Translation. Region: Russian Federal
Source: Novosibirsk State University –
An important disclaimer is at the bottom of this article.
A prototype of the world's first medical decision support system, the "Digital Physician Assistant 'Dr. Pirogov,'" was developed by scientists at the Research Center for Artificial Intelligence (AI Center) of Novosibirsk State University. It is based on a hybrid technical solution combining neural and semantic networks. The digital physician assistant contains information on 250 major diseases. This number will increase in the future, as the system will be expanded to include information on additional pathological conditions. Information on these conditions is currently being systematized and verified. The developers used a hybrid approach combining neural network methods and a specialized knowledge graph, ANDSystem, to ensure the interpretability of decisions. The prototype was created using research and development conducted at the Institute of Cytology and Genetics of the Siberian Branch of the Russian Academy of Sciences.
– Developing the Doctor Pirogov system, we set ourselves the task: to reduce the routine load on the doctor and reduce the duration of the patient without loss of quality. This system is also able to provide consulting support and support medical decisions in the provision of primary medical care in a wide range of medical directions, which is especially relevant with the current deficiency of narrow specialists in remote and sparsely populated regions. Our “Doctor Pirogov” allows the patient’s interactive survey. We plan to implement both voice and visual interface. The system “studies” medical documentation, analyzes clinical information and instrumental research, evaluates laboratory indicators and results of genetic testing. She does all this, relying on the semantic columns of knowledge, built on the basis of an analysis of the knowledge contained in its basis of medical and scientific literature. The results of the work of the Doctor Pirogov system are a list of probable diagnoses on pathophysiological justifications, a list of necessary additional examinations, therapy recommendations taking into account drug interactions, and recommended personalized preventive measures for each particular patient. At the moment, we have created a prototype of the system, the main functions of a digital assistant have been successfully worked out, ”said the leading researcher, project manager at the NSU Artificial Intelligence Center, the head of the laboratory of artificial intelligence and large genetic data ICIG SB RAS, and the head of the computer proteomics ICIG, Vladimir Ivanisenko.
The Doctor Pirogov system combines knowledge across 20 medical specialties, including internal medicine, cardiology, endocrinology, neurology, gastroenterology, infectious diseases, pediatrics, oncology, psychiatry, dermatology, hematology, nephrology, rheumatology, and others. This allows it to be used as a universal digital assistant for general practitioners, emergency physicians, and specialists.
The developers of the Doctor Pirogov system are confident that the digital physician assistant will be extremely useful in rural areas where paramedics see patients. It will assist these medical professionals in making treatment decisions and referring patients to diagnostic centers for additional testing or to other medical facilities for specialized care. Doctor Pirogov can be used for preliminary patient interviews in pre-hospital care settings and to support the physician during appointments. Patients can enter their personal information, complaints, and test results while waiting for their appointment—the user-friendly interface ensures a seamless experience. Based on the patient's initial information, the system will generate a list of possible conditions, ranked by risk. During the appointment, the physician will then conduct further interviews to confirm the diagnosis, if necessary, and Doctor Pirogov will recommend additional tests and treatment options. The physician decides whether to follow these recommendations.
"Doctor Pirogov" will be an assistant to healthcare professionals. It won't replace doctors, but it will significantly facilitate their work. When used in pre-hospital care settings, it will conduct an initial patient interview and analyze existing clinical data, issue a referral for lab tests or a doctor's appointment, including a recommendation for a specialist. This will allow for efficient patient flow, ensure preliminary triage and routing, and reduce the workload of community healthcare workers. Supporting the doctor during appointments is equally important. Here, our system will expedite the processing of patient complaints, their test results, and other clinical data. Supporting clinical decision-making based on a database of medical and scientific knowledge is also crucial. This will reduce the duration of patient appointments without sacrificing quality and reduce the risk of diagnostic errors. "Doctor Pirogov" will be especially useful for doctors in rural and remote areas. It will be able to provide consulting support in many medical specialties where specialist availability is limited and analyze lab data, instrumental examinations, and medical records. This will improve the quality of initial diagnosis and treatment, reduce the need to refer patients to the district center, and reduce the workload of rural doctors thanks to the AI system's interpretation of complex clinical data, explained Vladimir Ivanisenko.
It took scientists from Novosibirsk State University and the Institute of Cytology and Genetics SB RAS 10 years to create the semantic network embedded in the Doctor Pirogov system. This work was conducted at the Institute of Cytology and Genetics SB RAS under the leadership of Academician Nikolai Kolchanov. The developers were tasked with finding pharmacological targets and developing drugs for a number of common diseases. They needed to determine which gene a specific drug targets and then, based on this information, develop the chemical structure of the drug.
– A person has about 20 thousand genes, they interact with each other. At the same time, one gene is able to suppress or increase the activity of another. Depending on this, they can react differently to a particular drug. To solve the problem of choosing targets for the action of the medicine, it is necessary to determine which genes are associated with a certain disease and how these genes interact with each other. It was obvious to us that it was impossible to do without the help of artificial intelligence in this situation, so we began to build a semantic network. Its difference from the neural network is, when teaching neural networks, knowledge is distributed through the so -called Libra in the form of some numbers. To make the methods of AI interpreted, it is necessary to distribute the knowledge gained in training in the form of semantic networks. We had 20 thousand genes, each of them became a peak, and facts – about 30 thousand diseases. Between them are the ribs – relationships. Our system was supposed to take into account everything – risk factors, the influence of the external environment, mutations in genes, the physiological parameters of the body. All this information was contained in 50 million scientific publications. One person working 8 hours a day and spending 2 minutes on reading one article would take 300 years to perform this work. Every year, an average of 1.5 million publications appear. In this work, to extract facts from the texts, we involved neural networks, applying a thin tuning method to them, for which 25 thousand rules were previously manually registered. As a result, a semantic network was built, where about 40 thousand facts of the ratio of genes and symptoms with various diseases were installed, ”said Vladimir Ivanenko.
A large team of researchers was involved in this large-scale project. Students from NSU and other Novosibirsk universities also contributed to the creation of the semantic network of the specialized ANDSystem knowledge graph. Under the supervision of Vladimir Ivanisenko alone, 40 student papers were written. Each year, 10-12 students participated in the work as part of their summer internship. Six PhD dissertations were defended, and over 150 scientific articles were published, devoted to the analysis of various diseases using this information extraction method. Initially aimed at solving scientific problems, the developers later decided to adapt it for practical medicine. A semantic network, unlike the human brain, can store and, if necessary, retrieve a much larger volume of information about medications and their compatibility with each other, side effects and contraindications in the presence of comorbidities, and much more. The use of AI will help avoid errors in prescribing medications, determining patient management strategies, and their rehabilitation. Currently, there are no analogues of this system in the world, and only four similar systems are known that perform scientific tasks in the field of genetics.
The digital physician assistant's functions have now been successfully tested, and the developers of Doctor Pirogov are now faced with the task of implementing it on a large scale. To achieve this, it is necessary to obtain the appropriate permitting documentation for classification as medical software, register with Roszdravnadzor, and obtain an assessment of compliance with the requirements of the EAEU Technical Regulation and Standard 047/2018. Systematic solutions at the Russian government level are also needed: the creation of a regulatory sandbox for testing AI in clinical practice, a simplified registration procedure for AI-based medical solutions subject to physician oversight, and the development of Methodological Recommendations from the Russian Ministry of Health for integrating digital assistants into primary care. Once all these steps are completed, medical facilities will need to retrofit their offices with computerized workstations for patient-doctor interaction with the AI.
Clinical trials of the Doctor Pirogov system will begin next year. Internal verification of its scientific validity is currently underway.
Material prepared by: Elena Panfilo, NSU press service
Please note: This information is raw content obtained directly from the source. It represents an accurate account of the source's assertions and does not necessarily reflect the position of MIL-OSI or its clients.
