Canada Spends Logo

Explore our Budget 2026 analysis

Grant: Engage Grants Program

Dalhousie University

$24,768.00

Structured information extraction for mining free-text radiology reports

FY 2016-17

Summary
Palomino is a web document management company, with experience in the management of health documentsfor the Guysborough Antigonish Strait Health Authority in Nova Scotia. Radiology reports are dictated by thedoctor and speech is transcribed and stored as text. Turning text reports into structured information will be asignificant boost in the exploitation of electronic health information for population health. Such a capabilitywill significantly enhance Palomino's expertise in the space of electronic health records.Two challenges need to be addressed to achieve a sufficiently accurate system that extracts structured data fromunstructured text of radiology data. First, different types of errors in the text arising from the speech-to-textconversion need to be identified and corrected. Errors could be spelling mistakes (e.g., 'abnormaliti' as opposedto 'abnormality'), missing space or word segmentation errors (e.g., 'noevidence' as opposed to 'no evidence') orsymbol errors (e.g., 'intra abdominal' as opposed to 'intra-abdominal'). Second, entities (medical conditions)and their values (e.g. normal vs. abnormal) must be extracted from the corrected text, by reference to ataxonomy of medical conditions, and processing of the text, taking into account the abbreviated nature of thetext which may result in text that does not always contain fully-formed sentences in terms of syntax. Once acorpus of radiology reports has been converted into structured data, in the form of condition-value pairs,classification will be used over a large radiology report corpus to provide insights in population health.In this project, we will first design and implement a system by combining different algorithms to deal withdifferent types of errors in the text, we will then extract the medical conditions and the associated values tooutput the structured data.
Program
Engage Grants Program
Location
Nova Scotia, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data

As published

Application id
492418-2015
Name nom
Gu, Hong
Department département
Mathematics and Statistics
Organization id
79
Institution établissement
Dalhousie University
Province en
Nova Scotia
Province fr
Nouvelle-Écosse
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2016
Competition year année de concours
2015
Award amount
24768
Program id
EGP
Program name en
Engage Grants Program
Program name fr
Programme de subventions d'engagement partenariat
Group en
Research partnerships programs
Group fr
Programmes de partenariat
Committee code
1551
Committee name en
Atlantic Internal Decision Committee
Committee name fr
Comité de décision interne pour l'Atlantique
Area of application code
1211
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Medical and health sciences
Area of application fr
Sciences médicales et de la santé
Research subject code
2700
Research subject group en
Information technology
Research subject group fr
Technologie de l'information
Research subject en
Information technology
Research subject fr
Technologie de l'information
Partie
1
Nb partie
1
Application title
Structured information extraction for mining free-text radiology reports
Keyword
NSERC
Application summary
Palomino is a web document management company, with experience in the management of health documentsfor the Guysborough Antigonish Strait Health Authority in Nova Scotia. Radiology reports are dictated by thedoctor and speech is transcribed and stored as text. Turning text reports into structured information will be asignificant boost in the exploitation of electronic health information for population health. Such a capabilitywill significantly enhance Palomino's expertise in the space of electronic health records.Two challenges need to be addressed to achieve a sufficiently accurate system that extracts structured data fromunstructured text of radiology data. First, different types of errors in the text arising from the speech-to-textconversion need to be identified and corrected. Errors could be spelling mistakes (e.g., 'abnormaliti' as opposedto 'abnormality'), missing space or word segmentation errors (e.g., 'noevidence' as opposed to 'no evidence') orsymbol errors (e.g., 'intra abdominal' as opposed to 'intra-abdominal'). Second, entities (medical conditions)and their values (e.g. normal vs. abnormal) must be extracted from the corrected text, by reference to ataxonomy of medical conditions, and processing of the text, taking into account the abbreviated nature of thetext which may result in text that does not always contain fully-formed sentences in terms of syntax. Once acorpus of radiology reports has been converted into structured data, in the form of condition-value pairs,classification will be used over a large radiology report corpus to provide insights in population health.In this project, we will first design and implement a system by combining different algorithms to deal withdifferent types of errors in the text, we will then extract the medical conditions and the associated values tooutput the structured data.