Abstract: This paper describes basics of the Research Information System (RIS) format and Current Research Information Systems (CRIS) and their data models. The result of this research is the mapping scheme of data from RIS format to the Common European Research Information Format (CERIF) standard.

Abstract: This paper will present a new software system that enables automated assessment of the breast cancer risk from the mammographic images. The aim of the system is to provide for a quick and efficient detection of all potentially malignant abnormalities from the mammograms and determination of preliminary malignancy risk for each. Such preliminary risk assessment could, further, be used either to terminate the procedure (small risk), or as a priority by which the patient will be scheduled for following procedural steps. The system extracts the data necessary for risk assessment from the mammogram. For that purpose methods for digital image processing and analysis are used. Interpretation of the extracted data and the final risk assessment require a method capable to deal with imprecise and incomplete knowledge because no exact mathematical model for estimation of potentially malignant abnormalities risk exists. Therefore, risk assessment is implemented by means of a fuzzy controller capable to imitate the work of a human expert / radiologist. The paper presents architecture of the system for automated assessment of the breast cancer risk from mammograms, some specific details of its implementation, and discussion of the system’s accuracy.

Abstract: Evaluating a student’s programming knowledge is not always precise or objective. Students may give a correct answer to a question without understanding the underlying code. One approach to solving this problem is to track the student’s gaze during the examination by using an eye tracking device. Data acquired from the eye tracker can show if students are actually looking at the code they are explaining by providing the x and y coordinates of their gaze. These coordinates can then be mapped to the code displayed on the screen, allowing the comparison between the code the students were looking at and the code which they were expected to look at, thereby confirming or refuting their knowledge. This paper presents a solution to the aforementioned problem. The solution includes establishing communication with the Gazepoint eye tracking device, using a separately created Java library. Coordinates received from the device are transformed using the iTrace Eclipse plugin, modified to support tracking gazes across multiple files within a project in a single Eclipse session. The transformed data is then subjected to additional processing and filtering, during which redundant data is eliminated, such as all gazes that were flagged as non-code gazes, as well as attributes which were deemed irrelevant to the current project, such as the raw coordinates of the gaze, or pupil diameters. The median filter was also applied to the data, to reduce noise caused partially by the hardware itself, and partially by chaotic eye movements. The final output data is a sequence of gazes, each consisting of the line and column of the code that was watched, the name of the file containing the code, as well as the duration of the gaze. The output is then written to a JSON file that can be used later in the process of evaluating students’ knowledge.

Abstract: Service Oriented Architecture (SOA) is one of the most common styles used to design and develop contemporary software intensive systems. Microservices represent the latest trend to build SOA applications, offering interoperability, scalability, component sharing, improved fault isolation and etc. Although an emerging and still immature concept, it gains big impact in software development industry, as many companies try to migrate their projects to such architecture. Microservices lead to construction of highly flexible scalable and dynamic software systems. Such dynamism, however may provoke dynamic reallocation of running microservice instances which results into a change of their addresses. This raises some problems with discovery of microservices. This paper presents a platform for discovery of microservices, which tackles this problem.

Abstract: Cloud providers offers plenty of choice to their potential customers, spanning all standard types of services IaaS, PaaS, SaaS, FaaS, BPaaS... Sometimes it is quite hard to properly determine someone’s exact needs and evaluate the total cost of cloud-based solution, and even harder to compare if that cost, in a long run, is comparable to the cost of the on-premise solution. If we concentrate at the virtual machine market, all major cloud providers (Amazon WS, MS Azure, Google, IBM…) offer on demand instances of different computational characteristics, dedicated instances for demanding customers, but entry level configurations that comes with some strong limitations, are also available. Their price may be very attractive to small and medium enterprises, one that are usually struggling with internal IT support, and the one that may be best served by managed service offered by the cloud providers. Those entry level instances usually represent virtual machines created and run on spare capacity of cloud computing hardware that currently is not used for on demand services. Since the lifespan of these instances is pretty unpredictable, they are usually best utilized for applications that do the batch processing and can easily withstand loss of computing capacity. This paper discuss the possibility of using these entry level instances even for a bit more demanding applications, concentrating on Amazon Web Services spot instances.

Abstract: The referential integrity constraint (RIC) type is one of the constraint types of relational data model. It is defined between at most two, not necessarily different, relation schemes. In the course of the third normal form database design, there may be necessary to define the referential integrity constraint between more than two relation schemes. Therefore, an extended referential integrity constraint (ERIC) type is defined. A constraint of this type spans more than two relation schemes. ERIC is defined in the relational data model and its implementation in relational database management systems can be done procedurally, by means of database triggers, which is presented in this paper. Unlike relational data model, XML data model does not recognize ERIC type. In this paper we specify the extended referential integrity constraint type in XML data model and propose techniques for the implementation of ERICs in XML database management systems.

Abstract: In this paper, one of the aspects of the smart grids is analyzed. This aspect includes the utilization of the large amount of available digital information for creating smart models for planning and forecasting. The latest and new achievements in the field of machine learning are used for that purpose. Specifically, models based on deep belief networks are developed within this paper and it is examined whether these models may be applied for electricity price forecasting. For that purpose, the hourly data of the prices of the power exchanges in the region of Southeast Europe are used. The obtained results present the advantages of the developed models based on deep belief networks, compared to the traditional neural networks, when applied to electricity price forecasting. To this end, the mean average percent error of the deep belief network model is less than the minimum error of the traditional neural network model in each of the analyzed datasets.

Abstract: The integration of Internet of Things and Cyber Physical Systems principles and technologies in Enterprise Systems creates a new complex system category, with increased capability to process and manage information and knowledge. Modern Enterprise Systems offer innovative services that are developed in relation to core enterprise principles. This paper is focused on the investigation of using recently developed techniques in the area of process mining and work-flow identification, that contribute to the evolution of Sensing Systems towards Sensing Enterprise Systems. The processing steps related to document flow discovery in a mixed human – device - cyber enterprise environment are analyzed and an automated process mining solution is proposed and a case study is discussed.

Abstract: In this study novel machine learning based neural and neuro-fuzzy prognostic models for leukemia clinical outcome prediction have been developed, based on clinical and morphometric diagnostic data. Motivation was to enable better prediction of complete cytogenetic response (CCgR) for patients with chronic myeloid leukemia, compared to traditional and well-established scoring systems. Computational intelligence and machine learning have been applied to a wide range of problems to assist in decision-making, especially artificial neural networks, fuzzy systems and powerful hybrid neuro-fuzzy approaches have already proven their strong potentials in medicine. Despite that, applications in hematology are still scarce. This prospective study included a consecutive series of patients with chronic myeloid leukemia (CML) who were started on imatinib therapy. Analysis was performed with CCgR at different time intervals as the outcome variables. Machine learning based computationally intelligent neural and neuro-fuzzy models that were developed included EUTOS score on diagnosis and one of the angiogenesis morphometric parameters. The major finding of this study is that machine learning models using the morphometric parameters, available at diagnosis of chronic phase of the CML, may improve prediction of CCgR for patients on imatinib drug therapy, in comparison particularly to the EUTOS score being the standard prognostic scoring system and regression models using the same inputs.

Abstract: The automatic processes operated by machine learning algorithms that use information extracted from the images require input of data that can be used as parameters for object classification. The preprocessing step for image segmentation is often a non-trivial task, especially for outdoor environments, where acquisition conditions depend on environmental factors whose characteristics cannot be controlled. The paper addresses all necessary steps to segmentation and classification of vehicles and pedestrians to a traffic control system in order to capture vehicles’ infractions during pedestrian crossing the streets. A set of images from potential collision situation were detected from real traffic conditions. The text presents the performance evaluation of machine learning based algorithm combined with feature extraction methods applied in the identification of collision situations.

Abstract:

Abstract: This paper presents a proposal for calculating and visualizing flood prediction on the basis of rainfall prediction within given area in surface. Rainfall prediction and visualization is performed in multiple subsequent steps. As a basic, this approach uses quantified ground-level precipitation from the measured attenuation of microwave signals of cellular networks. Quantified precipitation along signal lines is used as an input into GIS module which performs spatial interpolation of precipitation for the area intersected by microwave signals. Interpolated values are used to calculate the estimated total amount of rainfall. Further, in combination with DEM model for the selected area, estimated rainfall is used to perform a flood prediction and visualization within the same GIS module.

Abstract: Manufacturing enterprises are presently faced with an array of industry 4.0 (I4.0) challenges. “Digital requirements” need to be really assessed by an accurate analysis and deep understanding of the operational and technological criticalities in the manufacturing operations. This may thus allow to effectively design and implement a digital twin that supports the transformation. The goal of the article is to present and analyze the criteria adopted in a design for the transformation of a real setting of an Italian company into a I4.0 testing learning-laboratory. The idea behind is that this laboratory-factory may play a key role in developing new solutions for the automation and the implementation of the I4.0 paradigm and it is a viable solution for the learning factory.

Abstract: In large cities, district heating system is a network of pipelines where heating is delivered to a number of customers from a centralized heating station, central plant. Ideal operational strategy is to deliver enough heat in order to satisfy heating demands of each customer with a minimum cost. Nevertheless, nowadays many district heating systems in Serbia are operated largely based on operational experience, without any available tools for predicting future states or demands. Due to the interactive, adaptive, and interconnected nature of the district heating systems, their components and outside environment, they must be capable of negotiating constantly changing scenarios. Being among large energy consumers, district heating systems, could be optimized with help of deep learning methodologies and extend capabilities of artificial neural networks (ANNs), such as convolutional neural networks (CNN) and long short-term memory (LSTM). In this paper control system of a district heating plant is empowered with a model that predicts the future heat demand – heat load forecast. Based on this model, controller generates values that meet the predicted heat demand without violating the system constraints. In addition to that, IoT devices that are part of the district heating system provide quantitative measurements of the processes they are involved in, which in turn generate data streams that have not existed before and provide for improvements of modelling and control of the system.

Abstract: During hip replacement procedure femoral head is removed and instead of it implant is inserted into the hollow femur. When an implant is inserted into the fractured bone the healing process starts to happen, which means that the newly formed bone interlocks with the inserted implant. Interlocking capacity is commonly analyzed using in vivo experiments, and only several papers have information on numerical approach. The goal of this study was to use the numerical approach to analyze the interlocking capacity of the modified hip implant surface. Numerical approach included implementation of the finite element method (FEM) for simulation of the interlocking capacity of modified surface topographies in the titanium alloys (Ti-6Al-4V) implants. Analyzed topographies were produced by the electron beam (EB) technique on the surface of the titanium alloy. The structures were analyzed by infinite focus microscopy and exported as a 3D profile for the application of the finite element method. The biggest advantage of this approach is that it can provide information about different types of topographies under a wide range of loading conditions before needing to insert an implant into an animal.

Abstract: Sleep apnea is a common sleep-related disorder caused by the obstruction of the respiratory tract and the absence of respiratory flow. 18 million Americans are estimated to suffer from sleep apnea, out of which 80% are thought to go undiagnosed. Nowadays, apnea detection is done using a Polysomnography (PSG) test during sleeping hours, which requires that the patient spends a night in a specialized center wearing several sensors to monitor his/her state. After that, medical trained staff checks the recordings/ markings and manually corrects the scoring. This process is time-consuming, which sets the motivation for this study to provide and validate an automatic apnea detection algorithm using the raw PSG recordings of 50 patients. Apnea is detected primarily employing the nasal flow signal, combined with oxygen saturation (SpO2) for hypopnea classification. The proposed approach achieves 77.124% accuracy (SD was 7.7) with around 55.3% sensitivity and around 82.5% specificity. Such an automatic algorithm will help drastically reduce the necessary time to analyze patients’ condition.

Abstract: The aim of this work was to computationally examine the biomechanical characteristics of the patient-specific AortoIliac Aneurysm (AIA) employing the fluid-solid interaction (FSI) and finite element method (FEM). Beside clinical assessment, computational modeling and simulations have great importance for improving aneurysmal examination and patient monitoring. In that order, the patient-specific 3D model of AIA was created, based on the computed tomography (CT) data. The blood flow simulation and interaction with the aortic wall, considering existence of the intraluminal thrombus (ILT), gave a more specific and more complete analysis of aneurysmal biomechanical characteristics. Assuming the blood as incompressible, viscous and laminar fluid, with an average properties and parabolic flow, the computational simulation of cardiac cycle was performed. The Von Mises stress, as the main parameter for evaluation of aortic wall degradation and rupture risk, was analyzed at the peak systolic moment. Presence of high stress at the aortic bifurcation and aneurysmal neck indicated the need for operative treatment. Furthermore, the obtained blood flow characteristics showed presence of stagnant blood flow at the maximal aneurysmal diameter, which affects the ILT and aneurysmal growth. In order to include the computational simulations in a daily clinical practice, which may lead to better prevention of aneurysmal rupture and evaluation of operative treatment need, a larger number of patients should be investigated in further studies.

Abstract: Integrating hand-written with generated code can sometimes be challenging part of implementing the MDSE approach, especially if the clean separation is not possible. To overcome this problem, we based our solution on Microsoft’s Roslyn project. During our research, we found that Roslyn’s Syntax Tree API is difficult to use due inherent properties of its implementation. In this paper, we present our RoseLib library that abstracts large part of this implementation, which liberates developers from remembering unnecessary details and makes development process much more efficient.

Abstract: In this paper, we propose a methodology for behavior variation and anomaly detection from acquired sensory data, based on temporal clustering models. Data are collected from smart cities that aim to become fully “elderly-friendly”, with the development and deployment of ubiquitous systems for assessment and prediction of early risks of elderly Mild Cognitive Impairments (MCI) and frailty. Our results show that Hidden Markov Models (HMMs) allow efficient (1) recognition of significant behavioral variation patterns and (2) early identification of pattern changes.

Abstract: The mandible is one of the frequently fractured bones in skeletal system of man. Reconstruction of the damaged mandible bone requires detailed and precise preoperative planning, which enable maxillofacial surgeons to improve execution of surgical interventions. For that purpose, 3D geometrical model of human mandible of specific patient is necessary in preoperative planning. In previous research, 3D parametric model of human mandible body was developed using Methods of Anatomical Features (MAF). The 3D parametric model is defined as point cloud, where coordinates of points are described by parametric functions, whose components are morphometric parameters (dimensions which are read on medical images). Ten central and bilateral morphometric parameters were used to create the parametric model of human mandible. The main benefit of this model is ability to adapt model to the specific patient, by applying patient specific values of morphometric parameters. The main focus of this study was to reduce number of parameters, while at the same time preserve geometrical accuracy of the personalized model. With this aim, statistical analysis (multiple regression) was applied in order to create “best fit” model with reduced number of morphometric parameters. Geometrical accuracy of the obtained parametric model was tested by the application of the deviations analysis in CATIA software. Deviations analysis was performed between input and resulting model. The results of analysis satisfied necessary accuracy of the models.