[Objectives]: To propose a method for extracting relations between diseases and radiological findings.
 [Materials and Methods]: 1,155 sentences relating to radiological findings were selected from an electronical medical textbook for this study. First, dependency trees of the sentences were determined using syntactic analysis, and next, subtrees relating to radiological diagnoses were extracted from dependency trees using medical attributes. Finally, relations between a disease name, which is an entry word, and each finding in a subtree were extracted with a positive or negative attribute.
 [Results]: Relations between 124 diseases and 794 radiological findings were extracted with positive or negative attributes. Recall of relation extraction was at a rate of 66%, and precision was at a rate of 95%.
 [Conclusion]: Our feasibility study suggests the efficiency of our method for extracting relations between diseases and radiological findings which are useful for building medical ontology for radiological diagnoses.

View full abstract

 We are working on the development of the reporting system that can present input support information by extracting medical knowledge from a past report, and using this.
 In this system, first of all, the existing, natural sentence interpretation of radiogram report is structurized because of the natural language processing. Next, the candidate word is presented as input support information according to the part and the modality by using the statistical information of the structurized data. And, the user makes the interpretation of radiogram report by selecting the presented candidate word.
 As structurizing processing used for this system, We examined the realizability of highly accurate structurizing processing by the technique of applying the algorithm that was called “Support Vector Machine (SVM)” in the natural sentence resolved to the morpheme, and giving the word the attribute.
 In this thesis, details of the technology that extracts “Medical knowledge” such as “Part” and “Opinion” by first using this “SVM” from a past report and the experiment result of it are described. Next, the proposal of the technique that uses “Paraphrase” to improve the extraction accuracy of “Medical knowledge” and the experiment result of it are described.

View full abstract

 Background: For the automatic retrieval of medical terms from clinical documents, the problems still remain in processing the noun-noun compounds. In addition, the semantic relations extracted from the compounds are useful for building the ontologies in the domain. Purpose: The purpose of the study is to investigate the distribution of the semantic relations between the two atomic medical terms of the noun-noun compounds extracted from the clinical documents. Methods: 100 clinical reports for CT examination generated in July 2005 were collected in a random way at Hokkaido University Hospital. The medical noun-noun compounds were extracted from the reports. We indexed the UMLS semantic type, group for each atomic medical term in the compounds, extracted the UMLS semantic relation according to the index, and analyzed the distribution of the semantic relations. Results: Only 29.9% of the compounds had the UMLS semantic relation defined (mainly including “location_of” and “adjacent_to”), and 70.1% of them undefined. Conclusion: The semantic relations defined in the UMLS are not enough for the noun-noun compounds extracted from the clinical reports, indicating further efforts are needed.

View full abstract

 The semantic web technology has recently drawn increasing attention in the medical domain. In this paper, we present an experimental reporting system. The system retrieves usefull information from the clinical database, and present it for a report author. The database contains clinical data which are extracted from past diagnostic imaging reports. In the database, data elements are related each other by RDF (Resource Description Framework). Our research aims at feasibility of such systems.
 The clinical database are created from the past MRI reports on cerebrovascular disorder cases in the Hospital of Hyogo College of Medicine. The database holds relationship among body pars, observation, and diagnosis. Based on the experiments with past reports, we conclude that our system can handle large variety of reports and achieve reasonably quick response time. Also, based on the experiments with chest computed radiography reports in the Osaka University Hospital, we concluded that our structured data model can be easily extended to various modalities and body parts.

View full abstract

 In this paper, we address the importance of nursing risk management, show computational models of nursing risk management, and address the necessity of building nursing ontologies to perform computational nursing risk management.
 First, we address the importance and necessity of nursing risk management. Then we review nursing accidents or incidents from the cognitive aspects of human errors. Based on cognitive features, we logically model nursing risk management. We briefly show an abduction-based model and a scenario violation model that is an extended model of abduction that can deal with time information. Finally, we show the necessity of building a knowledge base and a scenario base for the inferences. In addition, we point out the necessity of building nursing ontologies for an automatic knowledge base genaration and show a perspective of nursing ontology building.

View full abstract

 This is a study report of medical ontology development which is generated by porting from the MedDRA thesaurus into OWL. The generated ontology is named MedDRA ontology. As a result of this ontology making, we can easily browse meaning of the MedDRA ontology by using the ontology viewer of the semantic web engine which has been developed by the CyberEdge Corporation Ltd. Also, we can edit easily the ontology and can search semantically for various information of the MedDRA ontology.

View full abstract

 Learning for surgical procedures has been changing from experimental learning or textural learning to skill learning using virtual reality (VR) simulation. VR simulation is very useful to learn for surgical procedures but it is very difficult to make VR simulation environment. Therefore a simulation library which named MVL (Medical Virtual Reality Library) is build for developing VR environment very efficiently. On the other hand, textbooks on surgery have amount knowledge for VR simulation developing. In this paper, we suggest a new ontology between medical VR simulations and textbooks on surgery. If this ontology is developed, we can connect two knowledge.
 Concretely, knowledge as surgical procedure on textbooks can make a VR simulation environment through MVL with machine learning automatically.

View full abstract

 Life Science Dictionary (LSD) is a versatile database of English and Japanese terms based on the quantitative analyses of biomedical corpora. To develop a thesaurus of LSD terms for future application to computer-assisted text mining, we have evaluated the frequency of LSD terms in the literature-based corpora, and mapped the LSD terms to the MeSH tree. Coverage of LSD English terms in a PubMed-based corpus was 80%. In 65,000 MeSH tree terms, LSD-matched terms were 20%, which was increased to 40% in a subpopulation of terms occurred in the English corpus. The MeSH-unmatched LSD terms included abbreviations, verbs, adjectives, adverbs and MeSH-unclassified terms. These results indicate the requirement of new comprehensive thesaurus tree covering complex English-Japanese translations.

View full abstract