Trust and verify: how neural networks will start treating Russians
In 2026, Russia plans to introduce large-scale artificial intelligence technologies into the processing of electronic medical records of all Russians. Algorithms will analyze the research data, and the results will be sent to a single digital storage system. It is assumed that this will speed up diagnosis, increase the accuracy of medical decisions and reduce the burden on medical staff. At the same time, the final clinical conclusions will still remain with the doctor, the Ministry of Health emphasized. All the details are in the Izvestia article.
Why is a single medical database being created?
Artificial intelligence will start processing all electronic medical records of Russians this year. The research data that the AI will decode will be collected in a single database. The Ministry of Health plans to train health workers in the use of neural networks in professional practice. At the same time, AI will only become an assistant in decision-making, and the doctor will be responsible for diagnosis and treatment. This was stated on the sidelines of the State Duma by Deputy Minister of Health Vadim Vankov.
This year, doctors across the country will begin to undergo extensive training in working with artificial intelligence. AI will help relieve doctors and administrators by taking over some of the routine tasks in hospitals and polyclinics. In addition, soon all patient images (X-rays, CT scans, MRI scans) and half of the medical reports will be included in a single database. Artificial intelligence will analyze at least 85% of mammograms. With the help of AI, all electronic medical records will also be processed.
The doctors themselves perceive the transition to such an approach positively. Elena Satirova, oncologist of the highest category, Medical director of Neuromed LLC, emphasizes that the larger and better the database, the deeper and more accurate the answers that the system gives.
The key goal of the new AI infrastructure is to combine disparate medical data into a single digital circuit. This will allow doctors to see not individual episodes of treatment, but the entire history of the patient's condition: the results of tests, studies, disease dynamics, and prescribed therapy.
— The new unified database is a transition to a predictive healthcare model. We are talking about creating an end—to-end digital patient history that allows AI to analyze data over time and identify patterns," explains Alexander Zhegalov, head of AI at FMF.dev.
In the future, such systems can significantly change the very logic of medicine. Automation of the treatment process and the goals of clinical decisions are inevitable in the context of an avalanche-like increase in the amount of available information, says Alexander Gushchin, professor of the Department of Clinical Engineering and Artificial Intelligence Technologies at Volgograd State Medical University (VolgSMU).
According to the experts interviewed, the use of AI already makes it possible to detect diseases at earlier stages, as well as analyze medical statistics at the regional and national levels.
Where will medical data be stored?
One of the key elements of the new system will be the infrastructure for storing medical information. We are talking about creating a large—scale network of repositories where research results will be received, from laboratory tests to X-ray and tomographic images.
In particular, it is planned to streamline the work of the radiological service so that the research results flow into the Central Archive of Medical Images. Regional archives are already actively filling up: in a number of regions, their occupancy reaches 80-90%. At the same time, data volumes continue to grow rapidly: one medical image can take up to 2 GB.
In such a situation, the healthcare system needs powerful data centers capable of storing and processing huge amounts of information, experts say.
"Data storage will be organized through secure government data processing centers and specialized medical image archives," says Dmitry Ivankov, an expert at the SKB Kontur AI Center.
At the same time, experts note that we are not talking about a single physical database where all medical information will be stored. Rather, a distributed system will appear in which data remains in regional information systems, but is combined into a common analytical circuit, explains Andrey Povarenkin, Deputy General Director of the Third Opinion Platform.
This architecture will allow doctors to access the necessary information about the patient regardless of the region of his residence, as well as analyze medical data at the level of the entire healthcare system.
"The system will allow aggregating large amounts of clinical information, identifying patterns and reducing the burden on doctors by automating routine tasks," adds Sergey Golitsyn, Head of T1 AI.
How to keep medical information up-to-date
Another important issue is keeping medical data up—to-date. An artificial intelligence system is effective only when it works with accurate and timely updated information. Medicine is developing rapidly: new treatment protocols are emerging, clinical recommendations are being updated, and diagnostic methods are being improved. Therefore, information systems must constantly adapt to these changes.
This requires regular updating of algorithms and retraining of artificial intelligence models based on new data, emphasizes Stanislav Yezhov, Director of AI Development at Astra Group.
However, the discipline of working with medical information is no less important. If data is entered incompletely or with delays, this reduces the effectiveness of any analytical tools.
"The relevance of the system depends not so much on the AI itself, but on the discipline of updating data and the quality of its markup," warns Kirill Smelovets, chief architect of Cloud X artificial intelligence and machine learning services.
Automation of data transmission from diagnostic equipment can significantly improve the accuracy of data. In this case, the research results will be entered into the system almost immediately after the procedure.
"If information is automatically pulled up from diagnostic equipment and medical systems, it can be updated almost in real time," says Yuri Tyurin, MD Audit's technical director.
The less manual data entry, the lower the risk of errors and the more reliable the entire analysis system works, the expert is convinced.
How will AI errors be detected and corrected
Despite the high level of technology development, artificial intelligence systems are not completely autonomous. In medicine, they are used as a decision support tool, not as a substitute for a doctor. Algorithms analyze medical data, compare it with the accumulated knowledge base and come up with possible conclusions. However, the final decision on diagnosis and treatment is made by an on-site specialist, says Viktor Gombolevsky, a leading researcher at the AIRI Institute.
The quality control of algorithms is based on a multi-level verification system. Before implementation, medical AI services are tested, and then their work is constantly monitored in real clinical practice, says the commercial director of FabricaONE.AI Roman Smirnov.
Special attention is paid to error analysis. They make it possible to improve algorithms and improve diagnostic accuracy.
"False negative predictions, when AI says that there is no pathology, but it actually exists, are the most valuable growth area for improving the quality of AI systems," says Nikita Nikolaev, co—founder and CEO of Celsus. Such cases are used to retrain models, which gradually increases the accuracy of the system.
What problems are possible at the start of the AI application
Experts agree that the main difficulties of implementing artificial intelligence are related not so much to the algorithms themselves as to the quality of the source data and the features of the medical infrastructure. Many medical institutions use different information systems and diagnostic equipment, which can create difficulties when combining data into a single digital environment. In addition, the quality of medical images and the format of information storage differ.
— The first pains will not be from a lack of tools, but from data, — says Kirill Pronin, head of the development group at Neuromed LLC.
The heterogeneity of medical data can create additional "noise" for algorithms, reducing the accuracy of the analysis. Different devices, the number of images and protocols can lead to false alarms, warns Ilya Fomichev, head of AI at SimbirSoft.
Another challenge may be the adaptation of medical personnel to new digital tools. In parallel, the healthcare system will have to solve the tasks of unifying medical data, synchronizing regional information systems and ensuring reliable protection of patients' personal information. Without these conditions, the full-fledged operation of analytical tools is impossible.
The editorial board of Izvestia sent a request to the Ministry of Health of the Russian Federation. No response had been received at the time of publication.
Переведено сервисом «Яндекс Переводчик»
