![](/upload/iblock/585/hnmx6ir6f1gbrom8i4tkego7dx5e255g/doctor.png)
The use of artificial intelligence in healthcare was the topic of discussion at a meeting of the discussion club "Digital Reality" on August 17.
The guests of the stream were Igor Shulkin, Deputy Director for Prospective Development of the Moscow State Budgetary Health Institution "Scientific and Practical Clinical Center for Diagnostics and Telemedicine Technologies of the Moscow City Health Department", Alexander Gusev, Director for Business Development of K-Sky LLC (Webiomed), Deputy Head of the Department for Methodological Support of a Comprehensive Assessment of Technologies in Healthcare of the Federal State Budgetary Institution "Center for Expertise and Quality Control of Medical Care of the Ministry of Health of Russia" Filipp Gorkavenko and project manager in the direction of "Medicine" LLC "Ntech Lab" Dmitry Romanov. The meeting was moderated by Mikhail Anisimov, ICANN Stakeholder Engagement Representative for Eastern Europe and Central Asia.
The speakers brought up a wide range of current topics, including the role of artificial intelligence in medicine, the level of maturity of AI products for healthcare, access restrictions on anonymized medical data for AI training, and comparisons to the life cycle of medicines.
Igor Shulkin spoke about the current requirements for storing medical data (which, among other things, can be used for further training of AI) in radiology: “Images obtained during radiology are stored in a format that meets the international standard: there is a clear requirements and so on. But there is a problem of low-quality data: when not all the necessary tags are filled in or filled out non-standard. This is a matter of setting up diagnostic equipment and the work of engineers, and the people who use this data in order to study and analyze it must always check the quality of the original data.”
“For about ten years, both in our country and in the United States, there has been a discussion about the need to standardize electronic medical data in terms of coding clinical information. Unfortunately, the matter has not gone further than conversations - there is still no real compatibility and a unified approach in this part. Moreover, almost all systems for maintaining electronic medical records were not designed and operate in order to train Artificial Intelligence on this data. In other words, the current state of information encoding and storage in electronic medical records is not ready to be used by machine learning,” Alexander Gusev described the situation with electronic medical records.
“In my opinion, the future belongs to data that is purposefully collected as part of routine activities. You have a task, you organize the collection of data, make efforts to increase their standardization and completeness - then you can hope that they will be of high quality and complete. In other cases, the applicability of routinely collected data will be limited,” added Philip Gorkavenko.
“Everything that we are discussing here today is, of course, possible. But there is another side of the problem - infrastructure, high-speed networks, capacities with 3D accelerators. You need to have something where to implement artificial intelligence. Graphical medical data in many regions is usually stored in a loosely structured form - so that it is simply impossible to use it. For example, the banal task of picking up five hundred of the same type of studies with a specific pathology somewhere in the region most often requires manual labor and is not automated at all. Artificial intelligence is a high level of automation. We did an internal study: we took 1000 x-rays and asked three radiologists to describe them - the level of agreement on diagnoses was about 50%. The question arises: how much worse is an AI that has passed normal testing worse than a real doctor with accumulated fatigue and stress?” Dmitry Romanov said.
“There is a good trend - in recent years, healthcare leaders have clearly begun to understand that all medical data needs to be structured, learn how to process it, store it and draw conclusions based on it,” Igor Shulkin added to the words of his colleagues.
You can follow the discussion in detail on our YouTube channel.