In November, I was on a research trip in the U.S., where I participated in three conferences (the AAAI Fall Symposium Virtual Healthcare Interaction (VHI) (together with Hercules Dalianis), November 5-7, Arlington, Virginia, the Third i2b2 Shared Task and Workshop, November 13, San Francisco, and American Medical Informatics Association (AMIA) Annual Symposium, November 14-18, San Francisco). I also visited three departments at different universities, where I gave a seminar describing the work and research I am involved in here at DSV. During these research visits I also met researchers working in the field of Health Informatics. The departments I visited were the following: Department of Medical Informatics & Clinical Epidemiology, Oregon Health & Science University, Portland (09 November), School of Health Information Science, University of Victoria, Kanada (12 November) and Institute for Health Informatics, University of Minnesota, Minneapolis (19 November).

Notes and comments from the AAAI VHI symposium have been posted by Hercules Dalianis here.

I have also written some notes (in Swedish) from the rest of my trip. They can be found here.

Sumithra Velupillai, Hercules Dalianis and Martin Hassel attended ISHIMR 2009: The 14th International Symposium for Health Information Management Research in Kalmar, Sweden, September 14-16.

At ISHIMR 2009 Sumithra presented the paper The Stockholm EPR Corpus – Characteristics and Some Initial Findings, co-written with Hercules and Martin. In this paper, The Stockholm EPR Corpus is described and compared to a general language corpus in Swedish (PAROLE), as well as a description and discussion of four initiated experiments applied on the Stockholm EPR Corpus: hypothesis generation, detection of uncertain and certain expressions, automatic diagnose code suggestion and automatic synonym (or rather “associonym”) generation. The work is part of the KEA-project. The paper can be found here:

http://people.dsv.su.se/~sumithra/publications/ISHIMR/Stockholm-EPR-corpus-ishimr2009.pdf

Sumithra was awarded the best paper award after a successful presentation!

The conference gathered people from ten different countries, and many valuable contacts were established.

Some photos:

Martin and Sumithra in discussion at opening reception:

Spellbound audience during guided poster session:

Hercules at conference dinner at the Kalmar castle:

Sumithra presenting the paper:

Sumithra with the conference organizers Peter Bath and Göran Petersson, receiving award for best paper:

I will defend my licentiate thesis “Swedish Health Data – Information
Access and Representation” on October 6th, at the Department of Computer and Systems Sciences, Stockholm University.

Welcome!

Details:

Swedish Health Data – Information Access and Representation

Time: Tuesday, October 6th 2009 at 10.00
Room: Sal C, floor 4, Department of Computer and Systems Sciences,
Forum, Kista

Directions: http://dsv.su.se/en/about/contact/find

Respondent: Sumithra Velupillai
Opponent: Dr. Dimitrios Kokkinakis, Gothenburg University
Examiner: Prof. Louise Yngström, DSV, Stockholm University

Supervisors: Assoc. Prof. Hercules Dalianis and Dr. Martin Hassel, DSV, Stockholm University

Abstract

Health related research is an interdisciplinary, broad and growing research area. With the growth of digitalised systems that simplify and make work processes more efficient in many companies and organisations, the amount of available data is now immense. The information contained in health related digital data sets could be used for further research and also, in the long run, for improving health care, health care processes, and public health.

A large amount of the information contained in these data sets is often in unstructured, free text. Health related texts can comprise various types of text, such as scientific articles, questionnaire answers, (electronic) health records, information on web sites, and e-mail. What these texts all have in common is above all the use of a domain-specific vocabulary.

Information access methods applied to textual data require a language model. Many human language technology tools have been developed in order to improve and simplify representation models, primarily for English, and predominantly for general language use. For Swedish, several human language technology tools have been developed. How these tools work on domain-specific data such as health data, is still a relatively unchartered research area. We have investigated what properties the language use in Swedish electronic health records have compared to a large, general-purpose Swedish corpus, in order to identify if and where adaptation is necessary. We have also created a representation model based on phrases instead of words for Swedish scientific medical text.

Health related texts also contain a potentially large amount of previously unknown information, which could be valuable to exploit in further research. We have developed an iterative and interactive method for exploring large text sets, based on document clustering, where both structured and unstructured information is used for generating hypotheses from epidemiological questionnaire data and electronic health records.

One of the most important factors that influence the possibilities of performing research on health related data sets is availability. Although digital information is easy to store and obtain automatically, this type of data often contains sensitive and private information that makes it impossible to distribute for further research, unless identifiable information is deleted or replaced. We have initiated work on automatic de-identification for Swedish and created a manually annotated gold standard, which could be used both for evaluating de-identification systems as well as for training new systems.

The paper Developing a standard for de-identifying electronic patient records written in Swedish: precision, recall and F-measure in a manual and computerized annotation trial, which I’ve written together with Hercules Dalianis, Martin Hassel (both here at DSV) and Gunnar H Nilsson (at the Department of Neurobiology, Care Sciences and Society, Center for Family Medicine, Stockholm) has been accepted to the International Journal of Medical Informatics special issue on Mining of Clinical and Biomedical Text and Data. The work described in the paper has been part of the KEA project.

Abstract:

Background: Electronic patient records (EPRs) contain a large amount of information written in free text. This information is considered very valuable for research but is also very sensitive since the free text parts may contain information that could reveal the identity of a patient. Therefore, methods for de-identifying EPRs are needed. The work presented here aims to perform a manual and automatic Patient Health Information (PHI)-annotation trial for EPRs written in Swedish.

Methods: This study consists of two main parts: the initial creation of a manually PHI-annotated gold standard, and the porting and evaluation of an existing de-identification software written for American English to Swedish in a preliminary automatic de-identification trial. Results are measured with precision, recall and F-measure.

Results: This study reports fairly high Inter-Annotator Agreement (IAA) results on the manually created gold standard, especially for specific tags such as names. The average IAA over all tags was 0.65 F-measure (0.84 F-measure highest pairwise agreement). For name tags the average IAA was 0.80 F-measure (0.91 F-measure highest pairwise agreement). Porting a de-identification software written for American English to Swedish directly was unfortunately non-trivial, yielding poor results.

Conclusion: Developing gold standard sets as well as automatic systems for de-identification tasks in Swedish is feasible. However, discussions and definitions on identifiable information is needed, as well as further developments both on the tag sets and the annotation guidelines, in order to get a reliable gold standard. A completely new de-identification software needs to be developed.

Sumithra Velupillai attended the conference Using Corpora in Contrastive and Translation Studies (UCCTS) in Hangzhou, China, September 25-27. She presented the paper Automatic Dictionary Construction and Identification of Parallel Text Pairs which was written together with Martin Hassel and Hercules Dalianis. The experiments described in the paper have been part of the TvärSök project. We have looked at different methods for identifying whether a text pair is parallel or not. Such methods are important and useful for creating parallel corpora, which can be used for dictionary construction, machine translation, cross-lingual information retrieval, etc.

The full paper can be found here.

All the conference participants:

Sumithra Velupillai and Hercules Dalianis attended LOUHI ’08: The First Conference on Text and Data Mining of Clinical Documents in Turku, Finland, September 3-4.

At Louhi ’08 Sumithra and Hercules presented the work-in-progress proposal paper Diagnosing Diagnoses in Swedish Clinical Records. In this paper, planned text mining research on electronic clinical records written in Swedish are described. The work is part of the KEA-project and the proposed work will start this autumn. The paper can be found here:

http://people.dsv.su.se/~sumithra/publications/Louhi/Velupillai_Dalianis_Hassel_WOPP_Louhi08.pdf

The conference gathered people from all over the world working on similar research projects, and many valuable contacts were established. Hercules and Sumithra were proposed to continue the conference and organize it in Sweden next time. We were honoured and accepted of course to carry out the next conference in Sweden.

Some photos:

Sumithra and Hercules presenting their Work in Progress Proposal:

Hercules with Julia Medori and Cristiana Larizza:

Sumithra with Hanna Suominen and Cristiana Larizza at the poster session:

Sumithra Velupillai attended the Twelfth Conference on Computational Natural Language Learning (CoNLL) and the Workshop MMIES-2: Multi-source, Multilingual Information Extraction and Summarization (MMIES2), both held in conjunction with The 22nd International Conference on Computational Linguistics (Coling) in Manchester, August 16 – 23. Coling is one of the most important conferences in Computational Linguistics, with around 500 attendants.

At CoNLL Sumithra presented the article Mixing and Blending Syntactic and Semantic Dependencies together with her colleagues Yvonne Samuelsson, Oscar Täckström and Mark Fishel during a poster session at the conference. The work was the result of joining the shared task Joint Learning of Syntactic and Semantic Dependencies within the course Machine Learning 2 organized by GSLT. The group ranked 6 out of 20! The full article can be found here:

http://people.dsv.su.se/~sumithra/publications/CoNLL/CONLL_Mixing_Blending.pdf

At MMIES2 Sumithra presented the article Automatic Construction of Domain-specific Dictionaries on Sparse Parallel Corpora in the Nordic languages which was co-written with associate professor Hercules Dalianis. In this article Nordic domain-specific dictionaries were created automatically by utilizing multilingual corpora from the web site Hallå Norden and a word alignment approach. The work was part of the research project TvärSök.

The full article can be found here:

http://people.dsv.su.se/~sumithra/publications/MMIES2/Velupillai_Dalianis_MMIES2_ack.pdf

Some photos:

Sumithra at a demo session:

The opening reception, with Manchester’s mayor in the middle:

Sumithra Velupillai and Magnus Rosell (CSC/KTH) presented their article Revealing Relations between Open and Closed Answers in Questionnaires at the the Sixth International Conference on Language Resources and Evaluation, LREC 2008, Marrakech, Morocco, May 28-30, 2008. LREC is one of the major and largest conferences in Computational Linguistics, with around 1000 attendants from all over the world. The article describes experimental work on a questionnaire, where relations between open answers (regarding occupation) and closed answers (regarding smoking) were (semi-)automatically revealed using text clustering evaluation methods and a vector space visualization tool. Through the proposed method a previously unknown hypothesis (farmers smoke less than the average) was generated and evaluated through a literature study, and the method was proven successful. More experiments will be performed to further confirm its potential.

The full article can be found here:

http://people.dsv.su.se/~sumithra/publications/LREC08/rosellvelupillai08.pdf

Results from the research project TvärSök, (Cross language information retrieval) were also presented and published in form of an article with the title: Experiments to investigate the connection between case distribution and topical relevance of search terms in an information retrieval setting

The full article can be found here:

http://www.dsv.su.se/~hercules/papers/karlgrendalianisjongejan-LREC2008_final.pdf

The authors of the article are associate professor Jussi Karlgren, SICS, Bart Jongejan, CST, University of Copenhagen and associate professor Hercules Dalianis, DSV. The article compares lemmatization and stemming for German and Finnish respectively in a search setting for German and Finnish news article. The conclusions are two: the basic Porter stemmer used for German and Finnish is good enough for a search setting and that Finnish nouns in locative cases are less common in relevant documents. The TvärSök project is supported by VINNOVA and the search engine company Euroling AB that is developing and delivering web and intranet search engine SiteSeeker.