Issues in the Structuring and Acquisition of an Ontology for Medical Language Understanding

An efficient, large-scale, non-lattice-detection algorithm for exhaustive structural auditing of biomedical ontologies

One of the basic challenges in developing structural methods for systematic audition on the quality of biomedical ontologies is the computational cost usually involved in exhaustive sub-graph analysis. We introduce ANT-LCA, a new algorithm for computing all non-trivial lowest common ancestors (LCA) of each pair of concepts in the hierarchical order induced by an ontology. The computation of LCA is a fundamental step for non-lattice approach for ontology quality assurance. Distinct from existing approaches, ANT-LCA only computes LCAs for non-trivial pairs, those having at least one common ancestor. To skip all trivial pairs that may be of no practical interest, ANT-LCA employs a simple but innovative algorithmic strategy combining topological order and dynamic programming to keep track of non-trivial pairs. We provide correctness proofs and demonstrate a substantial reduction in computational time for two largest biomedical ontologies: SNOMED CT and Gene Ontology (GO). ANT-LCA achieved an average computation time of 30 and 3 sec per version for SNOMED CT and GO, respectively, about 2 orders of magnitude faster than the best known approaches. Our algorithm overcomes a fundamental computational barrier in sub-graph based structural analysis of large ontological systems. It enables the implementation of a new breed of structural auditing methods that not only identifies potential problematic areas, but also automatically suggests changes to fix the issues. Such structural auditing methods can lead to more effective tools supporting ontology quality assurance work.

Mining non-lattice subgraphs for detecting missing hierarchical relations and concepts in SNOMED CT

Objective

Quality assurance of large ontological systems such as SNOMED CT is an indispensable part of the terminology management lifecycle. We introduce a hybrid structural-lexical method for scalable and systematic discovery of missing hierarchical relations and concepts in SNOMED CT.

Material and Methods

All non-lattice subgraphs (the structural part) in SNOMED CT are exhaustively extracted using a scalable MapReduce algorithm. Four lexical patterns (the lexical part) are identified among the extracted non-lattice subgraphs. Non-lattice subgraphs exhibiting such lexical patterns are often indicative of missing hierarchical relations or concepts. Each lexical pattern is associated with a potential specific type of error.

Results

Applying the structural-lexical method to SNOMED CT (September 2015 US edition), we found 6801 non-lattice subgraphs that matched these lexical patterns, of which 2046 were amenable to visual inspection. We evaluated a random sample of 100 small subgraphs, of which 59 were reviewed in detail by domain experts. All the subgraphs reviewed contained errors confirmed by the experts. The most frequent type of error was missing is-a relations due to incomplete or inconsistent modeling of the concepts.

Conclusions

Our hybrid structural-lexical method is innovative and proved effective not only in detecting errors in SNOMED CT, but also in suggesting remediation for these errors.

[SIMS REIN: a multi-source information system for end-stage renal disease]

In France, the prevalence of End-Stage Renal Disease (ESRD) is not precisely known. The sources of information are scattered and not coordinated. Consequently, care is ill adapted to meet the demand. The Multi-Source Information System is the basis of the Renal Epidemiology and Information Network (REIN). It is dedicated to improve and organise our medical and epidemiological knowledge of ESRD and to aid public health decision-making in this area. The proposed approach is based on the datawarehouses. This model allows a unified vision of scattered data into distinct databases, for a better management, be it particular (patient follow-up) or global (regional follow-up), with a finality of aid in decision-making. Several categories of problems were considered: the global conception of the information system, the organisation of the datawarehouse, which offers different viewpoints of the data, the integration of heterogeneous data coming from different sources, data exchange and definition of a specific ontology.

A formal framework of knowledge to support rational psychoactive drug selection

Rational psychoactive drug selection is a data and knowledge intensive task which requires true expertise from clinical, pathophysiological and pharmacotherapeutic knowledge. This paper presents a framework of knowledge, which relates concepts from several disciplines required for psychoactive drug selection in a formal way. A framework, when based on formal semantics, avoids ambiguity and gives conceptual clarity and supports precise use of terminology which is required when many domain experts (clinicians, pharmacologists and basic science researchers) are involved. A formal framework permits linking of existing classification systems. It furthermore can serve as a knowledge base for drug selection decision support systems.

Reconciling users' needs and formal requirements: issues in developing a reusable ontology for medicine

A common language, or terminology, for representing what clinicians have said and done is an important requirement for individual clinical systems, and it is a pre-requisite for integrating disparate applications in a distributed telematic healthcare environment. Formal representations based on description logics or closely related formalisms are increasingly used for representing medical terminologies. GALEN's experience in using one such formalism raises two major issues, as follows: how to make ontologies based on description logics easy to use and understand for both clinicians and applications developers; what features are required of the ontology and description logic if they are to achieve their aims. Based on our experience we put forward four contentions: two relating to each of these two issues, as follows: that natural language generation is essential to make a description logic based ontology accessible to users; that the description logic based ontology should be treated as an "assembly language" and accessed via "intermediate representations" oriented to users and "perspectives" adapting it to specific applications; that independence and reuse are best supported by partitioning the subsumption hierarchy of elementary concepts into orthogonal taxonomies, each of which forms a pure tree in which the branches at each level are disjoint but nonexhaustive subconcepts of the parent concept; that the expressivity of the description logic must include support for transitive relations despite the computational cost, and that this computational cost is acceptable in practice. The authors argue that these features will be necessary, though by no means sufficient, for the development of any large reusable ontology for medicine.

Present and future trends with NLP

This paper reflects some desiderata on the role of Natural Language Processing (NLP) in the coming years as foreseen in the medical domain. During the Medical Informatics Europe (MIE) conference in 1997, the NLP track was composed by numerous papers on natural language, knowledge representation, nomenclatures and classifications. Indeed, the medical community is looking for solutions for the future which are going to emerge from more powerful desktop and for which the successful softwares of tomorrow are not yet identified. Presently this same community's needs remains unsatisfied with healthcare professionals writing at best their patient medical records in a text processing system. We stand today closer to the typewriter than to any modern solution which could radically transform the treatment of information related to patients. What kind of hurdles are there still in front of us before we reach this new territory?

A Dutch medical language processor

This paper describes the current state of a medical language processor for Dutch. The goal is to implement a language specific front-end compatible with some existing applications that aim at the intelligent extraction and processing of information from patient discharge summaries. A complete chain for processing and understanding Dutch medical documents will be the ultimate result. The text focuses mainly on the language specific aspects of the language processing chain. Evaluation results of the already functioning components are given as well as an outline for future developments and enhancements. A short theoretical background is provided (cf. also [1-3]: Rossi Mori et al., Proc. SCAMC 90, 1990, pp. 185-189; Wingert, in: Informatics and Medicine, an advanced course, Springer-Verlag. 1977, pp. 579-646; Wingert, Proc. MEDINFO 80, 1980, pp. 1321-1331) before the description of each component in order to familiarise the non-experienced reader with the basic notions of computational linguistics.