Using UMLS to construct a generalized hierarchical concept-based dictionary of brain functions for information extraction from the fMRI literature

Abstract

With a rapid progress in the field, a great many fMRI studies are published every year, to the extent that it is now becoming difficult for researchers to keep up with the literature, since reading papers is extremely time-consuming and labor-intensive. Thus, automatic information extraction has become an important issue. In this study, we used the Unified Medical Language System (UMLS) to construct a hierarchical concept-based dictionary of brain functions. To the best of our knowledge, this is the first generalized dictionary of this kind. We also developed an information extraction system for recognizing, mapping and classifying terms relevant to human brain study. The precision and recall of our system was on a par with that of human experts in term recognition, term mapping and term classification. Our approach presented in this paper presents an alternative to the more laborious, manual entry approach to information extraction.

BrainKnowledge: A Human Brain Function Mapping Knowledge-Base System

“…The detailed methodology for extracting fMRI related terms was discussed elsewhere (Hsiao et al 2009). BrainKnowledge extracts brain structure terms with the terms that are in NeuroNames (Bowden and Dubach 2003), a neuroanatomical thesaurus available in the Unified Medical Language System (UMLS), and in the Talairach atlas (Talairach and Tournoux 1988).…”

“…Extracting terms for brain functions and experimental tasks is more difficult as, to the best of our knowledge, there is no publicly available resource on brain function vocabulary that is comprehensive enough for our purpose at this time. Thus, we developed a generalized hierarchical concept-based dictionary (Hsiao et al 2009) of brain functions for subsequent named entity extraction based on the UMLS, which integrates many terminologies such as MeSH, Psychological Index Terms and similar vocabulary sources. A two-step approach to term recognition and classification was proposed to extract terms related to experimental tasks and brain functions (Hsiao et al 2009).…”

“…Thus, we developed a generalized hierarchical concept-based dictionary (Hsiao et al 2009) of brain functions for subsequent named entity extraction based on the UMLS, which integrates many terminologies such as MeSH, Psychological Index Terms and similar vocabulary sources. A two-step approach to term recognition and classification was proposed to extract terms related to experimental tasks and brain functions (Hsiao et al 2009). The first step, term recognition, combines a dictionary and a rule-based approach to identify words that describe experimental tasks and brain functions.…”

“…In our dictionary, we merged terms from 23 vocabulary sources to yield more complete coverage than any single source could provide. With such broad, though incomplete, coverage, BrainKnowledge can achieve 72% precision and 73% recall for term recognition, which is on par with human expert performance (Hsiao et al 2009). …”

“…As a result, most researchers must continue to rely on manually inputting information into databases, as is the case with BrainMap and the Brede database. In order to resolve this problem of data entry, in our previous study (Hsiao et al 2009), we constructed a generalized hierarchical concept-based dictionary of brain functions (also referred to as the generalized hierarchical brain function tree) and had developed an information extraction algorithm for extracting terms that were relevant to brain functions and experimental tasks. The precision and recall of our information extraction algorithm was on par with that of human experts.…”