10,000 foot view of what I am working on

Post on 01-Feb-2016

42 views 1 download

Tags:

description

10,000 foot view of what I am working on. Wendy W. Chapman, PhD. Biomedical Language Understanding. University of Pittsburgh. Dept of Biomedical Informatics. Background. U of Utah. Wisconsin. U of Utah. U of Pittsburgh. 1992. 1994. 2000. 2003. BA Linguistics. Post-doc BMI. - PowerPoint PPT Presentation

transcript

10,000 foot view of what I am working on

Wendy W. Chapman, PhD

University of Pittsburgh

Dept of Biomedical Informatics

Biomedical Language Understanding

BackgroundB

A L

ingu

istic

sB

A L

ingu

istic

s

Chi

nese

Lite

ratu

reC

hine

se L

itera

ture

1992 1994

PhD

Med

ical

Info

rmat

ics

PhD

Med

ical

Info

rmat

ics

Pos

t-do

c B

MI

Pos

t-do

c B

MI

Fac

ulty

DB

MI

Fac

ulty

DB

MI

20032000

U of UtahU of Utah WisconsinWisconsin U of UtahU of Utah U of PittsburghU of Pittsburgh

Biomedical Language Understanding

www.dbmi.pitt.edu/blulabHenk Harkema, Danielle Mowery, Mike Conway,Lee Christensen, Qi Li, Wendy Chapman

Temporality Schema

Topaz

NLP Repository

AnaphoricReference

Onyx

BLU Lab NLP Sampler

OntologyEnrichment

ResultsReview/Error

Analysis

OntologyFor

SyndromicSurveillance

Temporality Schema

Topaz

NLP Repository

AnaphoricReference

Onyx

BLU Lab NLP Sampler

OntologyEnrichment

ResultsReview/Error

Analysis

OntologyFor

SyndromicSurveillance

Topaz

• Named entity recognition– Maps UMLS concepts to higher-level concepts

• Contextual property assignment (ConText)– Existence (affirmed, negated)– Experiencer (patient, other)– Historicity (current, historical)– Realis (actual, non-specific/hypothetical)– Certainty (uncertain, certain)– Reason for exam (yes, no)– Quality of exam (diagnostic, limited)

Harkema, B Chapman, Hwa

ConText: Determine Values for Contextual Properties

Patient denies cough but complains of headache.No change in the patient’s chest pain.

trigger term

terminationtermpseudo-trigger

term

scope

Clinical condition: CoughNegation: Negated

ConText: Historical

Past history of pneumonia presentingtoday with cough and fever.

trigger term terminationterm

scope

Clinical condition: PneumoniaTemporality: Historical

Temporality Schema

Topaz

NLP Repository

AnaphoricReference

Onyx

BLU Lab NLP Sampler

OntologyEnrichment

ResultsReview/Error

Analysis

OntologyFor

SyndromicSurveillance

Onyx

Onyx

At (translucency, numberEight) &

surfaceOf (numberEight, mesial) &

stateOf (translucency, possible)

Semantic Models

Semantic Models

Syntactic AnalyzerSyntactic Analyzer

Context-free Grammar

Context-free Grammar

TrainingCorpusTrainingCorpus

Semantic AnalyzerSemantic Analyzer

Eight mesial might have a slight translucency

Haug, Schleyer

Knowledge-rich Frame-based Mapping

ProbabilisticFrames

SemanticNetwork

- Frame slots map to semantic network- Relationships between slots are probabilistic

Annotation Interfacewith active learning and help from Onyx

TemplatesTemplates

Semantic ModelSemantic Model

Speech NLP Chart

Onyx

Dental ExamsNumber one Is missing. Two is fine. Caries on Tooth 3.

Titus Schleyer, Lee Christensen, Peter Haug, Jeannie Irwin, Henk Harkema

Temporality Schema

Topaz

NLP Repository

AnaphoricReference

Onyx

BLU Lab NLP Sampler

OntologyEnrichment

ResultsReview/Error

Analysis

OntologyFor

SyndromicSurveillance

Ontology Development-Information Extraction (ODIE)

Ontology

Text

Ontology EnrichmentUse IE to find new concepts and relationships to add

Information ExtractionUse ontology to improve IE from text

Rebecca Crowley, Mayo Clinic, Stanford NCBO

SurgicalPathologySurgical

PathologyChest

RadiographyChest

Radiography

View Overlap of Ontologies

Suggest Concepts

Temporality Schema

Topaz

NLP Repository

AnaphoricReference

Onyx

BLU Lab NLP Sampler

OntologyEnrichment

ResultsReview/Error

Analysis

OntologyFor

SyndromicSurveillance

Results Review/Error Analysis

Temporality Schema

Topaz

NLP Repository

AnaphoricReference

Onyx

BLU Lab NLP Sampler

OntologyEnrichment

ResultsReview/Error

Analysis

OntologyFor

SyndromicSurveillance

Schema for Clinical Condition Properties

Properties of Condition Concept

Existence Yes, NoExperiencer Patient, OtherChange Unmarked, Unchanging, Changing, Increasing, Decreasing,

Improving, Worsening, RecurrenceIntermittent Unmarked, Yes, NoCertainty Unmarked, High, Moderate, LowMental State Yes, NoGeneralized/Conditional Yes, NoCurrent Visit Relation Before, Meets_Overlaps, After

Wiebe, Jordan, Mowery, Harkema

Schema for Temporal Relations

Time Words Points, DurationsOrdering Words Precedes, During, FollowsAspectual Words Initiation, Continuation, Culmination

Temporality Schema

Topaz

NLP Repository

AnaphoricReference

Onyx

BLU Lab NLP Sampler

OntologyEnrichment

ResultsReview/Error

Analysis

OntologyFor

SyndromicSurveillance

Temporality Schema

Topaz

NLP Repository

AnaphoricReference

Onyx

BLU Lab NLP Sampler

OntologyEnrichment

ResultsReview/Error

Analysis

OntologyFor

SyndromicSurveillance

Application Ontology for Syndromic Surveillance

Consensus of developers/users across countryConway, Buckeridge

Temporality Schema

Topaz

NLP Repository

AnaphoricReference

Onyx

BLU Lab NLP Sampler

OntologyEnrichment

ResultsReview/Error

Analysis

OntologyFor

SyndromicSurveillance

Anaphoric Reference in Clinical ReportsCrowey, Savova, Zeng

• Adapted MUC schema for clinical reports• Three experts annotated 180 reports

Five types—Mayo, UPMC

• identity• part/whole• set/subset

• Characterize anaphoric reference in reports• Train/test resolution algorithms