The study aims to analyze the opportunities and limitations of applying artificial intelligence (AI) technologies within the decision support system (DSS) of the All-Russian Disaster Medicine Service (ARDMS). Specifically, the authors cover the issue of lacking specialized databases for AI training, as well as the prospects of utilizing standard ARDMS emergency response algorithms as a foundation for developing domain-adapted large language models (LLMs). An analysis was conducted on the capabilities of AI-powered media monitoring for the early detection of emergency incidents and assessment of their scale.
Materials and methods: Regulatory documents governing the activities of the ARDMS were analyzed; an assessment of existing information systems, including those used by the ARDMS, was performed; a comparative analysis of the capabilities of modern LLMs was carried out; data on problems related to operational reporting during the mitigation of medical and sanitary consequences of emergencies was systematized.
Results and discussion: Systemic limitations for the application of AI in processing medical data of casualties were identified; an architecture for a hybrid DSS based on a domain-adapted LLM was proposed; the effectiveness of using AI for media monitoring and open-source intelligence analysis was considered.

The article examines the applicability of the industrial digital twin (DT) paradigm in medicine. An analysis of key definitions (GOST, ISO/IEC 30173, DTC) reveals that a mature digital twin implies bidirectional interaction with a physical object, whereas real-world medical applications are either digital models or are limited to one-way monitoring—the socalled "digital shadow."
Direct transfer of industrial approaches proves problematic due to fundamental differences between technical systems and living organisms. This concerns the inherent impossibility of creating a complete model of a human being due to complexity and biological variability, the ethical unacceptability of automatic physiological control, as well as computational complexity and strict regulatory requirements.
The article analyzes international and Russian projects, identifies specific risks, and proposes implementation principles. These include sufficient accuracy for a specific task rather than striving for absolute model completeness, mandatory physician participation in decision-making, and validation on real clinical data. It also argues that medical digital twins require their own methodology, and their success should be measured by clinical benefit rather than adherence to industrial standards.

As life expectancy increases, the number of people, including those of working age, with chronic noncommunicable diseases (CNCDs), increasing the workload of healthcare personnel, is growing. This increases the need for automated processing of medical data, particularly data from monitoring systems for key health indicators of patients with CNCDs. Key tasks in this area include accurately describing the patient's current health status and timely diagnosis. Addressing these challenges is crucial for the effective management of patients with CNCDs, supporting physicians in choosing the optimal treatment strategy. Fuzzy models offer significant potential for solving these problems due to the ambiguity of medical data, the ability to model the knowledge of medical professionals, and their low computational complexity. This article reviews, systematizes, and summarizes the results of 29 studies published between 2015 and 2025 on the descriptive and predictive analysis of numerical medical data from patients with CNDs using fuzzy models employing fuzzy sets and fuzzy logical inference. Particular attention is paid to assessing the accuracy of fuzzy models for various CNDs. The analysis of publications demonstrates the competitiveness and high efficiency of fuzzy models in data analysis, as evidenced by accuracy (from 90% to 99.61%) and sensitivity (from 80.94% to 98.57%) metrics, with the exception of cancer studies. The obtained results can serve as a basis for the development of medical decision support systems.

The aim of the study is to develop and evaluate the architecture of a neural network for automated analysis of electrocardiographic (ECG) signals to detect cardiac arrhythmias.
Materials and methods. Open ECG databases (34,570 records) were used for training. The study was conducted using data from six standard ECG leads (I, II, III, aVR, aVL, aVF). Signal preprocessing included polynomial trend removal, wavelet filtering, median filtering, smoothing, and normalization. A convolutional neural network, which processes signals from each of the six leads separately, was used for classification. The final decision is formed using a weighted voting method with empirically determined coefficients reflecting the contribution of each channel based on a preliminary analysis of metrics. A clinical annotation performed by cardiologists was used as a reference standard.
Results. The proposed model demonstrates high classification performance: accuracy – 0.97 (95% CI 0.96-0.98), precision – 0.98 (95% CI 0.97-0.99), recall – 0.98 (95% CI 0.97-0.99), specificity – 0.94 (95% CI 0.92-0.96), F1-score – 0.98 (95% CI 0.97-0.99), ROC-AUC – 0.99 (95% CI 0.98-1.00), PR-AUC – 0.96 (95% CI 0.94-0.97).
Conclusion. The obtained results confirm the effectiveness of the proposed method for diagnosing cardiovascular diseases. The proposed method can be adapted for diagnosing a wider range of cardiac diseases, making it relevant for implementation in practical cardiology.

Background. Spinal disorders are observed in a significant number of people of different ages. Improving the accuracy of diagnosing spinal disorders using medical artificial intelligence systems (MAIS) enables timely treatment initiation, preventing serious complications.
Research Objective. Rationale for the development and application of new AI services for the diagnosis of spinal diseases. Analysis of the diagnostic accuracy of AI services. Development of a physician-friendly AI system for the automatic analysis of spinal X-rays. Identification of conditions for increasing physician interest in AI services. Materials and Methods. To implement the study, a preliminary analysis of the accuracy of existing AI services for disease diagnosis and the potential for improving accuracy through the careful selection of neural network models (NNMs) was conducted. The process of creating a neural network-based software system for classifying and identifying pathologies based on spinal X-rays is described. The developed user interfaces for a web application and a Telegram bot providing quick access to diagnostic results using mobile devices are presented.
Results. Benchmarks for the accuracy of spinal disease diagnostics were determined. A dataset available on Kaggle. com was used for the first stage of developing and training a neural network for spinal disease diagnostics. The developed NNM was integrated into a web application and Telegram bot to provide automated diagnostic solutions. Potential conditions for increasing physician interest in AI services are demonstrated.
Conclusion. The developed MAIS prototype ensures the specified accuracy in diagnosing scoliosis and spondylosis using X-ray images from the test dataset and can be used to analyze labeled and prepared images provided by medical institutions. The results open up opportunities for further development and improvement of the developed system.

With the digitalization of healthcare, patients have gained access to their own medical records. However, poor clarity of medical texts often prevents them from interpreting it correctly. Large generative models (LGMs) have the potential to become a tool for adapting medical texts, but their use is currently fraught with risks. The aim of the study was to evaluate the safety of using LGMs to interpret radiology protocols for patients. Seven models performed a simplified text interpretation using eight computed tomography protocols as input. The generated interpretations were submitted to physicians and respondents without medical training for evaluation. The resulting scores were analyzed to determine the safety of implementing this technology and its feasibility.
All models generated text that met the main quality criteria. However, consistent ethical and safety violations were observed. A comparative analysis failed to identify a model that was superior across all criteria. The study also identified criteria for which assessments by physicians and respondents without medical training differed significantly.
It was demonstrated that, although large-scale generative models are formally successful in simplified interpretation of medical protocols, their direct application without a control system in clinical practice is extremely unsafe. The main problem is the distortion of the original information—the inclusion of additional recommendations, diagnoses, and prognoses, which contravenes patient communication standards. It was shown that despite the technology's potential within the field, safe implementation requires the preliminary development of a quality control system for large-scale generative models, a questionnaire that takes into account the competencies of both experts and non-experts, and clear threshold criteria. This work represents the first step toward creating such a system.

Identifying patients at high risk for fatal cardiovascular disease (CVD) complications is a critical task in reducing preventable CVD morbidity and mortality. Various risk assessment algorithms and scores are widely used for this purpose, but their limitations include a limited set of predictors and low accuracy. Machine learning methods offer the potential to address these shortcomings and personalize cardiovascular risk assessment.
Objective: To compare the accuracy of the SCORE scale and machine learning models in predicting fatal cardiovascular complications.
Materials and Methods: A multicenter retrospective study was conducted (1999–2018), including 3,891 treatment cases of 1,064 patients aged 40–69 years in the Russian Federation. Logistic regression, ensemble machine learning (ML) methods, and Multi-Layer Perceptron were used for forecasting. Comparison with SCORE was performed on an independent validation set consisting of 440 records.
Results: The CatBoost ML model demonstrated the best accuracy (AUROC 0.879; sensitivity 0.938; specificity 0.777). During validation, CatBoost demonstrated comparable discrimination to SCORE but outperformed the scale in specificity (0.653 vs. 0.408) and accuracy (0.673 vs. 0.45) when referencing patients to lowand intermediate-risk groups. Key predictors for the model were gender, age, smoking, systolic blood pressure, body mass index, heart rate, and lipid profile.
Conclusion: The machine learning model outperformed the SCORE scale in predicting fatal cardiovascular events. The use of machine learning in predicting cardiovascular risk can improve the effectiveness of CVD prevention and facilitate personalized patient care.