Beata Jankowska

How to Acquire and Structuralize Knowledge for Medical Rule-Based Systems?

The intention of a medical expert system is to help doctors make right diagnostic and therapeutic decisions concerning, sometimes not very well-known to them, diseases. This expert system needs a high quality knowledge base. In order to design such a base one has to reach sources containing knowledge that is current, rich and based on reliable medical experiments. At the same time, due to various formats of this knowledge storing, its acquisition and structuralization to the form required by expert systems is not an easy task. Focusing our attention on medical rule-based systems, we propose the algorithms and tools that will be useful while designing such a knowledge base.

How to speed up reasoning in a system with uncertainty

One of the difficulties of using Artificial Neural Networks (ANNs) to estimate atmospheric temperature is the large number of potential input variables available. In this study, four different feature extraction methods were used to reduce the input vector to train four networks to estimate temperature at different atmospheric levels. The four techniques used were: genetic algorithms (GA), coefficient of determination (CoD), mutual information (MI) and simple neural analysis (SNA). The results demonstrate that of the four methods used for this data set, mutual information and simple neural analysis can generate networks that have a smaller input parameter set, while still maintaining a high degree of accuracy.

Discovering Medical Knowledge from Data in Patients' Files

An Individual Patients File (IPF) includes information about a patient and the course of his illness over many years. During this period, the doctor, taking permanent care of the patient, puts various data and observations into his IPF. In the paper we propose a kind of probabilistic and statistical analysis of the data stored in collections of IPFs. The collections can be made accessible by means of a medical Semantic Web. The proposed analysis requires the data to be expressed in the form of Bernoulli variables. As a result, it gives us medical hypotheses, with their premises-facts, conclusions and reliabilities. They can be presented in the form of production rules, used in medical Rule-Based Systems for automatic reasoning.

Semantic Data Integration in the Domain of Medicine

Finding an efficient solution to a problem of medical data integration is an important challenge for computer science and technology. Medical data integration is necessary both to retrieve missing data of an individual patient (horizontal integration) and to merge similar data from varied medical repositories (vertical integration) in order to increase data reliability. In the paper, we propose to solve the problem of medical data integration by means of an algebraic approach. In this approach also heterogeneous data can be considered. If only we know data taxonomies, can interpret data schemas and design schema mappings, then semantic and syntax differences are not an obstacle to integration.

Machine Ranking of 2-Uncertain Rules Acquired from Real Data

There are many places (e.g. hospital emergency rooms) where reliable diagnostic systems might support people in their work. They could have form of RBSs with uncertainty and use the techniques of forward and backward chaining in their reasoning. The number and the contents of derived hypotheses depend then both on the form of the system’s knowledge base and on the inference engine performance. The paper provides detailed considerations on designing and applying particular uncertain rules, namely 2-uncertain rules. They are equipped with two reliability factors, representing a kind of second order probability. The rules can be acquired from real data of attributive representation. In the paper we propose a method for calculating the two reliability factors. We also suggest how to take advantage of the factors during reasoning, in order to obtain reliable hypotheses. The factors help to rank the rules and to fire them in the best order.

Cost-Effectiveness Analysis of Transferring Medical Data over P2P Network

In the paper, a transfer of medical dataover P2P network is considered. The data are homogenous, of textual type. They are transferred between the nodes representing medical systems, along transmission lines. Although only some of the nodes can produce, process or integrate data, all of the nodes are able to propagate them to their neighbours. The costs of building the transmission lines are high, so an optimum network topology is being searched. The random graphs with bounded degree are used for modelling of such networks. The process of transferring data over network is examined by means of the original algorithm RST for searching rooted spanning trees.