Shusaku Tsumoto

Automated extraction of medical expert system rules from clinical databases based on rough set theory

Abstract Automated knowledge acquisition is an important research issue to solve the bottleneck problem in developing expert systems. Although many inductive learning methods have been proposed for this purpose, most of the approaches focus onlu on inducing classification rules. However, medical experts also learn other information important for diagnosis from clinical cases. In this paper, a rule induction method is introduced, which extracts not only classification rules but also other medical knowledge needed for diagnosis. This system was evaluated on three clinical databases, whose experimental results show that our proposed method correctly induces diagnostic rules and estimates the statistical measures of rules.

Knowledge discovery in clinical databases and evaluation of discovered knowledge in outpatient clinic

Abstract Rule induction methods have been proposed in order to discover knowledge automatically from databases. However, conventional approaches do not focus on the implementation of induced results into an expert system. In this paper, the author focuses not only on rule induction but also on its evaluation and presents a systematic approach from the former to the latter as follows. First, a rule induction system based on rough sets and attribute-oriented generalization is introduced and was applied to a database of congenital malformation to extract diagnostic rules. Then, by the use of the induced knowledge, an expert system which makes a differential diagnosis on congenital disorders is developed. Finally, this expert system was evaluated in an outpatient clinic, the results of which show not only that the system performs as well as a medical expert, but also that the system is very useful for instruction to medical residents.

PRIMEROSE: PROBABILISTIC RULE INDUCTION METHOD BASED ON ROUGH SETS AND RESAMPLING METHODS

Automated knowledge acquisition is an important research issue in machine learning. Several methods of inductive learning, such as ID3 family and AQ family, have been applied to discover meaningful knowledge from large databases and their usefulness is assured in several aspects. However, since their methods are of a deterministic nature and the reliability of acquired knowledge is not evaluated statistically, these methods are ineffective when applied to domains essentially probabilistic in nature, such as medical domains. Extending concepts of rough set theory to a probabilistic domain, we introduce a new approach to knowledge acquisition, which induces probabilistic rules based on rough set theory (PRIMEROSE) and develop a program that extracts rules for an expert system from a clinical database, using this method. The results show that the derived rules almost correspond to those of the medical experts.

Modelling Medical Diagnostic Rules Based on Rough Sets

This paper discusses the characteristics of medical reasoning and shows the representation of these diagnostic models by the use of rough set theory. The key ideas are both a variable precision rough set model, which corresponds to an ordinal positive reasoning, and an upper approximation of a target concept, which corresponds to a focusing procedure. Acquired representation suggests that rough set model should be closely related with medical diagnosis.

Risk Mining in Medicine: Application of Data Mining to Medical Risk Management

Organizations in our modern society grow larger and more complex to provide advanced services due to the varieties of social demands. Such organizations are highly efficient for routine work processes but known to be not robust to unexpected situations. According to this observation, the importance of the organizational risk management has been noticed in recent years. On the other hand, a large amount of data on the work processes has been automatically stored since information technology was introduced to the organizations. Thus, it has been expected that reuse of collected data should contribute to risk management for large-scale organizations. This paper proposes risk mining, where data mining techniques were applied to detection and analysis of risks potentially existing in the organizations and to usage of risk information for better organizational management. We applied this technique to the following three medical domains: risk aversion of nurse incidents, infection control and hospital management. The results show that data mining methods were effective to detection of risk factors.

Automated discovery of positive and negative knowledge in clinical databases

In this article, the characteristics of two measures, classification accuracy and coverage, were discussed. We showed that both measures are dual, and that accuracy and coverage are measures of both positive and negative rules, respectively. Then, an algorithm for induction of positive and negative rules was introduced. The proposed method was evaluated on medical databases, and the experimental results show that induced rules correctly represented expert knowledge. Several interesting patterns were also discovered.

Evaluation of rule interestingness measures in medical knowledge discovery in databases

OBJECTIVE We discuss the usefulness of rule interestingness measures for medical KDD through experiments using clinical datasets, and, based on the outcomes of these experiments, also consider how to utilize these measures in postprocessing. METHODS AND MATERIALS We first conducted an experiment to compare the evaluation results derived from a total of 40 various interestingness measures with those supplied by a medical expert for rules discovered in a clinical dataset on meningitis. We calculated and compared the performance of each interestingness measure to estimate a medical experts interest using f-measure and correlation coefficient. We then conducted a similar experiment for hepatitis. RESULTS AND CONCLUSION The comprehensive results of experiments on meningitis and hepatitis indicate that the interestingness measures, accuracy, chi-square measure for one quadrant, relative risk, uncovered negative, and peculiarity, have a stable, reasonable performance in estimating real human interest in the medical domain. The results also indicate that the performance of interestingness measures is influenced by the certainty of a hypothesis made by the medical expert, and that the combinational use of interestingness measures will contribute to support medical experts to generate and confirm their hypotheses through human-system interaction.

Mining Complex Data

Session A1.- Using Text Mining and Link Analysis for Software Mining.- Generalization-Based Similarity for Conceptual Clustering.- Trajectory Analysis of Laboratory Tests as Medical Complex Data Mining.- Session A2.- Conceptual Clustering Applied to Ontologies.- Feature Selection: Near Set Approach.- Evaluating Accuracies of a Trading Rule Mining Method Based on Temporal Pattern Extraction.- Session A3.- Discovering Word Meanings Based on Frequent Termsets.- Quality of Musical Instrument Sound Identification for Various Levels of Accompanying Sounds.- Discriminant Feature Analysis for Music Timbre Recognition and Automatic Indexing.- Session A4.- Contextual Adaptive Clustering of Web and Text Documents with Personalization.- Improving Boosting by Exploiting Former Assumptions.- Discovery of Frequent Graph Patterns that Consist of the Vertices with the Complex Structures.- Session B1.- Finding Composite Episodes.- Ordinal Classification with Decision Rules.- Data Mining of Multi-categorized Data.- ARAS: Action Rules Discovery Based on Agglomerative Strategy.- Session B2.- Learning to Order: A Relational Approach.- Using Semantic Distance in a Content-Based Heterogeneous Information Retrieval System.- Using Secondary Knowledge to Support Decision Tree Classification of Retrospective Clinical Data.- POM Centric Multi-aspect Data Analysis for Investigating Human Problem Solving Function.

Rough representation of a region of interest in medical images

This paper introduces the rough representation of a region of interest (ROI) in medical images. The main advantage of this method is its ability to represent inconsistency between the knowledge-driven shape and image-driven shape of a ROI using rough approximations. The method consists of three steps including preprocessing. First, we derive discretized attribute values that describe the characteristics of a ROI. Next, using all attributes, we build up the basic regions in the image so that each region includes voxels that are indiscernible on all attributes. Finally, according to the given knowledge about the ROI, we construct an ideal shape of the ROI and approximate it by the basic categories. Then the image is split into three regions: a set of voxels that are (1) certainly included in the ROI (Positive region), (2) certainly excluded from the ROI (Negative region), (3) possibly included in the ROI (Boundary region). The ROI is consequently represented by the positive region associated with some boundary regions. In the experiments we show the result of implementing a rough image segmentation system.

Comparison of clustering methods for clinical databases

Clustering methods can be viewed as unsupervised learning from a given dataset. Even without domain knowledge or labels such as the names of diseases given by medical experts, these methods generate partition of datasets. In some cases, these new generated classes lead to discovery of a new disease or new concept. This paper discusses how clustering methods work on a practical medical data set. For comparison, the following four clustering methods were selected and evaluated on a dataset on meningoencephalitis: single- and complete-linkage agglomerative hierarchical clustering, Wards method and rough clustering. For comparison, a single similarity measure, a linear combination of the Mahalanobis distance between numerical attributes and the Hamming distance between nominal attributes was given to each clustering method. Usefulness of the clustering methods was evaluated from the following viewpoints: (1) the quality of generated clusters, (2) correspondence between the attributes used to generate the high-quality clusters and clinical knowledge. The experimental results showed that the best clusters were obtained using Wards method where the clinically reasonable attributes were selected, which also suggested that this similarity measure would be applicable to the medical data sets.