor before clear signs of decline.

6. Primary care: a prompt to the doctor during a visit, screening before the doctor and less load on the front desk.

7. Ophthalmology: analysis of retina images and screening where doctors are scarce.

8. Endoscopy: highlighting suspicious areas in real time while the doctor performs a colonoscopy.

9. Laboratory diagnostics: routine review of cells in a smear and explanation of results for the patient.

10. Drug development: faster search for a molecule, while trials and registration stay the same.

11. Clinical trials: preparation of long document packages and final reports.

12. Telemedicine and monitoring: reminders and data collection between visits, draft replies for the doctor.

13. Pediatrics: measurements on images, reminders to families and draft replies for the doctor under stricter rules.

14. Surgery: planning from CT, calculation of indices and navigation; there are no autonomous operations in the cases.

15. Medical education: an exam for a language model and training doctors to work with AI.

16. Healthcare in Russia: purchases of registered devices by regions, payment for AI scan reviews from compulsory health insurance funds, and infrastructure.

We do not repeat the field texts here. Examples of AI in medicine with numbers and lessons for clinics are in the reviews themselves, and all of them open from the section page.

What the state tells clinics

On 17 April 2026, at the ministry’s annual board meeting, Russia’s health minister Mikhail Murashko said the industry must adopt smart algorithms for non-medical, service tasks too. As examples he named smart scheduling and chatbots that route patient requests.

For a private clinic this means that AI in service matches the ministry’s course, and you can start with tasks that need no registration. What a clinic can do right now is covered in our article on service tasks of clinics.

Patient data

Even service AI works with information protected by medical confidentiality: the very fact that a person contacted a clinic. Which data can be given to AI, on what grounds and where it must be stored is covered on our page on patient data and AI.

Where a clinic should start

Start with organizational tasks that need no registration: booking, confirming and moving visits, reminders, answering patients about prices, address and preparation. Here the result is easy to measure: the share of requests that ended in a booking, and the no-show rate.

Service is better to start first because of timelines: such an agent is rolled out in weeks, while medical device registration, according to market players, takes years.

The assistant passes medical questions to a doctor or receptionist and collects no diagnoses or scans. What such a helper can do and where its limit is, is described on our page on the medical AI assistant.

Pilot: one AI agent on one process in 14 days from $1,499, with the success metric written into the contract, and the clinic sees it in its own medical information system before and after the start. After the pilot the clinic decides: stop, keep the agent on support or build an AI department of 3–5 agents.

A clinic that wants to take part in our research on AI in service will find the program, method and status on the page on AI research in medicine.

Questions and answers

What is medical AI?

AI software that helps a doctor diagnose and choose treatment, works with medical documents and data, or helps a clinic with booking and patient questions. Under article 38 of 323-FZ, Russia’s health protection law, software of the first group counts as a medical device and goes through state registration. In August 2026 there were 59 such Russian solutions.

Does AI for patient booking need registration?

No, if the software solves organizational tasks and takes no part in diagnosis and treatment: by the law’s definition it is not a medical device. Registration is needed for software that the maker intended for prevention, diagnosis, treatment and medical rehabilitation. If the software starts assessing symptoms, it crosses this line.

Sources and fact-checking

Checked by the editors on 16 October 2026.

  1. Vedomosti: Russia has registered 59 Russian medical solutions with AI, 18 August 2026 (vedomosti.ru)
  2. Russian Ministry of Health: annual board meeting for 2025 and plans for 2026, digital transformation, 17 April 2026 (minzdrav.gov.ru)
  3. ConsultantPlus: 323-FZ “On the basics of health protection of citizens”, article 38 “Medical devices” (consultant.ru)

Found a mistake? Tell us

Publications

Mike Mozg

Agentic AI use cases at AT&T: nearly 500 processes now run on agents

AT&T no longer uses AI only to help staff work faster: it is rebuilding whole business processes around agents. SiliconANGLE reported on 14 November 2026 that nearly 500 agentic AI use cases at AT&T are already in production. Here is how the company rebuilds them, where a human still checks the agent, and what other large companies can take from it.

News4 min
Contact us