Natalia Dobrenko
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Natalia Dobrenko has a total of 39 co-authors. Here are the top co-authors by number of shared publications:
- Natalia Gusarova
ITMO University
9 shared publications - Aleksandra Vatian
School of Translational Information Technologies, ITMO University, 197101 St. Petersburg, Russia
8 shared publications - Anatoly Shalyto
ITMO University
4 shared publications - Ivan Tomilov
ITMO University
4 shared publications - Alexey Zubanenko
ITMO UniversityITMO University
2 shared publications - Anna A. Tatarinova
Национальный медицинский исследовательский центр им. В.А. Алмазова 197341 Санкт-Петербург Российская Федерация
2 shared publications
Latest Publications
- Open Access
- 25 October 2026
- Published by ITMO University in Scientific and Technical Journal of Information Technologies, Mechanics and Optics
- Vol. 26 (5) , 1000-1008
Training visual language models for MRI-based diagnostics
In modern radiology, diagnostic Artificial Intelligence assistants for classifying Magnetic Resonance Imaging (MRI) brain images are typically based on Convolutional Neural Networks (CNNs), which work exclusively with the visual signal and are unable to use structured diagnostic descriptions of pathologies. Vision Language Models (VLMs) open up new possibilities: joint processing of visual and textual data allows for the consideration of the semantic context of pathologies, which is inaccessible to CNN classifiers. The key obstacle to their application remains the high resource intensity of retraining, which is incomparable with the capabilities of a typical clinic. This paper proposes a method for constructing a resource-efficient diagnostic AI assistant based on VLM Gemma3 (4b-it) using low-rank adaptation (LoRA). Retraining was performed on a dataset of 91,777 Fluid-Attenuated Inversion Recovery MRI brain images, distributed across eight pathology classes. Only low-rank adapters were trained — 12.5 % of the model parameters, which allowed training on a single NVIDIA RTX3090 Graphics Processing Unit with 24 GB of Video Random Access Memory without masked data annotation. The radiologist’s expert knowledge was integrated via a system prompt, which assigns the model the role of an experienced neuroradiologist with explicit analysis criteria. On the validation set (9178 examples), the model achieved: Accuracy 0.930, F1-score 0.929, Specificity 0.989, and Cohen’s Kappa 0.915. Comparison with the baseline ConvNeXt-Base classifier on an identical dataset showed the superiority of VLM+LoRA across all metrics: Accuracy increased by 7.4 percentage points, and Cohen’s Kappa by 0.090. The obtained results confirm the practical applicability of the proposed approach for clinics with limited computing resources.
- 25 October 2026
- Published by ITMO University in Scientific and Technical Journal of Information Technologies, Mechanics and Optics
- Vol. 26 (5) , 1000-1008
Open AccessTraining visual language models for MRI-based diagnostics
In modern radiology, diagnostic Artificial Intelligence assistants for classifying Magnetic Resonance Imaging (MRI) brain images are typically based on Convolutional Neural Networks (CNNs), which work exclusively with the visual signal and are unable to use structured diagnostic descriptions of pathologies. Vision Language Models (VLMs) open up new possibilities: joint processing of visual and textual data allows for the consideration of the semantic context of pathologies, which is inaccessible to CNN classifiers. The key obstacle to their application remains the high resource intensity of retraining, which is incomparable with the capabilities of a typical clinic. This paper proposes a method for constructing a resource-efficient diagnostic AI assistant based on VLM Gemma3 (4b-it) using low-rank adaptation (LoRA). Retraining was performed on a dataset of 91,777 Fluid-Attenuated Inversion Recovery MRI brain images, distributed across eight pathology classes. Only low-rank adapters were trained — 12.5 % of the model parameters, which allowed training on a single NVIDIA RTX3090 Graphics Processing Unit with 24 GB of Video Random Access Memory without masked data annotation. The radiologist’s expert knowledge was integrated via a system prompt, which assigns the model the role of an experienced neuroradiologist with explicit analysis criteria. On the validation set (9178 examples), the model achieved: Accuracy 0.930, F1-score 0.929, Specificity 0.989, and Cohen’s Kappa 0.915. Comparison with the baseline ConvNeXt-Base classifier on an identical dataset showed the superiority of VLM+LoRA across all metrics: Accuracy increased by 7.4 percentage points, and Cohen’s Kappa by 0.090. The obtained results confirm the practical applicability of the proposed approach for clinics with limited computing resources.
- 6 July 2025
- Published by Springer Nature
- p. 195-203
Enhancing Medical Image Analysis with Multi-Task Learning Using Visual Transformers
- 6 July 2025
- Published by Springer Nature
- p. 195-203
Enhancing Medical Image Analysis with Multi-Task Learning Using Visual Transformers
- Open Access
- 1 June 2023
- Published by ITMO University in Scientific and Technical Journal of Information Technologies, Mechanics and Optics
- Vol. 23 (3) , 595
Intelligent clinical decision support for small patient datasets
The ways of substantiating the clinical decision of doctors in the absence of clinical treatment protocols are considered. A comparative evaluation of various statistical methods for ranking clinical symptoms in terms of significance for predicting the outcome of the disease in a small sample of patients with COVID-19 and a history of cardiovascular diseases was performed. The data set (141 patients, 81 factors) was formed based on the materials of electronic medical records of patients of the Federal State Budgetary Institution “National Medical Research Center named after V.A. Almazov”. A subset of controllable risk factors (51 factors) was identified. Descriptive statistics methods (one-way ANOVA, Mann-Whitney and χ² tests) and dimensionality reduction methods (univariate linear regression combined with multiple logistic regression, generalized discriminant analysis, and various decision tree algorithms) were used to rank the factors. To compare the ranking results and evaluate the statistical stability, Kendall’s correlation was used, visualized as a heat map and a positional graph. It has been established that the use of descriptive statistics methods is justified when ranking on a small sample size of patients. It is shown that the ensemble of ranking results may be statistically inconsistent. It is concluded that the positions of the same features obtained by ranking them as part of a complete set and a subset of features do not match; therefore, when choosing a statistical processing method for expert evaluation, one should take into account the meaningful formulation of the problem. It is shown that the statistical stability of ranking under conditions of small samples depends on the number of features taken into account, and this dependence is significantly different for different ranking methods. The proposed method of intellectual support and verification of clinical decisions in terms of choosing the most significant clinical signs can be used to select and justify the tactics of managing patients in the absence of clinical protocols.
- 1 June 2023
- Published by ITMO University in Scientific and Technical Journal of Information Technologies, Mechanics and Optics
- Vol. 23 (3) , 595
Open AccessIntelligent clinical decision support for small patient datasets
The ways of substantiating the clinical decision of doctors in the absence of clinical treatment protocols are considered. A comparative evaluation of various statistical methods for ranking clinical symptoms in terms of significance for predicting the outcome of the disease in a small sample of patients with COVID-19 and a history of cardiovascular diseases was performed. The data set (141 patients, 81 factors) was formed based on the materials of electronic medical records of patients of the Federal State Budgetary Institution “National Medical Research Center named after V.A. Almazov”. A subset of controllable risk factors (51 factors) was identified. Descriptive statistics methods (one-way ANOVA, Mann-Whitney and χ² tests) and dimensionality reduction methods (univariate linear regression combined with multiple logistic regression, generalized discriminant analysis, and various decision tree algorithms) were used to rank the factors. To compare the ranking results and evaluate the statistical stability, Kendall’s correlation was used, visualized as a heat map and a positional graph. It has been established that the use of descriptive statistics methods is justified when ranking on a small sample size of patients. It is shown that the ensemble of ranking results may be statistically inconsistent. It is concluded that the positions of the same features obtained by ranking them as part of a complete set and a subset of features do not match; therefore, when choosing a statistical processing method for expert evaluation, one should take into account the meaningful formulation of the problem. It is shown that the statistical stability of ranking under conditions of small samples depends on the number of features taken into account, and this dependence is significantly different for different ranking methods. The proposed method of intellectual support and verification of clinical decisions in terms of choosing the most significant clinical signs can be used to select and justify the tactics of managing patients in the absence of clinical protocols.
- Open Access
- 1 December 2022
- Published by ITMO University in Scientific and Technical Journal of Information Technologies, Mechanics and Optics
- Vol. 22 (6) , 1166
Automated evaluation of ECG parameters during the COVID-19 pandemic
Algorithms for prompt automated evaluation of electrocardiogram parameters in the absence of specialized equipment and specialized specialists are considered. The patient’s electrocardiogram is recorded on a paper tape, then it is photographed on the primary care doctor’s mobile phone and processed by a specialized application. The application digitizes the photographed image of the electrocardiogram, evaluates its main parameters as well as calculates criteria for the differential diagnosis of certain diseases using approximate formulas. In addition, the digitized electrocardiogram image is transmitted to the server and processed using a machine learning system. Algorithms for digitizing and analyzing an electrocardiogram have been developed that make it possible to evaluate its elements that are important for diagnosis, and the average error in determining the position of the most complex (smoothed) peaks — P and T waves — was no more than 0.1 mm. An algorithm for the criteria analysis of an electrocardiogram is proposed to support the differential diagnosis of acute myocardial infarction with ST segment elevation and early ventricular repolarization syndrome which provides accuracy values of 0.85 and F-scores of 0.74. An alternative algorithm based on a deep neural network is proposed which provides the best values — 0.96 and 0.88, respectively, but requires large computing resources and is executed on the server. The algorithms are implemented as a set of library functions. They can be used both independently and as part of a full-scale clinical decision support system for automated evaluation of electrocardiogram parameters based on a client-server architecture. In addition, all calculation results, together with a photograph of the original electrocardiogram, can be promptly transferred to a qualified cardiologist in order to receive an advisory opinion remotely.
- 1 December 2022
- Published by ITMO University in Scientific and Technical Journal of Information Technologies, Mechanics and Optics
- Vol. 22 (6) , 1166
Open AccessAutomated evaluation of ECG parameters during the COVID-19 pandemic
Algorithms for prompt automated evaluation of electrocardiogram parameters in the absence of specialized equipment and specialized specialists are considered. The patient’s electrocardiogram is recorded on a paper tape, then it is photographed on the primary care doctor’s mobile phone and processed by a specialized application. The application digitizes the photographed image of the electrocardiogram, evaluates its main parameters as well as calculates criteria for the differential diagnosis of certain diseases using approximate formulas. In addition, the digitized electrocardiogram image is transmitted to the server and processed using a machine learning system. Algorithms for digitizing and analyzing an electrocardiogram have been developed that make it possible to evaluate its elements that are important for diagnosis, and the average error in determining the position of the most complex (smoothed) peaks — P and T waves — was no more than 0.1 mm. An algorithm for the criteria analysis of an electrocardiogram is proposed to support the differential diagnosis of acute myocardial infarction with ST segment elevation and early ventricular repolarization syndrome which provides accuracy values of 0.85 and F-scores of 0.74. An alternative algorithm based on a deep neural network is proposed which provides the best values — 0.96 and 0.88, respectively, but requires large computing resources and is executed on the server. The algorithms are implemented as a set of library functions. They can be used both independently and as part of a full-scale clinical decision support system for automated evaluation of electrocardiogram parameters based on a client-server architecture. In addition, all calculation results, together with a photograph of the original electrocardiogram, can be promptly transferred to a qualified cardiologist in order to receive an advisory opinion remotely.
- Open Access
- 1 January 2021
- Published by INSTICC
- p. 163-170
De-identification of Medical Information for Forming Multimodal Datasets to Train Neural Networks
- 1 January 2021
- Published by INSTICC
- p. 163-170
Open AccessDe-identification of Medical Information for Forming Multimodal Datasets to Train Neural Networks
- Open Access
- 17 October 2019
- Published by State University of Aerospace Instrumentation (SUAI) in Information and Control Systems
- No. 5,p. 64-75
Automata approach for personalized support of clinical processes in healthcare
Introduction: Within the framework of the National Healthcare Project, personalization of a physician’s activity is very important,forming a demand for a Clinical Decision Support System. The available systems miss the functions of prompting a doctor during theclinical process or identifying possible contradictions between different types of medical treatment offered to the patient. Purpose:Development of a solution, free from the above-mentioned problems, for personalized support of the clinical process. Methods:Automata (state machine) approach presenting the clinical process as a set of automata states and possible transitions between them,and a set of design patterns, namely: Abstract Factory, Facade, Adapter and Visitor. Results: A solution for personalized supportof clinical processes is proposed, based on the automata approach and design patterns. The automata approach allows you to dividethe clinical process into separate stages and automatically control the possible transitions and conditions for their implementation,including checking for c ontraindications. The use of design patterns provides a sufficient degree of generalization, allowing you,without affecting the structure of the main application code, to promptly connect the system to the necessary sources of information,and to enter the data about contradictions of various origins, taking them into account when making decisions on the treatment of aparticular patient. Practical relevance: The developed solution, as compared to the available systems, is more efficient at promptingthe doctor during a clinical process, and at identifying possible contradictions between the various types of medical treatment offeredto the patient.
- 17 October 2019
- Published by State University of Aerospace Instrumentation (SUAI) in Information and Control Systems
- No. 5,p. 64-75
Open AccessAutomata approach for personalized support of clinical processes in healthcare
Introduction: Within the framework of the National Healthcare Project, personalization of a physician’s activity is very important,forming a demand for a Clinical Decision Support System. The available systems miss the functions of prompting a doctor during theclinical process or identifying possible contradictions between different types of medical treatment offered to the patient. Purpose:Development of a solution, free from the above-mentioned problems, for personalized support of the clinical process. Methods:Automata (state machine) approach presenting the clinical process as a set of automata states and possible transitions between them,and a set of design patterns, namely: Abstract Factory, Facade, Adapter and Visitor. Results: A solution for personalized supportof clinical processes is proposed, based on the automata approach and design patterns. The automata approach allows you to dividethe clinical process into separate stages and automatically control the possible transitions and conditions for their implementation,including checking for c ontraindications. The use of design patterns provides a sufficient degree of generalization, allowing you,without affecting the structure of the main application code, to promptly connect the system to the necessary sources of information,and to enter the data about contradictions of various origins, taking them into account when making decisions on the treatment of aparticular patient. Practical relevance: The developed solution, as compared to the available systems, is more efficient at promptingthe doctor during a clinical process, and at identifying possible contradictions between the various types of medical treatment offeredto the patient.
- 10 February 2019
- Published by Springer Nature
- p. 138-153
Intellectualization of Knowledge Acquisition of Academic Texts as an Answer to Challenges of Modern Information Society
- 10 February 2019
- Published by Springer Nature
- p. 138-153
Intellectualization of Knowledge Acquisition of Academic Texts as an Answer to Challenges of Modern Information Society
- 9 November 2018
- Published by Springer Nature
- p. 195-203
Specifics Analysis of Medical Communities in Social Network Services
- 9 November 2018
- Published by Springer Nature
- p. 195-203
Specifics Analysis of Medical Communities in Social Network Services
- 13 October 2016
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-9
An ontology approach to storing educational information
The modern educational process is characterized by the fact that the competencies and the content of education quickly change and dynamically vary depending on the needs of the specific groups of students. Especially strongly it is shown in the professional education, for example in information technologies. The teacher should have an opportunity to quickly and effectively arrange the content of education for the parameters of a specific course (for instance: type of group, periods of training, etc.) In these conditions, the traditional method of storing the information, such as the relational data model, seems difficult and inconvenient to manage with the permanently and dynamically changing structure of data. In this regard, the use of network models of data structure is required. In this paper, we present an ontological model for an information system for storning educational information. The authors propose the use of a wiki-engine for building an information system- and the SCORM-based organization of the training material. The experimental results have shown that the ontological approach for storing the educational information tends to cover the needs of the teachers in universities.
- 13 October 2016
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-9
An ontology approach to storing educational information
The modern educational process is characterized by the fact that the competencies and the content of education quickly change and dynamically vary depending on the needs of the specific groups of students. Especially strongly it is shown in the professional education, for example in information technologies. The teacher should have an opportunity to quickly and effectively arrange the content of education for the parameters of a specific course (for instance: type of group, periods of training, etc.) In these conditions, the traditional method of storing the information, such as the relational data model, seems difficult and inconvenient to manage with the permanently and dynamically changing structure of data. In this regard, the use of network models of data structure is required. In this paper, we present an ontological model for an information system for storning educational information. The authors propose the use of a wiki-engine for building an information system- and the SCORM-based organization of the training material. The experimental results have shown that the ontological approach for storing the educational information tends to cover the needs of the teachers in universities.