=Paper=
{{Paper
|id=Vol-1613/paper_4
|storemode=property
|title=Extraction and Processing of Rich Semantics from Medical Texts
|pdfUrl=https://ceur-ws.org/Vol-1613/paper_4.pdf
|volume=Vol-1613
|authors=Kerstin Denecke,Yihan Deng,Thierry Declerck
|dblpUrl=https://dblp.org/rec/conf/esws/DeneckeDD16
}}
==Extraction and Processing of Rich Semantics from Medical Texts==
https://ceur-ws.org/Vol-1613/paper_4.pdf
Extraction and Processing of Rich Semantics
from Medical Texts
Kerstin Denecke1 , Yihan Deng2 , and Thierry Declerck3
1
Department of Medical Informatics,
Bern University of Applied Sciences
Höheweg 80, 2501 Biel, Switzerland
kerstin.denecke@bfh.ch
2
Innovation Center Computer Assisted Surgery,
Semmelweissstr 14, 04103 Leipzig, Germany
yihan.deng@medizin.uni-leipzig.de
3
German Research Center for Artificial Intelligence
Stuhlsatzenhausweg 3 66123 Saarbrücken
declerck@dfki.de
Abstract. Important information is captured in medical documents.
To make use of this information and intepret the semantics, technologies
are required for extracting, analysing and interpreting it. As a result,
rich semantics including relations among events, subjectivity or polar-
ity of events, become available. The First Workshop on Extraction and
Processing of Rich Semantics from Medical Texts, is devoted to the tech-
nologies for dealing with clinical documents for medical information gath-
ering and application in knowledge based systems. New approaches for
identifying and analysing rich semantics are presented. In this paper, we
introduce the topic and summarize the workshop contributions.
1 Introduction
Pharmaceutical companies and individual patients exploiting advances in trans-
lational medicine and informational infrastructure are joining clinical interests in
recording detailed patient records. The latter comprise a broad range of clinical
documents including nurse letters, discharge summaries and radiology reports
describing a patients health status, diagnoses, applied procedures and obser-
vations of the health care team. The rich semantics such as facts, experiences,
opinions or information that are hidden in those medical documents could when
extracted automatically - support a broad range of applications including clin-
ical decision support systems. Physicians could learn about the experiences of
their colleagues, get hints to critical events in the treatment of a specific patient
or receive information for improving treatment. Studies on the effectiveness of
clinical treatment could be realised based on the text material. The workshop
on Extraction and Processing of Rich Semantics from Medical Texts (RichMed-
Sem) brings together researchers and their work in this upcoming field. Relevant
topics include on the one hand extraction methods specifically designed for the
2 Kerstin Denecke, Yihan Deng, Thierry Declerck
extraction of rich semantics from medical texts. Further, the representation and
storage of extracted rich semantics for further analysis is an important aspect.
Semantic Web technologies are of particular interest, since they could provide
the means to link this type of medical information with other data sets in a
Linked Data infrastructure. What makes rich semantics, what are the challenges
to be addressed and which approaches are suggested? The work presented in the
RichMedSem workshop addresses these questions and answers are summarized
in the following.
2 Rich Semantics
With rich semantics, we refer to concepts and their relations and characteris-
tics described in written text. Existing methods for information extraction from
clinical texts addressed the extraction of mentions of diagnoses, clinical treat-
ments or medications mainly for the purpose of clinical coding, detection of drug
interactions or contraindications. Extraction or addition of rich semantics goes
beyond this. Rich semantics can include descriptions of clinical events, relations
among clinical events (e.g. causality relations), but also subjectivity, polarity,
emotion or even comparison, for example
– A change in the health status (e.g., a patient can suddenly feel better or
worse),
– Critical events, unexpected situations or specific medical conditions that
impact the patient’s life (e.g., tumour is malignant as such is a fact, but this
medical condition is negative for the patient since it might lead to health
problems or death),
– The outcome or effectiveness of a treatment (e.g., a surgery can be success-
fully completed),
– Experiences or opinions towards a treatment or a sort of drug (e.g., a patient
or a physician can describe serious adverse events after drug consumption),
– The certainty of a diagnosis (e.g., a physician can be certain of some diag-
nosis).
3 Challenges of Extracting Rich Semantics
One big challenge for extracting rich semantics from clinical texts are language
peculiarities, content diversity, streaming nature of clinical documents that pose
many challenges to an automatic processing, Finding the trade-off of filtering
noise at the cost of losing potentially relevant information is crucial. In contrast
to biomedical documents, clinical documents are often not well formulated. They
can for example contain verbless clauses, writing errors, many idiosyncratic ab-
breviations and sentence complexity of such document ranges from few word
phrases to complex sentence structures. New technologies are required for deal-
ing with the peculiarities of clinical documents, in particular for extracting,
Extraction and Processing of Rich Semantics from Medical Texts 3
analysing or adding semantics, which can be included into corresponding knowl-
edge based systems. The challenges of extracting rich semantics stem from the
ultimate ambiguities caused by the objective nature of the medical text. The
medical knowledge bases have also a deep influence on the outcome of the se-
mantic meanings. For instance, events such as a bleeding can be positive or neg-
ative, critical or less critical. The phrase blood pressure decreased could express
a positive or negative change depending on the previous state of blood pressure.
A decrease of blood pressure can be good when it was too high before. This also
shows that sentiment in clinical narratives cannot always be manifested in single
terms of phrases, but the context is important.Moreover, the medical seman-
tics include a large amount of aspects, a coherent definition and standard data
schema should be established to facilitate the linking and the usage of current
available ontologies in the biomedical domain.
4 Summary of the Workshop
The papers accepted in the RichMedSem workshop this year have focused on
the extraction, retrieval of biomedical semantics and corpus generation in the
biomedical domain, which covered the foundation of text analysis in the biomed-
ical domain. Shafahi et.al [1] present a controlled experiment with the task of
clinical guideline updating. Text-based and semantics-based methods are eval-
uated with the corpus obtained from the PubMed query service. The proposed
methods simulated the updating process of the Dutch national guidelines of
breast cancer from 2004 to 2012. New existing research literatures was used to
generate updates. The approach has focused on the retrieval of new evidences
at linguistic and semantic level in a specific domain. Natural language process-
ing and concept based methods were used. Deng and Denecke [2] describe the
creation of a corpus for the task of biomedical sentiment analysis. The senti-
ment analysis in the biomedical domain is different to the sentiment analysis
in the general domain. In the biomedical domain, the polarity of the patient
status is related to the clinical events. The paper describes the generation of a
corpus with 300 intensive care unit nurse letters and the corresponding anno-
tation guidelines for biomedical sentiment. Zhukova et.al [3] introduce a system
architecture to process medical documents in Russian, stressing the necessity to
process unstructured documents in the medical domain and to transform them
in structured information that can be managed by computers. Schmidt et.al [4]
present a system that allows users to perform faceted navigation over a large
corpus of medical documents. These documents concern either clinical research
or patient records. The case presented here focusses on the domain of nephrology
and makes use of large database of patients’ records. Anna Kolliakou describe
in her invited talk Social media platforms and clinical records: Trend detection
and intervention a very interesting aspect of interrelating medical information
detected in social media and the information included in patient records. The
aim is to enable healthcare professionals and policy makers to evaluate online
information for emerging rumours and other health-related issues. Findings can
4 Kerstin Denecke, Yihan Deng, Thierry Declerck
then be used to (i) to develop education materials for service users and the gen-
eral public and (ii) link to analysis of electronic health records (EHR). Analyses
of clinical data utilised in clinical record resources.
5 Concluding remarks
There is an increased awareness that rich semantics such as sentiments, opinions
and other qualitative factors are relevant in ensuring individualized care. New
research topics are coming up (e.g. sentiment analysis from clinical texts). Rich
semantics can be in the future be used in clinical decision support systems, with
particular support of semantic web technologies. There is still a huge potential to
connect semantic web technologies with extraction and storage of rich semantics
from biomedical texts. Clinical decision support can only be realised when the
amounts of unstructured texts can be processed, on the one hand to establish
databases with rich semantics; on the other hand to extract decision relevant
information for specific patients targeting at improving treatment.
References
1. Mohammad Shafahi, Qing Hu, Hamideh Afsarmanesh, Zhisheng Huang, Annette
ten Teije, and Frank van Harmelen. A task-based comparison of linguistic and
semantic document retrieval methods in the medical domain. In Proceedings of 1st
Workshop International Workshop on Extraction and Processing of Rich Semantics
from Medical Texts at ECSW, May 29, 2016 to June 2, 2016 in Anissaras, Crete,
Greece, 2016.
2. Yihan Deng and Kerstin Denecke. The generation of a corpus for biomedical senti-
ment analysis. In Proceedings of 1st Workshop International Workshop on Extrac-
tion and Processing of Rich Semantics from Medical Texts at ECSW, May 29, 2016
to June 2, 2016 in Anissaras, Crete, Greece, 2016.
3. Nataly Zhukova, Mikhail Lushnov, Timur Safin, and Maksim Lapaev. Medical text
processing for smda project. In Proceedings of 1st Workshop International Workshop
on Extraction and Processing of Rich Semantics from Medical Texts at ECSW, May
29, 2016 to June 2, 2016 in Anissaras, Crete, Greece, 2016.
4. Danilo Schmidt, Hans-Juergen Profitlich, and Daniel Sonntag. Integrated informa-
tion extraction and facetted search applications in nephrology. In Proceedings of 1st
Workshop International Workshop on Extraction and Processing of Rich Semantics
from Medical Texts at ECSW, May 29, 2016 to June 2, 2016 in Anissaras, Crete,
Greece, 2016.