Overview 

Division of electronics continues to work on research and development of novel intelligent data and signal analysis techniques, and their application in areas of highest scientific interest such as: biomedicine, computational biology and bioinformatics, as well as on development of advanced measurement and signal processing systems. With the start of the 6th European Framework STREP project HEARTFAID, Laboratory for information systems initiated research in knowledge engineering, primarily focused on knowledge representation and management in the field of heart failure disease. Laboratory for stochastic signals and processes research is continuing development of measurement methodology and data acquisition protocols for human locomotion and human jaw kinematics analysis with Faculty of Kinesiology and with School of Dental Medicine from University of Zagreb.

Achievements

Machine learning and data mining research

     Main developments in this area were concentrated on techniques that foster either descriptive or predictive capabilities of existing methods such as subgroup discovery, SVM, Parallel Random Forests in combination with diverse feature construction and filtering methods. An enhanced machine learning methodology for the analysis of 1D electrophoresis experiments has been developed and applied for plant tissue discrimination. Machine learning models based on support vector regression technique and self-organizing-maps are developed for the prediction of mechanism of action and assess antitumor potential of a new class of chemical entities, based on biological responses from in-vitro experiments. Applications of optimized methodologies range from molecular biology (Gamberger D et al., Inform Med Slov 2006: 46:51) and drug-discovery, to information processing and engineering.

Knowledge engineering

HEARTFAID project is a EU6FP ICT-STREP project focussed on the research and development of knowledge enabled platform for the support of healthcare for the elderly patients suffering from heart failure syndrome (Conforti D et al., J Inform Tech Healthcare 2006: 4). Laboratory for Information Systems is leading work package 4 – Knowledge representation, discovery and management. During first 10 months of the project we have worked on knowledge representation issues and ontology formation for the heart failure domain, based on deliverables originating from medical partners.

Overview

       Division of electronics continues to work on research and development of novel intelligent data and signal analysis techniques and their application in biomedicine, computational biology and bioinformatics, as well as on development of advanced measurement techniques. Strong multi-disciplinary orientation this year was confirmed with strong publication record in different fields.    This year, Division was granted a long term research programme entitled “Computational knowledge discovery in scientific applications (PI Dragan Gamberger), financed by the Croatian Ministry of Science, Education and Sports. The research programme involves 5 related research projects, three of them within the Division, including a number of researchers from different academic institutions of the University of Zagreb.  Besides these scientific projects, Division works on a number of international multilateral (FP6 project HEARTFAID) and bilateral projects. Laboratory for stochastic signals and processes started collaboration with Končar Institute for Electrical Engineering on reliability of programmable logic devices in industrial embedded systems.   Our researchers contribute to high education providing both undergraduate and postgraduate courses at the University of Zagreb.

Achievements

Knowledge discovery methodology and applications

A new technique, called iterative subgroup discovery, has been developed and applied for the data analysis of patients suffering from brain ischemia (Gamberger et al., 2007). Advanced data mining methodology has also been applied for intelligent analysis of the data collected under the clinical study of posttraumatic stress disorder (Marinić I et al., 2007).  A complex knowledge discovery procedure including a sequence of data processing and filtering algorithms together with support vector regression model development, has been designed and successfully applied for estimating of antitumor activity on tumor cell lines (IC₅₀ values, in-vitro testing) of a series of crown ether compounds.

Meta-modeling of complex real-world processes using machine learning algorithms and data mining tools is relatively unexplored field. An application to the problem of so called dose-rate buildup factors in radiation protection has proved the potential and advantages of machine learning paradigm in formulating accurate and reliable approximate models of complex processes (Trontl K, Šmuc T and Pevec D, 2007).

Knowledge representation and reasoning for healthcare

Within the activities related to HEARTFAID project (EU6FP ICT-STREP) focused on the research and development of knowledge enabled platform for the support of healthcare for the elderly patients suffering from heart failure syndrome (Conforti D et al., J Inform Tech Healthcare 2006: 4). Laboratory for Information Systems is leading work package 4 – Knowledge representation, discovery and management. During the second year our focus was on definition of knowledge representation and formulation of ontology for the heart failure domain. Within this activity, a new ontology reasoner-interpreter called CWI, based on OWL language (Ontology Web Language) has been built for testing the new heart failure disease ontology, and as a prototype for a decision support services of the HEARTFAID platform (see Fig. 1).

Advanced measurement systems and signal processing techniques

A numerical procedure for the computation of natural gas molar heat capacity, the isentropic exponent, and the Joule-Thomson coefficient has been derived using fundamental thermodynamic equations, DIPPR AIChE generic ideal heat capacity equations, and AGA-8 extended virial-type equation of state. The procedure enables precise calculation of natural gas properties and can be efficiently applied to flow-rate measurements (Marić, I, 2007). Development of new methods for fractal features extraction from time series based on data embedding in pseudo phase space. Trajectory vectors projection on principal axes reveals scale invariant statistics. Method is illustrated on human heartbeat time series (interbeat interval). An ongoing research activity in signal processing field is the development of the integrated approach for the quantification of recurrences in short sequences. Coupling of the so called Recurrence Quantification Analysis technique, with machine learning algorithms, into a new software system will enable automated analysis of short sequences or signals (from few tens to  few hundred points), involving simultaneous optimization of the RQA transform with respect to the particular annotation/classification problem (such as protein sequences, heart rate variability analysis).           

Education

1.         Knowledge Discovery in Medical Domains,  Dragan Gamberger, PhD Program at the Medical School, University of Zagreb.

2.         Optical communication networks, Branka Medved Rogina,  Graduate Program, Faculty of Electrical Engineering and Computing, Zagreb,

3.         Algorithms in Bioinformatics, Strahil Ristov, PhD Program, Faculty of Electrical Engineering and Computing,  University of Zagreb

4.         Algorithms and data structures, Strahil Ristov, Graduate Program, Faculty of Electrical Engineering and Computing,  University of Zagreb

Projects

Projects supported by the Ministry of Science, Education and Sport (MSES)

1.         Machine Learning Algorithms and their Application, Dragan Gamberger

2.         Computational Intelligence Methods in Measurement Systems, Ivan Marić

3.         Real Life Data Measurement and Characterization, Branka Medved Rogina

(These projects are under umbrella of the long-term research programme that includes

5 related projects, under the name: Computational knowledge discovery in scientific applications, lead by Dragan Gamberger

Research, developmental and international projects

1.         HEARTFAID – A knowledge based platform of services for supporting medical-clinical management of heart failure within elderly population, Dragan Gamberger (EUFP6, ICT-STREP project)

2.         Intelligent Data Analysis, Dragan Gamberger, (Croatian-Slovenian bilateral project)

3.         Inductive Databases for Genomics and Proteomics, Tomislav Šmuc, (Croatian-Slovenian bilateral project)

4.         Reliability of programmable logic devices in industrial embedded systems, Branka Medved Rogina, (R&D project with Končar Institute for Electrical Engineering)

Selected publications

1. Gamberger D, Lavrač N, Krstačić A, Krstačić G. Clinical data analysis based on iterative subgroup discovery: Experiments in brain ischaemia data analysis.  Appl Intell 2007: 27: 205.

2.      Trontl K, Šmuc T, Pevec D: Support vector regression model for the estimation of γ -ray buildup factors for multi-layer shields. Ann Nucl En 2007: 34(12): 939.

3. Stipčević M, Medved Rogina B.: Quantum random number generator based on photonic emission in semiconductors.  Rev Sci Instrum 2007: 78:  45104.

4.      Marić I: A procedure for the calculation of the natural gas molar heat capacity, the isentropic exponent, and the Joule-Thomson coefficient. Flow Meas Instrum 2007 18: 18.

5. Marinić I, Supek F, Kovačić Z, Rukavina L, Jendričko T, Kozarić-Kovačić D. Posttraumatic Stress Disorder: Diagnostic Data Analysis by Data Mining Methodology. Croat Med J 2007: 48(2): 185.

OVERVIEW

Main topics of research in Division of electronics comprise of research and development of novel intelligent data and signal analysis techniques, knowledge representations for information systems, and variety of tailored applications of these techniques in biomedicine, computational biology and bioinformatics, as well as in development of advanced measurement techniques. Strong multi-disciplinary orientation is best reflected in publication record of Division’s members, and collaborations in multidisciplinary projects. In 2008, Division was granted a new research project entitled “Machine learning of predictive models in computational biology”, which is to be financed by the Croatian Ministry of Science, Education and Sports, in the period 2008-2011. Members of the Division organized a 2 day workshop for the research programme “Computational knowledge discovery in scientific applications” comprised from 5 related research projects, three of them within the Division, including a number of researchers from different faculties of the University of Zagreb and industrial partners (http://lis.irb.hr/KDSA2008/).   Besides these scientific projects, Division works on a number of international multilateral (EU FP6 project HEARTFAID) and bilateral projects, as well as R&D projects with industry. Division’s contribution to the high education curriculum was extended with a graduate course on artificial intelligence at the Faculty of Science.

Achievements

Machine learning and knowledge technology research

Filtering proteomic signals

New computational method applied to measurements derived from chromatographic separations of biomolecular mixtures (for instance capillary or gel electrophoresis), has been developed in collaboration with the group of prof. Krsnik-Rasol at the Department of Biology, at Faculty of science. The method involves several steps in which a larger number of measurements is processed using machine learning techniques (PCA, RF) enabling at the end reliable determination of relevant fractions for the particular discrimination problem, and improved computational models for class distinction based on a filtered set of relevant fractions (Fig 1).

Knowledge representation and reasoning for healthcare

During the final year of the research and development efforts within HEARTFAID project (EU6FP ICT-STREP) included the coupling of Bayesian network reasoning to already developed ontology based reasoner. This extension significantly enhances decision support capabilities of the system, providing reasoning in situations in which some information is missing. Besides the work on decision support part of the platform, extensive analyses of the proprietary follow-up study database has been done in order to form new models for the prognosis of elderly patients suffering from heart failure syndrome.

Finite-state data representation algorithms

We have developed an efficient method for the implementation of a sub-sequential transducer. A transducer is a finite state automaton that on every input sequence symbol produces a specific output symbol during transition from state to state. If the whole output sequence is associated with the final state, transducer is called sub-sequential. Our implementation is the most efficient one regarding space requirements while preserving high speed inherent to automata structures. The applications for transducers are numerous and vary from natural language processing and human machine interface, to machine learning and data mining.

Advanced signal processing techniques and measurement systems

Biomedical applications

Two aplplications that were in our current research focus are related to human walk and human jaw movement. First method comprises from the extraction of dynamic features of the human walk and the embedding of related time series data in high pseudo phase space. Subsequent projection of the embedded vectors on principal axes or/and on main diagonal of embedding space enables extraction of fractal features from the data (Fig. 2). This research is part of our collaboration with Faculty for Kinesiology University of Zagreb. Second application is based on measurements using tri-axial MEMS wireless acceleration sensor wherein accelerometric profiles of averaged vertical jaw acceleration and velocity are measured.  In collaboration with School of Dental Medicine University of Zagreb, a study including healthy subjects without any signs or symptoms of temporomandibular disorders, is used to obtain standard profiles necessary for future studies.