Runs - Medical 2011¶

AEHRC1¶

Run ID: AEHRC1
Participant: AEHRC
Track: Medical
Year: 2011
Submission: 8/8/2011
Type: automatic
Task: main
MD5: 7e12eafb1b9902a1cbbd3adf8e16886c
Run description: Combined SNOMED-concept and term based run.

AEHRC2¶

Run ID: AEHRC2
Participant: AEHRC
Track: Medical
Year: 2011
Submission: 8/8/2011
Type: automatic
Task: main
MD5: 491de2d9180daa6ed57334d8a90e8aa6
Run description: SNOMED concepts search. boths queries and documents transformed to SNOMED concepts.

baselucene¶

Results | Participants | Input | Summary | Appendix

Run ID: baselucene
Participant: DUTCHHATTRICK
Track: Medical
Year: 2011
Submission: 8/15/2011
Type: automatic
Task: main
MD5: 75b9190f9f4ab0634ffd55eb5672b79c
Run description: Preprocessing of the corpus consisted of automatic spelling correction, section and sentence splitting, and applying the ConText algorithm to remove text that was either negated, about the history of the patient, or about some other than the patient. Frequently occurring sections that were deemed uninformative (e.g. 'Family History') were removed. This is a baseline run using Lucene without query expansion. First, a match score per report was calculated. Visit scores were calculated as the maximum report score per visit.

BiTeMbase¶

Run ID: BiTeMbase
Participant: BiTeM
Track: Medical
Year: 2011
Submission: 8/4/2011
Type: automatic
Task: main
MD5: b57eecc88ad59decbc8cf8a317bc0a83
Run description: The baseline run, with only free-text features, and just a couple of pre processing techniques.

BiTeMmEsh¶

Run ID: BiTeMmEsh
Participant: BiTeM
Track: Medical
Year: 2011
Submission: 8/4/2011
Type: automatic
Task: main
MD5: bb7191730f10f528b63efffff85a623f
Run description: The MeSH run was computed with a MeSH concepts oriented document representation, both applied to queries and documents. The computed run was then combined with the previous baseline run.

BiTeMmhICD¶

Run ID: BiTeMmhICD
Participant: BiTeM
Track: Medical
Year: 2011
Submission: 8/4/2011
Type: automatic
Task: main
MD5: ba2ac02268114baad243ea3d54117be0
Run description: This mh&ICD run was computed using the mEsh run (combination of a free-text and a MeSH concepts document representation). For this presumed most achieved run, we used the MeSH diseases mapped in queries, we found the associate ICD9 code with the UMLS identifiers, then we boosted documents that contained this code in their discharge diagnosis.

BiTeMsnomed¶

Run ID: BiTeMsnomed
Participant: BiTeM
Track: Medical
Year: 2011
Submission: 8/4/2011
Type: automatic
Task: main
MD5: ccdecf293daf6d0fe6cb939d5436d24e
Run description: The MeSH run was computed with a SNOMED concepts oriented document representation, both applied to queries and documents. The computed run was then combined with the previous baseline run.

buptpris01¶

Run ID: buptpris01
Participant: BUPT_WILDCAT
Track: Medical
Year: 2011
Submission: 8/9/2011
Type: manual
Task: main
MD5: c63edc015b83f112db6ece5f986dc17b
Run description: We produce the results with our own data mining algorithm.The ranking strategy is based on LSA external resources : Xapian, Indri

buptpris02¶

Run ID: buptpris02
Participant: BUPT_WILDCAT
Track: Medical
Year: 2011
Submission: 8/9/2011
Type: manual
Task: main
MD5: 3451e8ecfbc74e4d0a8d2656ff816236
Run description: run2 We produce the results with our own data mining algorithm.The ranking strategy is based on LSA external resources : Xapian, Indri

CengageM11R1¶

Run ID: CengageM11R1
Participant: Cengage
Track: Medical
Year: 2011
Submission: 8/15/2011
Type: automatic
Task: main
MD5: fda2e2d2331418f3f687f949fcaf5e55
Run description: Report Text + UMLS Terms + Filters External resources: UMLS Lucene GATE

CengageM11R2¶

Run ID: CengageM11R2
Participant: Cengage
Track: Medical
Year: 2011
Submission: 8/15/2011
Type: automatic
Task: main
MD5: 61632200a00e323d379b14095c1264a5
Run description: Report Text + UMLS Terms + UMLS Database Related Terms + Filters External Resources: UMLS Lucene GATE

CengageM11R3¶

Run ID: CengageM11R3
Participant: Cengage
Track: Medical
Year: 2011
Submission: 8/15/2011
Type: automatic
Task: main
MD5: 65b48598c664af9db579aaa468bd18b5
Run description: Report Text + UMLS Terms + UMLS Related Terms + Gale Virtual Library Reference Medical Encyclopedia Expansion + Filters External Resources: UMLS Lucene GATE Cengage GVRL

CengageM11R4¶

Run ID: CengageM11R4
Participant: Cengage
Track: Medical
Year: 2011
Submission: 8/15/2011
Type: automatic
Task: main
MD5: a8ed8acd730f9e31c9ebe12ff4bde31f
Run description: Report Text + UMLS Terms + UMLS Network Query Expansion + Filters External Resources: UMLS Lucene GATE Lexical Tools Neo4j

csirTC1¶

Results | Participants | Input | Summary | Appendix

Run ID: csirTC1
Participant: CSIRO
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: ccd8fa6fae47b0410953a94bc958d615
Run description: TopSig plaintext run, up to 1000 results per topic

csirTC2¶

Results | Participants | Input | Summary | Appendix

Run ID: csirTC2
Participant: CSIRO
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: 7157594597910105464b55da1c1884f8
Run description: TopSig run with MetaMap concept mappings added.

CWI1¶

Results | Participants | Input | Summary | Appendix

Run ID: CWI1
Participant: CWI
Track: Medical
Year: 2011
Submission: 8/15/2011
Type: automatic
Task: main
MD5: 39a49f0b5ac5c2b8b95faed029b11dc2
Run description: text retrieval with BM25

CWI2¶

Results | Participants | Input | Summary | Appendix

Run ID: CWI2
Participant: CWI
Track: Medical
Year: 2011
Submission: 8/15/2011
Type: automatic
Task: main
MD5: 9bc72d8cdda297aaef4f7fcf7f6a078d
Run description: - \alpha(text retrieval with BM25) + (1-\alpha)rank docs with ICD codes - ICD codes are retrieved by original query using BM25

CWI3¶

Results | Participants | Input | Summary | Appendix

Run ID: CWI3
Participant: CWI
Track: Medical
Year: 2011
Submission: 8/15/2011
Type: automatic
Task: main
MD5: 51bf3c876191b00cf920078cf29ba34d
Run description: - query expansion with ICD descriptions - (1-\alpha)expanded query + \alpha original query - BM25 for retrieval

CWI4¶

Results | Participants | Input | Summary | Appendix

Run ID: CWI4
Participant: CWI
Track: Medical
Year: 2011
Submission: 8/15/2011
Type: automatic
Task: main
MD5: 72df03dd6461b7e98f9860a4f76738f5
Run description: - find top X ICD descriptions with original query - for each ICS descriptions extract top N terms - Using each N terms to generate a ranked list, aggregated into one - Linearly Combine the ranked list using original query and the ranked list based on ICD terms

DEMIR1BASE¶

Run ID: DEMIR1BASE
Participant: DEUCENG
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: 6e56e155042e98f20fa2e513dc0c1f9a
Run description: Used DFR_BM25 weighting model,removal stopwords,used porterstemmer.It is our baseline.

DEMIR2TW¶

Run ID: DEMIR2TW
Participant: DEUCENG
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: fe78f5e7e00a1c755a78f5eedfd45e0f
Run description: Used DFR_BM25 weighting model,process with lexical tools,using term that it has semantic type

DEMIR4TWP¶

Run ID: DEMIR4TWP
Participant: DEUCENG
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: 14e3709fc0a98d14ae55fddaffea471f
Run description: Used DFR_BM25 weighting model,process with lexical tools,using term and term phrase that it has semantic type

emc1¶

Results | Participants | Input | Summary | Appendix

Run ID: emc1
Participant: emc
Track: Medical
Year: 2011
Submission: 8/16/2011
Type: automatic
Task: main
MD5: 2037d1604202a8932f694ee19527e697
Run description: Concept based recognition. Fully automatic system. External resource used: http://code.google.com/p/negex/

emc2¶

Results | Participants | Input | Summary | Appendix

Run ID: emc2
Participant: emc
Track: Medical
Year: 2011
Submission: 8/16/2011
Type: automatic
Task: main
MD5: 1dea9a6d30574323240a71ec235b1cc1
Run description: Automatic method. Overlap calculated from ACCCA, concept recognition and chunking modules. External resources: http://code.google.com/p/negex/ http://www.ncbi.nlm.nih.gov/books/NBK3827/?rendertype=table&id=pubmedhelp.T43 http://pages.cs.wisc.edu/~bsettles/abner/ http://incubator.apache.org/opennlp/ http://alias-i.com/lingpipe http://mmtx.nlm.nih.gov/ http://nlp.stanford.edu/software/CRF-NER.shtml

emc3¶

Results | Participants | Input | Summary | Appendix

Run ID: emc3
Participant: emc
Track: Medical
Year: 2011
Submission: 8/16/2011
Type: automatic
Task: main
MD5: 8361b450aa7b97c5035201966586b307
Run description: Automatic run. ACCCA system comparison. External resources: http://pages.cs.wisc.edu/~bsettles/abner/ http://alias-i.com/lingpipe http://incubator.apache.org/opennlp/ http://nlp.stanford.edu/software/CRF-NER.shtml

emc4¶

Results | Participants | Input | Summary | Appendix

Run ID: emc4
Participant: emc
Track: Medical
Year: 2011
Submission: 8/16/2011
Type: automatic
Task: main
MD5: b39787db6a254e7c8e24e201f2d25d79
Run description: Automatic run. Chunking comparison using noun phrases. External resources: http://incubator.apache.org/opennlp/

EssieAuto¶

Results | Participants | Input | Summary | Appendix

Run ID: EssieAuto
Participant: LHC
Track: Medical
Year: 2011
Submission: 8/16/2011
Type: automatic
Task: main
MD5: 7ae2925c209a48a7b461a4dc2b02c86d
Run description: Automatic query formulation using syntactic-semantic PICO frames. External resources: UMLS, MeSH, Google.

explucene¶

Results | Participants | Input | Summary | Appendix

Run ID: explucene
Participant: DUTCHHATTRICK
Track: Medical
Year: 2011
Submission: 8/15/2011
Type: automatic
Task: main
MD5: 8ade91e3c2858c86134e5755b1e9fc6a
Run description: Preprocessing of the corpus consisted of automatic spelling correction, section and sentence splitting, and applying the ConText algorithm to remove text that was either negated, about the history of the patient, or about some other than the patient. Frequently occurring sections that were deemed uninformative (e.g. 'Family History') were removed. Query expansion was done using Wikipedia: Text of the queries was used to identify corresponding Wikipedia pages, where queries were also used for disambiguation. Wikipedia page names, redirects, and link tag texts were used to identify synonyms. A combination of UMLS and Drugbank was used to expand drug classes (e.g. 'atypical antipsychotics') to individual drugs. This is a run using the Lucene search engine with query expansion. First, a match score per report was calculated. Visit scores were calculated as the maximum report score per vist. External sources used: Wikipedia, Drugbank, UMLS, ConText algorithm

INDRI¶

Results | Participants | Input | Summary | Appendix

Run ID: INDRI
Participant: KobeU
Track: Medical
Year: 2011
Submission: 8/13/2011
Type: automatic
Task: main
MD5: 1e64cef38037511755afad35e4d1539e
Run description: Retrieved by only indri.

IRITa1¶

Run ID: IRITa1
Participant: IRIT
Track: Medical
Year: 2011
Submission: 7/28/2011
Type: automatic
Task: main
MD5: 93ec77289df0362fcd064b677d86ba07
Run description: In_expB2 term weighting model with c=5.0, no query expansion

IRITa1QE1¶

Run ID: IRITa1QE1
Participant: IRIT
Track: Medical
Year: 2011
Submission: 7/28/2011
Type: automatic
Task: main
MD5: 2405536812db7e4d9654affe92fae30e
Run description: In_expB2 model (c=5.0) Bo1 QE model (20 terms, 20 top ranked documents)

IRITm1¶

Run ID: IRITm1
Participant: IRIT
Track: Medical
Year: 2011
Submission: 7/28/2011
Type: automatic
Task: main
MD5: 72f440295d0cca39966309cf5ff6d4c4
Run description: Automatic indexing, manual query removal of less significant words Term weighting model: In_expB2 (c=5.0)

IRITm1QE1¶

Run ID: IRITm1QE1
Participant: IRIT
Track: Medical
Year: 2011
Submission: 7/28/2011
Type: manual
Task: main
MD5: ba83e21884ebf1871a95dacd54476aca
Run description: Automatic indexing, manual query removal of less significant words Term weighting model: In_expB2 (c=5.0) Query expansion model: Bo1 (terms = 20, docs=20)

mayo2noprop¶

Run ID: mayo2noprop
Participant: MayoClinicNLP
Track: Medical
Year: 2011
Submission: 9/15/2011
Type: manual
Task: main
MD5: 995533b3c89a9807b3bb5c79eac1964a
Run description: This run uses NLP concept identification tools that compares queries and documents in the form of weighted ontologies.

mayoauto¶

Run ID: mayoauto
Participant: MayoClinicNLP
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: a8309da9d62ddff257db8eb32127c41d
Run description: Using the results of concept identification and dictionary lookup tools, creates an empirically-weighted 'mask' for each topic (automatically) and compares it to a 'profile' for each visit. Empirical weightings benefit from a Mayo Clinic internal corpus of millions of medical records.

mayobaseline¶

Run ID: mayobaseline
Participant: MayoClinicNLP
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: manual
Task: main
MD5: 54f8e813eb66c89bb8e181d21d438e57
Run description: Using the results of concept identification and dictionary lookup tools, creates an empirically-weighted 'mask' for each topic and compares it to a 'profile' for each visit. Empirical weightings benefit from a Mayo Clinic internal corpus of millions of medical records.

mayocooccur¶

Run ID: mayocooccur
Participant: MayoClinicNLP
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: manual
Task: main
MD5: 0bce833d5adc26cf3af2d74cd38c1554
Run description: Using the results of concept identification and dictionary lookup tools, creates an empirically-weighted semantic 'mask' for each topic (including co-occurrence information) and compares it to a semantic 'profile' for each visit. Empirical weightings benefit from a Mayo Clinic internal corpus of millions of medical records.

mayolbra¶

Run ID: mayolbra
Participant: MayoClinicNLP
Track: Medical
Year: 2011
Submission: 9/15/2011
Type: automatic
Task: main
MD5: 1f8981b25449051673adb9610949a4c9
Run description: This run uses NLP concept identification tools that compares queries and documents in the form of weighted ontologies. Co-occurrences stats are based on a large Mayo corpus of clinical reports. The query is automatically processed using string matches to the UMLS.

mayolbrst¶

Run ID: mayolbrst
Participant: MayoClinicNLP
Track: Medical
Year: 2011
Submission: 9/15/2011
Type: manual
Task: main
MD5: 7dd64315fc1ec749b5cca5bb7ce1cf59
Run description: This run is based on NLP concept matching tools that compares queries and documents in the form of weighted ontologies. Co-occurrences were used, based on a large Mayo corpus of clinical reports. The section in which concepts were found and the type of note both played a role in the similarity between visit and query.

mayoubr¶

Run ID: mayoubr
Participant: MayoClinicNLP
Track: Medical
Year: 2011
Submission: 9/15/2011
Type: manual
Task: main
MD5: be73e34b5e644e8d8f840c1b0cebd43a
Run description: This run uses cTAKES, a NLP concept identification tool. It compares queries and documents in the form of weighted ontologies. Co-occurrences stats are based on a large Mayo corpus of clinical reports.

merckkgaamer¶

Run ID: merckkgaamer
Participant: MERCKKGAA
Track: Medical
Year: 2011
Submission: 8/9/2011
Type: manual
Task: main
MD5: 440d26f7883a031a1719dd2db0f59f6d
Run description: The UIMA Skill Cartridge "Medical Entity Recognition" version 2011 (www.luxid.com) has been used to automatically tag the records and the queries. The LUXID user interface (www.luxid.com) has then been used to manually query the tagged output, and to export the documents hits as Excel formatted files. The tool Knime (www.knime.org) has been used to combine the results of the 35 runs, and generate the submission file. The tool we use has a cutoff that prevents low scoring hits to be displayed, therefore most of our runs have less than 1000 hits.

newton¶

Results | Participants | Input | Summary | Appendix

Run ID: newton
Participant: Michigan
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: f0252390894d2b9233672c56e39ba47e
Run description: Metamap, Lemur.

newtonA¶

Results | Participants | Input | Summary | Appendix

Run ID: newtonA
Participant: Michigan
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: ef77ba8f1ea009367bdfa52763663371
Run description: Non-pooled results.

newtonB¶

Results | Participants | Input | Summary | Appendix

Run ID: newtonB
Participant: Michigan
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: 22bd91d0f340b8f0e2af3ad1eea30ded
Run description: Non-pooled results.

newtonC¶

Results | Participants | Input | Summary | Appendix

Run ID: newtonC
Participant: Michigan
Track: Medical
Year: 2011
Submission: 9/15/2011
Type: automatic
Task: main
MD5: ee346fd0a94786b8a10cd4cc2efe9c38
Run description: Non-pooled results.

newtonD¶

Results | Participants | Input | Summary | Appendix

Run ID: newtonD
Participant: Michigan
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: 27bab0a9dd1358923c8298324906c593
Run description: Non-pooled results.

NICTA1¶

Run ID: NICTA1
Participant: NICTA_BioTALA
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: 7571fa4e26fbb012999c508a244f944b
Run description: Document processing: All the reports were first processed to extract the sections of interest, such as: gender, age range, history, chief complaint, medications, allergies, admit and discharge diagnoses, and the full text of the report. The ICD9 codes in admit and discharge diagnoses were expanded in the documents, keeping the original code together with the expanded forms of both the code and the heading under which it appears. All related reports of a visit were concatenated to make up one "document" for indexing. The fine-grained data was then indexed using separate fields. Query processing: Run1- Query transformation (QT): Using a flexible pattern matching, queries were automatically processed against a manually constructed set of patterns to both assign their terms to their respected fields, and translate some of their terms to the language of the reports. The patterns were formed based on the sample clinical questions and covered seven broad categories of age, weight (using body mass index), diagnostics, treatments, medications, history, allergies, and abbreviations. For example, if a query contained "elderly patients", we replaced "elderly" with an equivalent age field that covered people in their 60s to 90+. Some of the abbreviations, such as ER (emergency room), were expanded in the queries. For this run, we also included the entire query to be searched over the report text only. Indexing and Searching: The search engine used for indexing and searching in our runs was Apache Lucene (v3.2). We also used the BM25 ranking algorithm for Lucene described in: Joaqun Prez-Iglesias, Jos R. Prez-Agera, Vctor Fresno, Yuval Z. Feinstein. (2009). Integrating the Probabilistic Models BM25/BM25F into Lucene. CoRR. We relied on field search in all the runs, that is a Boolean search followed by ranking. No stemming was done in our experiments.

NICTA2¶

Run ID: NICTA2
Participant: NICTA_BioTALA
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: 9657c6f966a306c0494724aa61e4b364
Run description: Document processing: All the reports were first processed to extract the sections of interest, such as: gender, age range, history, chief complaint, medications, allergies, admit and discharge diagnoses, and the full text of the report. The ICD9 codes in admit and discharge diagnoses were expanded in the documents, keeping the original code together with the expanded forms of both the code and the heading under which it appears. All related reports of a visit were concatenated to make up one "document" for indexing. The fine-grained data was then indexed using separate fields. Query processing: Run2- Query Expansion (QE): queries were first processed using Metamap to identify their medical terms. Phrases that belonged to medication-related semantic types were expanded using DBpedia, searching for synonyms and other terms under that category; phrases related to more general interventions were expanded with synonyms only, and filtered by a MetaMap-based filter. The remaining terms were allocated to their corresponding fields as specified originally by Metamap. We did not weight the expanded and original terms differently. Indexing and Searching: The search engine used for indexing and searching in our runs was Apache Lucene (v3.2). We also used the BM25 ranking algorithm for Lucene described in: Joaqun Prez-Iglesias, Jos R. Prez-Agera, Vctor Fresno, Yuval Z. Feinstein. (2009). Integrating the Probabilistic Models BM25/BM25F into Lucene. CoRR. We relied on field search in all the runs, that is a Boolean search followed by ranking. No stemming was done in our experiments.

NICTA3¶

Run ID: NICTA3
Participant: NICTA_BioTALA
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: 5b89c7668143e463e414abbcc616ccf2
Run description: Document processing: All the reports were first processed to extract the sections of interest, such as: gender, age range, history, chief complaint, medications, allergies, admit and discharge diagnoses, and the full text of the report. The ICD9 codes in admit and discharge diagnoses were expanded in the documents, keeping the original code together with the expanded forms of both the code and the heading under which it appears. All related reports of a visit were concatenated to make up one "document" for indexing. The fine-grained data was then indexed using separate fields. Query processing: Run3- Combined QT and QE: queries were a combination of runs NICTA1 and NICTA2, with duplicate removal. We removed the full-query search in the report field from the QT queries in this run. Indexing and Searching: The search engine used for indexing and searching in our runs was Apache Lucene (v3.2). We also used the BM25 ranking algorithm for Lucene described in: Joaqun Prez-Iglesias, Jos R. Prez-Agera, Vctor Fresno, Yuval Z. Feinstein. (2009). Integrating the Probabilistic Models BM25/BM25F into Lucene. CoRR. We relied on field search in all the runs, that is a Boolean search followed by ranking. No stemming was done in our experiments.

NICTA4¶

Run ID: NICTA4
Participant: NICTA_BioTALA
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: a9468bfd7cb54dbe9071e1de7c1a995a
Run description: Query expansion relying on dbpedia and UMLS. Indexed with Lucene.

NICTA5¶

Run ID: NICTA5
Participant: NICTA_BioTALA
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: 512bf8fb767e4bedb821779eb9d542ca
Run description: Variation of NICTA4. Query transformation and query expansion, report indexing, tf-idf. Fielded indexing.

NICTA6¶

Run ID: NICTA6
Participant: NICTA_BioTALA
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: bfb64b4f509e8957655769f89937c873
Run description: Variation of NICTA4. Query transformation and query expansion, report indexing, tf-idf.

NICTA7¶

Run ID: NICTA7
Participant: NICTA_BioTALA
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: 70762b11a6dc20a262632042062dee54
Run description: Variation of NICTA4. Query expansion, report indexing, tf-idf. No fields.

NLMAutoLuc¶

Results | Participants | Input | Summary | Appendix

Run ID: NLMAutoLuc
Participant: LHC
Track: Medical
Year: 2011
Submission: 8/16/2011
Type: automatic
Task: main
MD5: 209b0aeba376069e0778db697bc5cad9
Run description: Automatic query formulation using syntactic/semantic PICO frames with Lucene document retrieval. External resources: UMLS, MeSH, Google.

NLMLucene¶

Results | Participants | Input | Summary | Appendix

Run ID: NLMLucene
Participant: LHC
Track: Medical
Year: 2011
Submission: 9/13/2011
Type: automatic
Task: main
MD5: 0a4a51739be475bcf7413036f8031c6f
Run description: Automatic retrieval using Lucene based on the topics. No external resources.

NLMLucenePS¶

Results | Participants | Input | Summary | Appendix

Run ID: NLMLucenePS
Participant: LHC
Track: Medical
Year: 2011
Submission: 9/13/2011
Type: automatic
Task: main
MD5: 41c9b66629f4cf37a93a04cfeb78079b
Run description: Automatic retrieval in which only the positive and speculative text from the documents was indexed with Lucene. No external resources.

NLMManual¶

Results | Participants | Input | Summary | Appendix

Run ID: NLMManual
Participant: LHC
Track: Medical
Year: 2011
Submission: 8/16/2011
Type: manual
Task: main
MD5: e416a843219ca2c7565917dd72177eaf
Run description: Manual query formulation using Essie corpus mining tool. External resources: UMLS, MeSH, Google.

NLMManualLuc¶

Results | Participants | Input | Summary | Appendix

Run ID: NLMManualLuc
Participant: LHC
Track: Medical
Year: 2011
Submission: 8/16/2011
Type: manual
Task: main
MD5: a55b84af47f5b8392e6de1520992b77a
Run description: Manual query formulation using Essie corpus mining tool with Lucene document retrieval. External resources: UMLS, MeSH, Google.

nobel¶

Results | Participants | Input | Summary | Appendix

Run ID: nobel
Participant: Michigan
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: 42c73e1e3b8a4ddd14e58b1aaec62594
Run description: this run incorporated query expansion

NORERANK¶

Results | Participants | Input | Summary | Appendix

Run ID: NORERANK
Participant: KobeU
Track: Medical
Year: 2011
Submission: 8/13/2011
Type: automatic
Task: main
MD5: 3b109c79b9160e1927267cdf42feea5c
Run description: Learning to rank result. (Not reranked by sex and age)

NORERANK2¶

Results | Participants | Input | Summary | Appendix

Run ID: NORERANK2
Participant: KobeU
Track: Medical
Year: 2011
Submission: 9/15/2011
Type: automatic
Task: main
MD5: 6cbd242302dc801690bbb4778a4f1b37
Run description: Learning to rank result. (Not reranked by sex and age)

NORERANK2N¶

Results | Participants | Input | Summary | Appendix

Run ID: NORERANK2N
Participant: KobeU
Track: Medical
Year: 2011
Submission: 9/15/2011
Type: automatic
Task: main
MD5: 04ec4df3dcc98fd386b5ff826d84f62c
Run description: Negation Processed. Learning to rank result. (Not reranked by sex and age)

ohsuManAll¶

Run ID: ohsuManAll
Participant: OHSU
Track: Medical
Year: 2011
Submission: 8/16/2011
Type: manual
Task: main
MD5: a05c1663af4021d43d08de651d248bd3
Run description: Fully manual run; Boolean queries with wildcards assembled by a clinician and the system's developer.

ohsuManLim¶

Run ID: ohsuManLim
Participant: OHSU
Track: Medical
Year: 2011
Submission: 8/16/2011
Type: manual
Task: main
MD5: a3441a54b8329e877929ac01290c60f5
Run description: Fully manual run; Boolean queries with wildcards assembled by a clinician and the system's developer. Search limited to ED visit summaries and discharge summaries.

RERANK¶

Results | Participants | Input | Summary | Appendix

Run ID: RERANK
Participant: KobeU
Track: Medical
Year: 2011
Submission: 8/13/2011
Type: automatic
Task: main
MD5: fcfbfa92145a16d5b228dad510df32c2
Run description: Reranked by sex and age.

RERANK2¶

Results | Participants | Input | Summary | Appendix

Run ID: RERANK2
Participant: KobeU
Track: Medical
Year: 2011
Submission: 9/15/2011
Type: automatic
Task: main
MD5: 4d5e08b460db38b77efb935362c52938
Run description: Learning to rank result. Reranked by sex and age.

RERANK2N¶

Results | Participants | Input | Summary | Appendix

Run ID: RERANK2N
Participant: KobeU
Track: Medical
Year: 2011
Submission: 9/15/2011
Type: automatic
Task: main
MD5: 2569bec92f2ca33eacc19d840eabe491
Run description: Negation Processed. Learning to rank result. Reranked by sex and age.

RMIT1¶

Run ID: RMIT1
Participant: RMIT
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: 46cc5d8ce66095d85afa17f84c0dd680
Run description: This run translates queries by identifying key terms that need translation to match the vocabulary of the original reports, such as age and gender, we used PL2 weighting model and remove terms with low idf.

RMIT2¶

Run ID: RMIT2
Participant: RMIT
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: ba92681ebb93e90760aba8be856f0ae9
Run description: This run translates queries, and expands terms using wikipedia only. We Use PL2 for weighting terms and low idf terms are stopped. A simple pattern matching is used to find the terms for translation, and those terms are translated to the language of the reports. Search Engine Used : Terrier 3.5

RMIT3¶

Run ID: RMIT3
Participant: RMIT
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: 1fc475651aad14543e581f4feb9d7cba
Run description: This run extracts key terms, and combines the methods that have been submitted in run1 and run2.

RMITN1¶

Run ID: RMITN1
Participant: RMIT
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: 2f78a1231060766d54ea2dcafca5919a
Run description: This run uses a simple indexer, taking each report as a document and indexing salient fields of each report. Ranking metric used was PL2, advanced model from Divergence-from-Randomness, using Terrier search engine. Automatic query processing and translation was used to expand particular words that need expansion to match the vocabulary of clinical records.

RMITN2¶

Run ID: RMITN2
Participant: RMIT
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: 81b0003eed8df9b0cc77c295edc1a630
Run description: This run uses the same indexer as run1, ranking metric used was the lemur version of TF_IDF using Terrier search engine. UMLS knowledge source was used to expand the queries and translates query term into a language to match clinical records.

RMITN3¶

Run ID: RMITN3
Participant: RMIT
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: 10194158540b5d4a010b34410aa885ee
Run description: This run is similar to the first with an additional query processing, using metamap tool to disambiguate terms and combining the same strategy to translate queries as the second run.

SCAIMED1¶

Run ID: SCAIMED1
Participant: fraunhofer.scai
Track: Medical
Year: 2011
Submission: 8/5/2011
Type: manual
Task: main
MD5: 47d1adf03dd00f3e747f39ce79c3b8fe
Run description: Keyword search in documents

SCAIMED2¶

Run ID: SCAIMED2
Participant: fraunhofer.scai
Track: Medical
Year: 2011
Submission: 8/5/2011
Type: automatic
Task: main
MD5: 27e5f9f22411bde4990367cd2481790e
Run description: search UMLS concepts in documents

SCAIMED3¶

Run ID: SCAIMED3
Participant: fraunhofer.scai
Track: Medical
Year: 2011
Submission: 8/5/2011
Type: automatic
Task: main
MD5: 092e69b6694b8ed4b977633d93982bd4
Run description: search dictionary (MeSH,MedDRA,ATC,DrugBank) concepts in documents + automatic query expansion

SCAIMED4¶

Run ID: SCAIMED4
Participant: fraunhofer.scai
Track: Medical
Year: 2011
Submission: 8/5/2011
Type: automatic
Task: main
MD5: 0ed85067e5edeadcee47ee6378a03295
Run description: search dictionary (MeSH,MedDRA,ATC,DrugBank) concepts in documents + automatic query expansion + search children concepts and their synonyms

SCAIMED5¶

Run ID: SCAIMED5
Participant: fraunhofer.scai
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: d133f8f8327f23347f89d05363b77836
Run description: Search using Keywords + SemRep + Concepts @ concept level

SCAIMED6¶

Run ID: SCAIMED6
Participant: fraunhofer.scai
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: 73f01d5e44f8051ce4802d0b6fcc547b
Run description: Search using SemRep + Concepts @ concept level and ontological level

SCAIMED7¶

Run ID: SCAIMED7
Participant: fraunhofer.scai
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: bb1a80976d440cd91864e527fb25bb88
Run description: Search using keywords + SemRep + Concepts @ concept level and ontological level

shannon¶

Results | Participants | Input | Summary | Appendix

Run ID: shannon
Participant: Michigan
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: 68b3540f411299b33f66606a03242e08
Run description: in this run, i removed the effects of all negations.

SQM¶

Results | Participants | Input | Summary | Appendix

Run ID: SQM
Participant: DUTCHHATTRICK
Track: Medical
Year: 2011
Submission: 8/15/2011
Type: automatic
Task: main
MD5: ca559a4941f75388507dc51ccb5c3940
Run description: Preprocessing of the corpus consisted of automatic spelling correction, section and sentence splitting, and applying the ConText algorithm to remove text that was either negated, about the history of the patient, or about some other than the patient. Frequently occurring sections that were deemed uninformative (e.g. 'Family History') were removed. Query expansion was done using Wikipedia: Text of the queries was used to identify corresponding Wikipedia pages, where queries were also used for disambiguation. Wikipedia page names, redirects, and link tag texts were used to identify synonyms. A combination of UMLS and Drugbank was used to expand drug classes (e.g. 'atypical antipsychotics') to individual drugs. This run uses language models in combination with query expansion. First, a match score per report was calculated. Visit scores were calculated as the maximum report score per vist. External sources used: Wikipedia, Drugbank, UMLS, ConText algorithm

UCDCSIrun3¶

Run ID: UCDCSIrun3
Participant: UCD_CSI
Track: Medical
Year: 2011
Submission: 8/12/2011
Type: manual
Task: main
MD5: f8b8a151d403387ee7e99726bc829394
Run description: Visits are returned using a language modeling system, with concept re-ranking then performed.

UCDCSIrun4¶

Run ID: UCDCSIrun4
Participant: UCD_CSI
Track: Medical
Year: 2011
Submission: 8/12/2011
Type: manual
Task: main
MD5: 5231e32fa2933b963f06bf292705e3b1
Run description: Reports are returned using first a language modeling technique, with concept re-ranking performed. A second run using okapi also performed. The scores of these two runs a combined to create a new ranking for the reports. Again, post-processing is performed to create visits from reports.

UCDCSIrunOne¶

Run ID: UCDCSIrunOne
Participant: UCD_CSI
Track: Medical
Year: 2011
Submission: 8/12/2011
Type: manual
Task: main
MD5: 90b1cc0d4f292404fab5d59121200ca4
Run description: Baseline run using language modeling to retrieve reports. Post-processing creates the visits by concatenating the retrieved reports.

UCDCSIrunTwo¶

Run ID: UCDCSIrunTwo
Participant: UCD_CSI
Track: Medical
Year: 2011
Submission: 8/12/2011
Type: manual
Task: main
MD5: 03619e61ef2c521625ac96244851f6c8
Run description: Reports are retrieved by a system using language modeling techniques. Concept re-ranking is then performed. Post-processing is performed to create visits from retrieved reports.

udelbl¶

Run ID: udelbl
Participant: udel
Track: Medical
Year: 2011
Submission: 9/12/2011
Type: automatic
Task: main
MD5: 1128d9fd4962c6a3caec1cfeecf14845
Run description: 1. Ad hoc retrieval using Indri. 2. Replace all diagnosis codes with their short descriptions

udelgn¶

Run ID: udelgn
Participant: udel
Track: Medical
Year: 2011
Submission: 9/12/2011
Type: automatic
Task: main
MD5: ee5638d561f3d2e9c8e0421903b64f3c
Run description: 1. Ad hoc retrieval using Indri. 2. Replace all diagnosis codes with their short descriptions 3. Used TREC Genomics Track 2007 data for query expansion

udelmx¶

Run ID: udelmx
Participant: udel
Track: Medical
Year: 2011
Submission: 9/12/2011
Type: automatic
Task: main
MD5: 945d3ea816554621bde88d64f7369828
Run description: 1. Ad hoc retrieval using Indri. 2. Replace all diagnosis codes with their short descriptions 3. Used TREC Genomics Track 2007 and one-day PubMed query log for query expansion

udelpm¶

Run ID: udelpm
Participant: udel
Track: Medical
Year: 2011
Submission: 9/12/2011
Type: automatic
Task: main
MD5: ac486d42012b654de7bf9ac122a81f5f
Run description: 1. Ad hoc retrieval using Indri. 2. Replace all diagnosis codes with their short descriptions 3. Used one-day PubMed query log for query expansion

UDMedBL¶

Run ID: UDMedBL
Participant: Udel_Fang
Track: Medical
Year: 2011
Submission: 8/19/2011
Type: automatic
Task: main
MD5: 90f2b5b0bae5c6a7b79ad5d8706ddbff
Run description: Applied F2-exp retrieval function.

UDMedComb¶

Run ID: UDMedComb
Participant: Udel_Fang
Track: Medical
Year: 2011
Submission: 8/15/2011
Type: automatic
Task: main
MD5: be0d4ab6bb3060b70130e1a8587609af
Run description: Applied F2-exp retrieval function. Applied aspect-based term proximity. Used information from www.healthline.com to do query expansion.

UDMedDiv¶

Run ID: UDMedDiv
Participant: Udel_Fang
Track: Medical
Year: 2011
Submission: 8/15/2011
Type: automatic
Task: main
MD5: 51fb11aa60b5a14900aa240294cb3506
Run description: Applied disease diversity based on the information from www.healthline.com

UDMedExp¶

Run ID: UDMedExp
Participant: Udel_Fang
Track: Medical
Year: 2011
Submission: 8/15/2011
Type: automatic
Task: main
MD5: 89c712c8c17a6ea67f4fb7e1916b07dd
Run description: Apply F2-exp retrieval function. Used information from www.healthline.com to do query expansion.

UDMedProx¶

Run ID: UDMedProx
Participant: Udel_Fang
Track: Medical
Year: 2011
Submission: 8/15/2011
Type: automatic
Task: main
MD5: 3b86990f27e60dae976d85c735f860f6
Run description: Apply F2-exp retrieval function. Apply aspect-based term proximity.

UHU1BL¶

Run ID: UHU1BL
Participant: MedicalMiner
Track: Medical
Year: 2011
Submission: 9/12/2011
Type: automatic
Task: main
MD5: f9d138097cbdfa921739d6b684221811
Run description: Universidad de Huelva - BaseLine

UHU2MFB¶

Run ID: UHU2MFB
Participant: MedicalMiner
Track: Medical
Year: 2011
Submission: 9/12/2011
Type: automatic
Task: main
MD5: 953b090c6e49f8e87d29f8a131f7144b
Run description: Universidad de Huelva - Medical Face Base uses MedicalMiner annotator developed by Vigo University

UHU3BWRTDDD¶

Run ID: UHU3BWRTDDD
Participant: MedicalMiner
Track: Medical
Year: 2011
Submission: 9/12/2011
Type: automatic
Task: main
MD5: e86807bca492e9129326ded42c447149
Run description: Universidad de Huelva - Uses ICD9 codes to retrieval.

UHU4BWRTDDD2¶

Run ID: UHU4BWRTDDD2
Participant: MedicalMiner
Track: Medical
Year: 2011
Submission: 9/12/2011
Type: automatic
Task: main
MD5: 588b8daf6f1259af24be5ddf0ce82bb9
Run description: Universidad de Huelva - Uses ICD9 codes to retrieval rated by gender and age.

UIICTSmed01¶

Run ID: UIICTSmed01
Participant: UI_ICTS
Track: Medical
Year: 2011
Submission: 8/16/2011
Type: automatic
Task: main
MD5: 3a1ae79c5beb5eb25607b4f3e74779e0
Run description: Text only, no diagnoses. ULMS concept recognition with limited concept expansion.

UIICTSmed02¶

Run ID: UIICTSmed02
Participant: UI_ICTS
Track: Medical
Year: 2011
Submission: 8/16/2011
Type: automatic
Task: main
MD5: 4b567cf1325462f241e70fce90531988
Run description: Text and diagnoses. ULMS concept recognition with limited concept expansion.

UIICTSmed03¶

Run ID: UIICTSmed03
Participant: UI_ICTS
Track: Medical
Year: 2011
Submission: 8/16/2011
Type: automatic
Task: main
MD5: 51742a52da667ce16356b744e58d70e6
Run description: Text and diagnoses with concept count ranking. ULMS concept recognition with limited concept expansion.

UIICTSmed04¶

Run ID: UIICTSmed04
Participant: UI_ICTS
Track: Medical
Year: 2011
Submission: 8/16/2011
Type: automatic
Task: main
MD5: c4cde4ef26cf302a198eb2e8f8cb3906
Run description: Text and diagnoses with frequency count ranking. ULMS concept recognition with limited concept expansion.

UIICTSmed05¶

Run ID: UIICTSmed05
Participant: UI_ICTS
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: 97907f25e1997641861855a5b98c46e6
Run description: free text concept extraction tempered with NegEx-style negation, count ranking.

UIICTSmed06¶

Run ID: UIICTSmed06
Participant: UI_ICTS
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: dad7a93676c45087f7d620612e0c5088
Run description: free text concept extraction tempered with NegEx-style negation, frequency sum ranking.

UIICTSmed07¶

Run ID: UIICTSmed07
Participant: UI_ICTS
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: bb704599a8bd8c14203015eab9125394
Run description: free text concept extraction, plus diagnosis tempered with NegEx-style negation, count ranking.

UIICTSmed08¶

Run ID: UIICTSmed08
Participant: UI_ICTS
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: 4912986b776043f33d3c440fc3dfb4b1
Run description: free text concept extraction, plus diagnosis tempered with NegEx-style negation, frequency sum ranking.

UIowaSMED1¶

Run ID: UIowaSMED1
Participant: UIowaS
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: 17f73aedb75e22ffc6472385b0052da9
Run description: Data Processing: For this run we first removed the files that did not have a checksum visit id mapping and also 845 report files that were dropped. As a preprocessing step all ICD9CM codes of the reports(mentioned in the admit diagnosis and discharge diagnosis fields) were translated to their literal form using the Unified Medical Language System (UMLS) database. An index of these files was created using Indri. Particular fields like admit and discharge diagnosis and report text were also indexed. Query Expansion: As a preprocessing step, a filter of stopwords containing terms like treated, take, etc. was applied to each query. Next, the queries were run through MetaMap and their semantic classes and CUIs were identified. Certain semantic classes like "Patient or Disabled Group", "Functional Concept", "Activity", etc. were removed for futher processing. The remaining concepts (CUIs) were expanded using UMLS. For this run we considered 4 separate strategies. Sub_Run_1: Concepts were expanded using stringent conditions where related concepts can only have "RN" (narrower relationship), "SY" (source asserted synonymy) or "CHD" (child relationship) relationships to the query CUI. Also they have to be in the same semantic class as the original query. The source vocabularies were limited to SNOMEDCT and ICD9CM. The final query does binary AND (#band) of synonymous relationships (#syn). The query structure in Indri looks like this: #band ( #syn( #1(a1) #1(a2) ) #syn( #1(b1) #1(b2) ) ...) Sub_Run_2: Same as Sub_Run_1 but we make the range of related concepts broader by incorporating "PAR" (parent relationship) and "RB" (broader relationship) relationships. Sub_Run_3: Same as Sub_Run_2 but we also include "RL" defined in UMLS as "the relationship is similar or "alike". the two concepts are similar or "alike". In the current edition of the Metathesaurus, most relationships with this attribute are mappings provided by a source, named in SAB and SL; hence concepts linked by this relationship may be synonymous, i.e. self-referential: CUI1 = CUI2. In previous releases, some MeSH Supplementary Concept relationships were represented in this way." Sub_Run_4: This is the least stringent query condition. Here instead of binary AND (#band) we use the combine operator (#combine). Combine does not require all the conditions to be true to satisfy a query. The Sub_Run_(1-4) were combined using a Balanced Round Robin Strategy where we take 3 results from Sub_Run_1, 3 results from Sub_Run_2, 2 results from Sub_Run_3 and 2 results from Sub_Run_4 and continue until each list is exhausted in a round robin way.

UIowaSMED2¶

Run ID: UIowaSMED2
Participant: UIowaS
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: 88cfc035b03107fe1cf6a6eebe5adb56
Run description: For Run2, we repeated the same steps as Run 1 (UIowaSMED1) but the index is built from multiple files grouped into a single file based on same visit ids.

UIowaSMED3¶

Run ID: UIowaSMED3
Participant: UIowaS
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: b66cc59796ea2db9f2de5534dad526c7
Run description: For Run3, we use the same index as in Run 2 (UIowaSMED2). But queries for Sub_Run_(1-3) here are run on individual fields like admit_diagnosis, discharge_diagnosis and report_text only.For Sub_Run_4, the query tries to find all the words of the query within an unordered window of 4 words. Results are finally combined using the Balanced Round Robin technique as described in Run 1, but taking 2 results from Sub_Run_1, 3 each from Sub_Run_2 and Sub_Run_3 and 2 from Sub_Run_4.

UIowaSMED4¶

Run ID: UIowaSMED4
Participant: UIowaS
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: 31a22b84188fd1c9151eaa286bf9c6cf
Run description: For Run4, we use the index used in Run 1 (UIowaSMED1). But instead of ranking results by the binary AND the results are ranked by first select documents satisfying all the terms (or its synonyms) in the query, and then rank the results according to a query term using a probabilistic model. The rest of the processing, including merging of results by the Balanced Round Robin strategy (described in UIowaSMED1), is done in the same way as in Run 1 (UIowaSMED1). Tools used: 1. NLM's MetaMap 2. Unified Medical Language System (UMLS) 3. INDRI IR system

unityranked¶

Run ID: unityranked
Participant: UCSOM_BTMG
Track: Medical
Year: 2011
Submission: 9/13/2011
Type: automatic
Task: main
MD5: 775b16ff30e95b5b77fbdfe47fbd18f9
Run description: This run consisted of running several Perl scripts and assembling their combined output for submission using trivial file manipulation. Both the original reports and the provided queries were run through the NLM Metamap tool with negated concepts enabled. Only concepts metamapped with 1.000 certainty were considered. Reports were ranked according to how many query concepts they contained, which were then translated into visit ids and the top 1,000 candidate visits were returned. Visits of equal scoring were returned arbitrarily.

uogTrDeNfCE¶

Run ID: uogTrDeNfCE
Participant: uogTr
Track: Medical
Year: 2011
Submission: 9/14/2011
Type: automatic
Task: main
MD5: 47fb4ac4e031d249d66b43608b01058b
Run description: This run combines Divergence from Randomness and Voting Model approaches to medical visit ranking. External resources (e.g. MeSH, ICD hierarchy, SnoMed, RxDrug) are used to drive a novel expansion technique.

uogTrDeNIo¶

Run ID: uogTrDeNIo
Participant: uogTr
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: 666649f1d81ef34a583c075492b9ea2e
Run description: This run combines Divergence from Randomness and Voting Model approaches to medical visit ranking. ICD & Wikipedia are used as an external resources to drive a novel expansion technique.

uogTrDeNsEc¶

Run ID: uogTrDeNsEc
Participant: uogTr
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: 0378cfdfb52cf0abebe2e471105daf3f
Run description: This run combines Divergence from Randomness and Voting Model approaches to medical visit ranking. MesH is used as an external resource to drive a novel expansion technique.

uogTrDeNSo¶

Run ID: uogTrDeNSo
Participant: uogTr
Track: Medical
Year: 2011
Submission: 8/16/2011
Type: automatic
Task: main
MD5: 94e7be8a596caf5a7594185edd394f23
Run description: This run combines Divergence from Randomness and Voting Model approaches to medical visit ranking. Aggregates are ranked at two levels. MesH is used as an external resource.

uogTrMDeNFo¶

Run ID: uogTrMDeNFo
Participant: uogTr
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: c39c803df0a0fb8e8854f335c94ce1a4
Run description: This run combines Divergence from Randomness and Voting Model approaches as well as shallow NLP to achieve effective medical visit ranking.

UTDHLTCIR¶

Run ID: UTDHLTCIR
Participant: UTD_HLT
Track: Medical
Year: 2011
Submission: 9/13/2011
Type: automatic
Task: main
MD5: 9736c20f39e813aba89e6382ef6a0d1b
Run description: Scoring is performed by a weighted vote between standard Lucene, Indri, and a modified Lucene system. Query expansion is performed using UMLS synonyms, Wikipedia redirects, SNOMED relations, and using Normalized Google Distance to find related phrases from the PubMed Central Open Access Subset. Irrelevant documents are filtered through the use of age, gender, and negation detection based on NegEx.

UTDHLTCIRLS¶

Run ID: UTDHLTCIRLS
Participant: UTD_HLT
Track: Medical
Year: 2011
Submission: 9/13/2011
Type: automatic
Task: main
MD5: 043d1fd130fed42368de53539a3d3a2a
Run description: Scoring is performed by a weighted vote between standard Lucene, Indri, and a modified Lucene system. Query expansion is performed using UMLS synonyms, Wikipedia redirects, SNOMED relations, and using Normalized Google Distance to find related phrases from the PubMed Central Open Access Subset. Irrelevant documents are filtered through the use of age, gender, and negation detection based on LingScope negation and hedging.

UTDHLTMK¶

Run ID: UTDHLTMK
Participant: UTD_HLT
Track: Medical
Year: 2011
Submission: 8/15/2011
Type: automatic
Task: main
MD5: f1018f94f00f850f4e86ceb7cbdf034e
Run description: The initial document ranking metric promotes documents containing multiple keywords and demotes documents matching one keyword many times. Query expansion is performed using UMLS synonyms, Wikipedia redirects, SNOMED relations, and using Normalized Google Distance to find related phrases from the PubMed Central Open Access Subset. Irrelevant documents are filtered through the use of age, gender, and negation detection.

UTDHLTSL¶

Run ID: UTDHLTSL
Participant: UTD_HLT
Track: Medical
Year: 2011
Submission: 8/15/2011
Type: automatic
Task: main
MD5: 4a8e7a854db9b1c139716bb47e8d8fe6
Run description: Initial document ranking is performed using the standard Lucene scoring metric. Query expansion is performed using UMLS synonyms, Wikipedia redirects, SNOMED relations, and using Normalized Google Distance to find related phrases from the PubMed Central Open Access Subset. Irrelevant documents are filtered through the use of age, gender, and negation detection.

WWOCorrect¶

Results | Participants | Input | Summary | Appendix

Run ID: WWOCorrect
Participant: DUTCHHATTRICK
Track: Medical
Year: 2011
Submission: 8/15/2011
Type: automatic
Task: main
MD5: baf7de54d7c689058970b277c0bc2b08
Run description: Preprocessing of the corpus consisted of automatic spelling correction, section and sentence splitting, and applying the ConText algorithm to remove text that was either negated, about the history of the patient, or about some other than the patient. Frequently occurring sections that were deemed uninformative (e.g. 'Family History') were removed. Query expansion was done using Wikipedia: Text of the queries was used to identify corresponding Wikipedia pages, where queries were also used for disambiguation. Wikipedia page names, redirects, and link tag texts were used to identify synonyms. A combination of UMLS and Drugbank was used to expand drug classes (e.g. 'atypical antipsychotics') to individual drugs. This run uses a variation of the Okapi search model, where matches between queries and reports were weighted by type of match (exact, or spread across sentences or sections), type of section (several sections such as 'Chief complaint' were given more weight), and reliability of query-term relation. A per-document winner-take-all approach was used for expansion of the queries. First, a match score per report was calculated. Visit scores were calculated as the maximum report score per visit. External sources used: Wikipedia, Drugbank, UMLS, ConText algorithm

york11CB1¶

Run ID: york11CB1
Participant: york
Track: Medical
Year: 2011
Submission: 8/16/2011
Type: automatic
Task: main
MD5: dd9f8fa7b77b3e35cb608dee8cf7483f
Run description: In this run, we use the classic BM25 model as the retrieval model. The parameters of BM25 are set as: k1=1.2, b=0.75 and k3=8. The original retrieved document list is based on checksum which works as the report ID. We map the checksum to visitID and merge the results belonging to the same patient. Finally, there are at most 1000 visitIds for each topic in the submitted run.

york11CQ2¶

Run ID: york11CQ2
Participant: york
Track: Medical
Year: 2011
Submission: 8/16/2011
Type: automatic
Task: main
MD5: 2ec80892630224f1237aef52f6f53e23
Run description: In this run, we use the classic BM25 model in the first retrieval phase and Rocchio's relevance feedback for the query expansion process. We set k1=1.2, b=0.75 and k3=8 for BM25. Top 3 documents are selected as the candidate feedback documents, and 10 terms are selected as the feedback terms. Besides, we extract the gender and age information from the collection automatically and use that to filter the results obtained by the BM25+Rocchio's relevance feedback.

york11mQeD1¶

Run ID: york11mQeD1
Participant: york
Track: Medical
Year: 2011
Submission: 8/16/2011
Type: automatic
Task: main
MD5: f06cb9d27c370c782227e852c9dfd2ac
Run description: This run is based on automatic query expansion using disease synonym dictionary, (available online at http://wishart.biology.ualberta.ca/polysearch/include/disease_IDlist.txt). First, we used biolabler medical annotation tool (available at http://www.biolabeler.com/bioLabeler/) to extract the disease keywords from the query. The disease query keywords are then matched with the dictionary entries to extract a set of synonym sets. The synonym sets that present a maximum overlap with the disease query keywords are selected and used to append the query with up 40 most representative terms.

york11mSB1¶

Run ID: york11mSB1
Participant: york
Track: Medical
Year: 2011
Submission: 8/16/2011
Type: automatic
Task: main
MD5: 57c200126f2329e5b0679534835df4e4
Run description: This run used an online service known as BioLabeler (http://www.biolabeler.com/bioLabeler/). BioLabeler is biomedical text mining tool which extracts Unified Medical Language System (UMLS) concepts from any given text that is proved to the said service. All UMLS sources that were available were used to generate two indexes of medical concepts for each record: index of diseases and procedures. Each medical concept contained statistical information to determine the relevance of a particular medical concept to the record, specifically its normalized weight and regular weight. To obtain the final results, we performed concept matching where we got a list of concepts for each topic, and matched it with corresponding concepts for each medical record. The medical records that were chosen to appear on the final result were the ones in which had the highest normalized weight, then ordered in descending order. The entire process from retrieving medical concepts from BioLabeler for the medical record and topic files, to obtaining the final result file was automated.

zen¶

Results | Participants | Input | Summary | Appendix

Run ID: zen
Participant: Michigan
Track: Medical
Year: 2011
Submission: 8/17/2011
Type: automatic
Task: main
MD5: 4a15f87162554476ee520f47053a5511
Run description: Lemur