Committed to the Open Source Movement in Healthcare
Established
16 October 1998
Copyright
© 1998–2008
Health informatics Europe
The Centre for Medical
Informatics at the University of Maribor, Slovenia  Head: Prof. Dr. Peter Kokol Researchers: Špela Hleb - Babi, Vili Podgorelec, Milan Zorman, and Matej Šprogar The main research areas covered by the Centre for Medical Informatics are:
Design methods for medical information systems Mechanical systems increase our physical abilities (we use cranes to lift vast amounts, telescopes to see farther, etc.), but. Intelligent systems are power tools for heavy lifting in the information world - they "complement, extend, and amplify" our ability to think and solve problems. The difference between intelligent systems and more »usual« expert systems or knowledge-based systems is that intelligent systems are not necessarily smarter than expert or knowledge-based systems in terms of the quantity or types of knowledge or reasoning they employ; in fact, the research and application challenges are largely the same. However, intelligent systems for the most part exploit additional tools and technologies that make them easier to use, easier to build and maintain, easier to integrate with conventional information systems and in addition an appropriate human interface is required, targeted at the intended users. To be able to design successful intelligent medical systems (IMS) we must design them with an appropriate design methodology – but very few, if any such methods exist in the real world. Thereafter as a design framework we use the MetaMet - a two-leveled system design approach developed by the authors of this summary. The MetaMet is defined as a process in which a problem situation is transformed into an improved situation - employing a computerised IMS - appreciating the epistemology of the MetaMet, the characteristics of the problem situation, recent developments in IMS and related research and science in general (standards, norms, technology, etc.) and the principle of uncertainty performed in an endless learning loop. Intelligent systems Medical knowledge is expanding rapidly. Every-day responsibilities demand that medical staff generate a vast amount of statistical reports concerning detailed patient information, accounting purposes, reimbursement, insurance needs, treatments, research and a myriad of others. The appearance of the new computer-based information technology and especially the introduction of intelligent systems with their ability to learn has initiated the possibility to enormously ease these routine activities and enable the medical staff to devote more time to enhanced creative work. When looking at intelligent medical systems, there is always the problem that only successful systems are reported and that therefore it is impossible to accurately estimate the success rate. In the past twenty years, hundreds of systems have been reported on, but few are used. One of the success stories is the database of drug interactions for pharmacies. Why have so few of the systems been successful in the medical field? A review done by one journal for the whole of 1997 revealed 13 articles that dealt with intelligent systems in medicine. Of these 13, one was evaluated thoroughly and is probably usable outside the department where it was developed. The others come under the "look what we can build" heading. One that was being developed as a teaching tool had no discussion of the human interface or of a trial with a novice. Statistical analysis where used was often not an appropriate analysis. By contrast, neural networks themselves, evolutionary programming, automated decision trees and similar can be used for statistical analysis especially of non-parametrically distributed data. Once again an appropriate human interface is required, targeted at the intended user in addition to the possibility of easy integration into existing medical information systems. According to that we will develop new design methods and new machine learning methods. One of our most successful methods is a hybrid one. In it we first construct some decision trees and some neural networks using conventional building techniques and perform evolutionary computation on them. The best individual neural networks are transformed into decision trees and these then compete with the best decision tree individuals and finally after some iterations the optimal decision tree and the optimal neural network are constructed which can then be used as advanced adaptive statistical tools. The whole construction process is improved with the use of adaptive evolutionary cost function, which represent a great advancement over traditional genetic algorithmic methods. Quality and complexity of information systems Although the software metrics are becoming more and more recognised in software engineering, the field of measuring the complexity of software is still not yet successful enough. The traditional complexity metrics are language dependent. We have to use a different metric to assess the complexity of a program written in a higher level programming language (even worse we have to use a different metric or at least a different tool for each programming language), another metric for a program written in an assembler language and some other metrics to asses the complexity of object ore executable code. But that is not all. A program or an information system can be represented in many other forms: requirements, specification, documentation, user interfaces, etc. and all those representations can be manifested in very different appearances: written text, graphical, symbolic, formal languages, etc. In addition the output of a traditional complexity metric is a number, usually without any "physical" meaning and unit. To overcome above problem we propose a "language and form" independent metric with critical values related to complexity, information content and entropy called Alpha metrics. Alpha metric is based on the fact that a computer program is a string of symbols, and we can asses the complexity of a string with the calculation of long range correlation between symbols – approach successfully used in the DNA decoding and recently on human writings. Long-range power law correlations (LRC) have been discovered in a wide variety of systems. Recognising a LRC is very important for understanding the system’s behaviour, since we can quantify it with a critical exponent - alpha. Quantification of this kind of scaling behaviour for apparently unrelated systems allows us to recognise similarities between different systems, leading to underlying unification. For example, the recent research has shown that DNA sequences, human writings and computer programs can be analysed using very similar techniques. In our research we would like to analyse the use of alpha metric in:
In both theoretical settings and real world applications. The CMI co-operates in some national and international projects. The most important are:
|
