Dr Inna Skarga-Bandurova

The paper proposes an integrated approach to the automated diagnosis of cervical intraepithelial neoplasia (CIN) in epithelial patches extracted from digital histology images. Experiments were conducted to determine the most suitable deep learning model for the dataset and fuse patch predictions to decide the final CIN grade of the histology samples. Seven candidate CNN architectures were assessed in this study. Three fusion methods were applied to the best CNN classifier. The model ensemble combined CNN classifier and highest performing fusion method achieved an accuracy of 94.57%. This result shows significant improvement over the state-of-the-art classifiers for cervical cancer histopathology images. It is hoped that this work will contribute towards further research to automate the diagnosis of CIN from digital histopathology images.

The study proposes an integrated approach to automated cervical intraepithelial neoplasia (CIN) diagnosis in epithelial patches extracted from digital histology images. The model ensemble, combined CNN classifier, and
highest-performing fusion approach achieved an accuracy of 94.57%. This result demonstrates significant improvement over the state-of-the-art classifiers for cervical cancer histopathology images and promises further improvement in the automated diagnosis of CIN.

The paper presents the base techniques for data processing, knowledge management, and integration in a smartphone-based system designed for both short-term and long-term self-monitoring of the progression of Parkinson’s disease. The system is enriched with multimodal functionality and includes collecting data from internal measurement units in two modes. This enables to assess the performance of specific tasks and run specific non-obtrusive passive sensing tests based on routine activities. Information about physical activities and symptoms is processed and displayed in the form of a diary. The base components of knowledge and data acquisition are introduced. Some aspects of data analysis and data fusion are also discussed. Classification accuracy achieved with one data processing method for tremor data is 84%. Insofar as the MeCo system uses different internal measurement units for tremor assessment, the enhancement of the classification accuracy may be achieved by a combination of criteria from different techniques. The fusion technique in Parkinson's disease symptom assessment provides up to 10% more accurate results in comparison with a single score.

In this paper, we propose a procedure for real-time database incident response and forensic preparation via detecting changes in its performance parameters. More specifically, it includes a five-step procedure that starts from choosing parameters and then exploits a score-based version of the cumulative sum (CUSUM) to detect a change-point in the time series. This procedure works well when incidents clogging normal system operation and slowing down the overall functionality of the database, but for security purposes, it is possible to continue to power on systems and monitor them throughout the monitoring tools when the internal investigation makes absolute sense.

The paper presents a smartphone-based system for long-term self-monitoring patients with Parkinson’s disease. Particularly promising multimodal functionality includes collecting data from different internal measurement units in two modes, evaluating the performance of specific tasks and special non-obtrusive passive sensing features based on day-basis activities. Information about physical activities and symptoms is processed and displayed in the form of a diary. The general system architecture and functional architecture are introduced. We outlined the main design principles and provisions on data completeness and engagement of participants during the first two study months. Some aspects of data analysis are also discussed. The classification accuracy with one data processing method for tremor data is 89,7%. Insofar as MeCo system uses different tremor components, the enhancement of the classification accuracy may be achieved by combination of criteria from different techniques. The combination of criteria provides up to 10% more accurate results in comparison with a single analysis.

The Arctic regions of the world have in recent years experienced an increase in human activity not seen before in modern times. Receding polar ice and climate change have contributed to the opening of new sea routes, creating opportunities in intercontinental shipping and tourism. Increased accessibility has enabled the extraction of natural gas and oil, metals, and other resources. The cold climate provides natural cooling for data centers and other computational facilities. Economic activities are coupled with the expansion of military and civilian infrastructure, including for telecommunications, scientific installations, ports, and other intermodal transportation facilities. Information technology promotes efficiency and technologies such as fiber optic cables, satellite communications, radio, and others enable accessibility to these infrastructures from locations outside the Arctic. However, the reliance on information and communication technology and the connectedness of most critical infrastructures (electricity, communications, information, financial and government services, etc.) result in new vulnerabilities exposed by natural disasters or environmental accidents and which adversarial agents can exploit. Cyber security and resilience play a central role in ensuring the safety and security of communities in this age of interconnectedness and big data. Due to their often remote and extreme conditions, Arctic regions face unique challenges of cyber security and resilience for their critical infrastructure. This chapter summarizes discussions and lessons learned from a working group at a NATO Advanced Research Workshop on Governance for Cyber Security and Resilience in the Arctic as it pertains to critical infrastructure, held in Rovaniemi, Finland on 27-30 January, 2019. It aims to provide perspectives on cyber security in the context of Arctic infrastructure from multiple disciplines, including engineering and computer science, international relations, social sciences, law, and governance. Each perspective identifies challenges and opportunities in cyber security and resilience, in particular ones characteristic to Arctic regions. This includes documenting available theory and methods, including analogous methods from other fields, and describing data availabilities and needs. Lessons are derived from past and ongoing scenarios and incidents and methods for forecasting emerging and future scenarios are reviewed. Recommendations for research and practice to increase the cyber security and resilience of infrastructure are provided.

Technological advancements in smart assistive technology enable navigating and manipulating various types of computer-aided devices in the operating room through a contactless gesture interface. Understanding surgeon actions is crucial to natural human-robot interaction in operating room since it means a sort of prediction a human behavior so that the robot can foresee the surgeon's intention, early choose appropriate action and reduce waiting time. In this paper, we present a new deep network based on Convolution Long Short-Term Memory (ConvLSTM) for gesture prediction configured to provide natural interaction between the surgeon and assistive robot and improve operating-room efficiency. The experimental results prove the capability of reliably recognizing unfinished gestures on videos. We quantitatively demonstrate the latter ability and the fact that GestureConvLSTM improves the baseline system performance on LSA64 dataset.

The paper refers to the problem of decision making and choosing appropriate ways for decreasing the level of input information uncertainty related to absence or unavailability some values of mixed data sets. Approaches to addressing missing data and evaluating their performance are discussed. The generalized strategy to managing data with missing values is proposed. The study based on real pregnancy-related records of 186 patients from 12 to 42 weeks of gestation. Three missing data techniques: complete ignoring, case deletion, and random forest (RF) missing data imputation were applied to the medical data of various types, under a missing completely at random assumption for solving classification task and softening the negative impact of input information uncertainty. The efficiency of approaches to deal with missingness was evaluated. Results demonstrated that case deletion and ignoring missing values were the less suitable to handle mixed types of missing data and suggested RF imputation as a useful approach for imputing complex pregnancy-related data sets with missing data.

This paper presents a novel global path planning algorithm adapted to grid maps extending the Bug algorithm family. During experiments with Bresenham’s line algorithm, it was found that this algorithm allows natural obstacle avoidance. Based on this observed property, a recursive path planning algorithm was developed that operates on the grid maps represented by a masked array and solves potential looping problems using a state machine-based loop breaking mechanism. BRP was compared with existing methods from the Bug family of algorithms as well as with more classical algorithms such as A*, D*, and Dijkstra. The Piano Movers problem was also touched upon. The time performance of the BLA algorithm for different line thicknesses was compared. Based on these experiments, future optimisation of the BRP algorithm for motion planning tasks is possible. Through experiments, BRP was found to show significant improvement in path planning time for certain map cases compared to existing algorithms while acknowledging a slight increase in route length, approximately 0.43% above the optimal path. The presented work improves the field of global path planning by providing efficient and adaptable solutions that can be applied to specific applications. The algorithm is available at https://github.com/qr34t0r/BRP

The paper presents a comparative analysis of popular deep learning-based object detectors, focusing on evaluating their real-time performance on embedded platforms. The objective was to identify the optimal algorithm that can efficiently operate within limited computational resources and memory constraints of embedded systems while minimizing energy consumption. We tested a set of popular detectors, such as Faster R-CNN, YOLO, and SSD MobileNet and evaluated their performance on the testing platform with CUDA cores. A faster variant of bounding box suppression, blazing fast Non-Maximum Suppression (NMS), was employed to enhance efficiency. Additionally, the generalized intersection over union (GIoU) metric was adopted as the optimal testing method, allowing for more accurate object localization and detection. Results demonstrate that the SSD MobileNet-V2 and Tiny YOLO detectors exhibit promising real-time performance on embedded platforms, although their accuracy is diminished compared to larger models.

The understanding and prediction of surface properties in nanomaterials are essential for optimizing product performance. This research presents a computer modelling study focused on nanoscale surface treatment and the relationship between surface roughness and the thickness of a film applied during titanium nitrogen ion implantation into stainless steel. Optical microscopy was employed to analyse the surface, while the RIO software was utilized to simulate the implantation parameters. The study revealed a correlation between the height of the surface layer and the thickness of the applied film. The results demonstrated a regular pattern, supported by a high reliability value for the cubic equation and a correlation coefficient close to 1. This finding indicates that the technique can be effectively used to predict the surface roughness of samples before further processing, allowing for adjustments in the implantation modes based on intended material purposes.

The research focuses on developing a decision-making methodology that combines human intelligence (HI) and AI, emphasizing interpretability of AI results. The methodology formalizes decision trials using feature vectors and probability distributions for AI recommendations and HI proposals. The reliability of both AI and HI decisions is crucial for effective decision-making. Trust and interpretability in AI-generated clinical decisions are essential for successful implementation. An experiment involving image classification tasks was conducted, examining human attitudes, trust, and decision-making behaviour concerning AI recommendations. Three scenarios – HI-decision, AI-decision, and joint HI-AI decision – were evaluated. Expected Calibration Errors (ECEs) were below 10%, with AI exhibiting an ECE AI of 9.7% and human an ECE H of 6.2%. ECEs were used as uncertainty scores to optimize joint decision-making rule. The trust of humans in AI was evaluated, leading to improved HI accuracy. The final decision relied on the interpretability of AI results, resulting in a 6% improvement in initial HI accuracy.

Within the medical image analysis domain, the lack of extensive and well-balanced datasets has posed a significant challenge to traditional machine learning approaches, resulting in poor generalization ability of the models. In light of this, we propose a novel approach to evaluate the efficacy of neural network learning on small imbalanced datasets. The proposed methodology uncovers the relationships between model generalization ability, neural network properties, model complexity, and dataset resizing. This research highlights several key findings: (1) data augmentation techniques effectively enhance the generalization ability of neural network models; (2) a neural network model with a minimal number of each layer type can achieve superior generalization ability; (3) regularization layers prove to be a crucial factor in achieving higher generalization ability; (4) the number of epochs is not a determining factor in enhancing generalization ability; (5) complexity measures exhibit no significant correlation with generalization ability in the described scenarios. The findings from this study offer a practical roadmap for model selection, architecture search, and evaluation of the methods’ effectiveness in medical image analysis.

One of the priority tasks in constructing a quadrotor is fault-tolerant control and stabilizing its flight. This paper proposes a quadrotor stabilization system based on onboard sensors, an accelerometer, a gyroscope, a complementary filter that combines their readings, and a PID controller. The complementary filter is used to improve the accuracy of tilt angles relative to the ground, and the PID controller is used to calculate the deviation compensation and bring the quadrotor to the required position. The filter makes it possible to level the gyroscope zero drift and discrete integration errors. At the same time, the PID controller compensates for the deviation of the quadrotor from the horizontal position. Experimental validations show that our approach is able to accurately control the position of a quadrotor without any external sensors and achieve sufficiently high-quality flight stabilization with RMSE 2°.

The paper analyses cyber security challenges of smart cities with a particular focus on the intelligent integrated and interconnected electric vehicle (EV) charging infrastructure. The analysis indicates that not all innovative elements and smart city solutions have adequate cybersecurity protection. Digital technologies vary considerably in terms of the level of potential risks, with certain novel technologies — such as V2G, smart charging, and smart energy management — posing higher risks than others. It is intended to lay a foundation for securing EV charging infrastructure by analysing problem context and data to be protected, including attack surfaces and cybersecurity threats and vulnerabilities in the EV ecosystem, analysing standardisation for the EV connection to the charging infrastructure, and providing a set of recommendations and best practices to securing EV charging infrastructure.

The region of interest (RoI) identification has a significant potential for yielding information about relevant histological features and is imperative to improve the effectiveness of digital pathology in clinical practice. The typical RoI is the stratified squamous epithelium (SSE) that appears on relatively small image areas. Hence, taking the entire image for classification adds noise caused by irrelevant background, making classification networks biased towards the background fragments. This paper proposes a novel approach for epithelium RoI identification based on automatic bounding boxes (bb) construction and SSE extraction and compares it with state-of-the-art histology RoI localization and detection techniques. Further classification of the extracted epithelial fragments based on DenseNet made it possible to effectively identify the SSE RoI in cervical histology images (CHI). The design brings significant improvement to the identification of diagnostically significant regions. For this research, we created two CHI datasets, the CHI-I containing 171 colour images of the cervical histology microscopy and CHI-II containing 1049 extracted fragments of microscopy, which are the most considerable publicly available SSE datasets.

The current technological horizon in the Internet of Things, computer vision, deep learning and healthcare systems makes it possible to monitor some pre-existing conditions as acne vulgaris, automate the assessment of the acne severity by photo and monitor the skin health by specialists. Remote analysis of the skin condition and automatic image classification has several challenges. One of the most critical problems is imbalanced data raised because the number of clinical cases for each acne grade differs. This paper proposes a deep oversampling technique for 4-level acne classification that enables to deal with imbalanced datasets. The method was validated using several criteria. The experimental results obtained for imbalanced data sets revealed that the acne classification via proposed deep oversampling outperforms benchmark approaches.

The paper investigates the influence of the load factor of the shared memory on the efficiency of multicore systems. Typically, all cores serve threads of one program in parallel by the OpenMP programming technology or execute independent programs. There are no interactions between threads and independent programs, but conflicts can occur when accessing the shared memory. Models of program execution in one core and a multicore computer are developed, considering the probabilities of successful calls and service times of all levels of the shared memory subsystem. The load factor of the first level cache is determined through the ratio of the Ll cache load time to the total execution time of the program. The execution of various types of programs is simulated. A technique for the acceleration coefficient of a multicore computer based on the total load factor of the shared memory has been proposed. Based on this insight, we apply our model to determine the acceleration coefficient for 4-, 8-, 12- and 16-core systems for different combinations of system parameters.

The problem of constructing regression models is that it is necessary to specify the structure of the model, in addition, in models of large dimensions, poor conditioning of the matrices is possible, which leads to an unstable solution. The paper considers methods of self-organization (methods of group accounting of arguments), which use an iterative procedure for the simultaneous synthesis of the structure of the model and the calculation of its coefficients. The advantages and disadvantages of the known methods of self-organization are analyzed. A self-organization method for the synthesis of a multiplicative model with linear private descriptions has been developed. The effectiveness of the method has been tested on test cases.

This work is devoted to evaluating the effectiveness of the electrocardiological study process without using and using cardiological decision support systems. To assess the effectiveness, analytical expressions of the probabilistic-time characteristics of the developed structural model of the electrocardiological study process are used. An analysis of the time characteristics of the model is performed when different initial conditions are set for three different types of electrocardiological studies: the study is conducted for the first time, the study is repeated as a result of screening, the study is repeated after treatment. The work shows that the use of cardiological decision support systems based on the developed methods for analyzing biomedical signals with locally concentrated features reduced the average time required for the electrocardiological study of each of the considered types.

Detection and avoiding obstacles in real time is an important design requirement for any autonomous mobile platforms. In this paper we propose a new algorithm for real time obstacle avoidance. To improve the localization of a small autonomous mobile robot equipped with a microcontroller and four ultrasonic sensors, a linear recursive Kalman filter was used. The application of the Kalman filter enables the robot to avoid additional obstacles when transferring data from the sensor, adaptively adjust the intensity of noise, which ensures a smoother movement of the robot. Simulation results show that the developed approach enables to obtain the adequate realtime localization and demonstrate the efficiency in obstacle avoidance with resulting RMSE of 4.15 s and 0.07 m.

Recently, smart grids have become a vector of the energy policy of many countries. Due to structural and operation features, smart grids are a constant target of combined and simultaneous cyberattacks. To maximize security and to optimize existing network schemes to prevent cyber intrusion, in this paper, we propose an approach to decision support in finding and identifying the most potent attack combinations that can set the system to maximum damage. The main purpose is to identify the most severe combinations of attacks on smart grid components that potentially can be implemented from the perspective of the attacker. In this context, the problem of finding weaknesses points in the network configuration of a smart grid and assessing the impact of events on cyberinfrastructure is considered. The technique for detecting and investigating the strongest combinations of cyberattacks on the smart grid network is given with an example of the analysis of the spread of pandemic software in a system with arbitrary structure.

The paper outlines some aspects of developing a cyber-forensic framework for Smart Grid cyber-crime investigations. In this research, we examine a key forensic instrument in reconstructing events, the timeline, followed by correlation of data from different sources. Then, we deal with the tasks of collecting and storing the monitored data. The paper also covers some aspects of the legal ramifications from collecting this data and touches on the preconditions that must be met to enable network forensics. Then we present the logging architecture, based on the recommendations of the UK National Cyber Security Center. The final part presents the methodological framework that is the result of applying the OSCAR methodology and relevant open source tools in order to ensure that necessary forensic information can be collected, stored and used as legal evidence in court.

The framework based on a convolutional neural network (CNN) with adding long short-term memory layers (LSTM) for recognizing hand gestures from a video stream in real-time is presented. A review and analysis of existing models relating to gesture recognition in deep learning are considered. The task was to perform hand gesture classification using deep convolutional neural networks and obtain a simple, precise and resource-efficient system for visual recognition of letters and digits in sign language. The model is stable to rather wide angles of hand rotation and independent of lighting due to the using of contour patterns. In experiments with CNN, 98.46% accuracy on the test dataset has been obtained.

In this paper, we present an overview of the problems associated with the cross-site scripting (XSS) in the graphical content of web applications. The brief analysis of vulnerabilities for graphical files and factors responsible for making SVG images vulnerable to XSS attacks are discussed. XML treatment methods and their practical testing are performed. As a result, the set of rules for protecting the graphic content of the websites and prevent XSS vulnerabilities are proposed.

Database (DB) stability is a core component of all IT services. Losses from poor performance and unexpected database failures cost millions for companies and take time to recover. In this paper, we introduce a new formalism for finding insight in DB monitoring data, such as performance, query complexity, and events history to increase the ability of the database administrator to respond on different performance issues. A data fusion technique to predict the probability of a critical state of the DB system is proposed. A case study for sets of time series data before DB shutdown is presented. Testing results conclude that the proposed methodology gives better understanding a combination of data, specific behavior of parameters and their contribution to the current problem.

Real-time operation systems (RTOS) have become very important to the development of autonomous mobile robots. The choice of RTOS has tremendous sway with processor utilization, response time, and real-time jitter. In this paper, we present experimental trials and analyze the feasibility of RTOS on a single-board computer for image recognition and vision-based navigation of small autonomous robot. Several real-time (RT) patches Linux frequently used not only in robotics are implemented and tested on the Raspberry Pi2 equipped with a native camera board. To study the speed of image recognition (classification) OpenCV library was used. Test results show that the RT Patch Linux can produce higher throughput compared to Xenomai, but it can be seen that RT systems almost did not affect the speed of static images recognition systems almost did not affect the speed of image recognition.

Smart Grids is an emerging technology promising significant changes in the economy and the social sphere. One among many challenges in their development and distribution is security. Considering recent hackers attacks on energy grids and taking into account the distributed structure of these systems the use of traditional means of computer protection and the search for a crime figure becomes more difficult or impossible. In this article, we introduce some application areas of smart grid forensic science, discuss the opportunities, and outline the open issues in the topic. We summarized challenges for forensic in Smart Grids in connection with a Blockchain and proposed a decentralized transaction platform based on Blockchain tailored to the energy sector with all the latest technology such as advanced metering infrastructure, distributed generation, etc.

In this paper, we present an overview of the problems associated with the analysis and security of medical data and offer a solution that will provide the basis for improving the quality of medical services. We propose the architecture of a decentralized health data ecosystem based on a blockchain that will allow us to operate with vast volumes of clinical data, while also protecting confidential medical data. An example of a blockchain solution based on Exonum framework for state-scale use in healthcare is discussed. The deployments of such systems will the benefit to medical data safety, extend the base of clinical data collections, and create an effective shared health infrastructure.

In this work, we take aim towards increasing the effectiveness of surgical assistant robots. We intended to make assistant robots safer by making them aware about the actions of surgeon, so it can take appropriate assisting actions. In other words, we aim to solve the problem of surgeon action detection in endoscopic videos. To this, we introduce a challenging dataset for surgeon action detection in real-world endoscopic videos. Action classes are picked based on the feedback of surgeons and annotated by medical professional. Given a video frame, we draw bounding box around surgical tool which is performing action and label it with action label. Finally, we presenta frame-level action detection baseline model based on recent advances in ob-ject detection. Results on our new dataset show that our presented dataset provides enough interesting challenges for future method and it can serveas strong benchmark corresponding research in surgeon action detection in endoscopic videos.

Nowadays perspective of the creation of autonomous cognitive agents opens the way to the enriching of human-computer interaction by means of the building of intelligent assistants and automation of cognitive tasks. We suppose that the autonomous behavior of cognitive agents may be the consequence of a causal gap between physical processes and self-referential meaningful processing of information, which is related but not determined by physical processes. Cognitive representations as the agent’s subjective estimations are a base for indeterminism and unpredictability of an agent’s behavior. We propose to consider conceptual spaces as these cognitive representations. Conceptual spaces enable qualitative modeling of the quantitative subjective estimation. We suppose that the agent’s behavior is based on an agent’s subjective estimations in the conceptual space that are changed by means of perceptions and reasoning. Human activities affect the social (external) conceptual space. The latter can be influential on the agent’s (internal) conceptual space. These cognitive representations must be constantly reevaluated on the basis of the new data to defend against deception. This is the way to the understanding of human’s ability to make broad judgments and the possibility to understand the essence of things. To predict and evaluate the influences on cognitive spaces, we propose cognitive capital as an assessment of factors that can influence agent’s (internal) and social (external) senses, meanings, values, preferences, and as a result, utilities. The bidirectional transformation of these conceptual spaces could be a quantitative indicator of cognitive capital variations. This influence on the conceptual spaces possible due to digital, cognitive, and neuromarketing technologies as instruments to create, invest in, and manage cognitive capital assets. Considering that market interactions are based on the utilities, cognitive capital could be a measure of evaluating and development of a market structure. This relates to fighting against fake news as a component of the “marketing of statehood”. The relationship of the cost of creating and delivering impact (information and media) to outputs (benefits) can be considered as the indicator of the effectiveness of influencing on the conceptual spaces. Not ideologies or positions are fighting now, but virtual images that often drive users into information traps. For example, wide-spreading fake news in mass media (especially social networks) and political campaigns can be attributed to their (prior predicted) high effectiveness. These calculations of effectiveness should be taken for a decision about investments in the physical, human, social, or cognitive capital. This decision can be made on the basis of marginal productivity theory when the most effective investment strategies are based on the marginal product of every type of capital.