Our human annotators on Share used 2012AB. I mention it because when I have done manual spot-checks between human and system annotations I had head-scratchers that ended up being differences in the UMLS version. I first noticed these discrepancies before I had started working on the fast lookup (that is to say: when working with the default lookup).
From: Kim Ebert [mailto:kim.eb...@perfectsearchcorp.com] Sent: Friday, December 19, 2014 1:40 PM To: dev@ctakes.apache.org Subject: Re: cTakes Annotation Comparison Sean, I don't think that would be an issue since both the rare word lookup and the first word lookup are using UMLS 2011AB. Or is the rare word lookup using a different dictionary? I would expect roughly similar results between the two when it comes to differences between UMLS versions. [IMAT Solutions]<http://imatsolutions.com> Kim Ebert Software Engineer [Office:]801.669.7342 kim.eb...@imatsolutions.com<mailto:greg.hub...@imatsolutions.com> On 12/19/2014 11:31 AM, Finan, Sean wrote: One quick mention: The cTakes dictionaries are built with UMLS 2011AB. If the Human annotations were not done using the same UMLS version then there WILL be differences in CUI and Semantic group. I don't have time to go into it with details, examples, etc. just be aware that every 6 months cuis are added, removed, deprecated, and moved from one TUI to another. Sean -----Original Message----- From: Savova, Guergana [mailto:guergana.sav...@childrens.harvard.edu] Sent: Friday, December 19, 2014 1:28 PM To: dev@ctakes.apache.org<mailto:dev@ctakes.apache.org> Subject: RE: cTakes Annotation Comparison Several thoughts: 1. The ShARE corpus annotates only mentions of type Diseases/Disorders and only Anatomical Sites associated with a Disease/Disorder. This is by design. cTAKES annotates all mentions of types Diseases/Disorders, Signs/Symptoms, Procedures, Medications and Anatomical Sites. Therefore you will get MANY more annotations with cTAKES. Eventually the ShARe corpus will be expanded to the other types. 2. Keeping (1) in mind, you can approximately estimate the precision/recall/f1 of cTAKES on the ShARe corpus if you output only mentions of type Disease/Disorder. 3. Could you send us the list of files you use from ShARe to test? We have the corpus and would like to run against as well. Hope this makes sense... --Guergana -----Original Message----- From: Bruce Tietjen [mailto:bruce.tiet...@perfectsearchcorp.com] Sent: Friday, December 19, 2014 1:16 PM To: dev@ctakes.apache.org<mailto:dev@ctakes.apache.org> Subject: Re: cTakes Annotation Comparison Our analysis against the human adjudicated gold standard from this SHARE corpus is using a simple check to see if the cTakes output included the annotation specified by the gold standard. The initial results I reported were for exact matches of CUI and text span. Only exact matches were counted. It looks like if we also count as matches cTakes annotations with a matching CUI and a text span that overlaps the gold standard text span then the matches increase to 224 matching annotations for the FastUMLS pipeline and 2319 for the the old pipeline. The question was also asked about annotations in the cTakes output that were not in the human adjudicated gold standard. The answer is yes, there were a lot of additional annotations made by cTakes that don't appear to be in the gold standard. We haven't analyzed that yet, but it looks like the gold standard we are using may only have Disease_Disorder annotations. [image: IMAT Solutions] <http://imatsolutions.com><http://imatsolutions.com> Bruce Tietjen Senior Software Engineer [image: Mobile:] 801.634.1547 bruce.tiet...@imatsolutions.com<mailto:bruce.tiet...@imatsolutions.com> On Fri, Dec 19, 2014 at 9:54 AM, Miller, Timothy < timothy.mil...@childrens.harvard.edu<mailto:timothy.mil...@childrens.harvard.edu>> wrote: Thanks Kim, This sounds interesting though I don't totally understand it. Are you saying that extraction performance for a given note depends on which order the note was in the processing queue? If so that's pretty bad! If you (or anyone else who understands this issue) has a concrete example I think that might help me understand what the problem is/was. Even though, as Pei mentioned, we are going to try moving the community to the faster dictionary, I would like to understand better just to help myself avoid issues of this type going forward (and verify the new dictionary doesn't use similar logic). Also, when we finish annotating the sample notes, might we use that as a point of comparison for the two dictionaries? That would get around the issue that not everyone has access to the datasets we used for validation and others are likely not able to share theirs either. And maybe we can replicate the notes if we want to simulate the scenario Kim is talking about with thousands or more notes. Tim On 12/19/2014 10:24 AM, Kim Ebert wrote: Guergana, I'm curious to the number of records that are in your gold standard sets, or if your gold standard set was run through a long running cTAKES process. I know at some point we fixed a bug in the old dictionary lookup that caused the permutations to become corrupted over time. Typically this isn't seen in the first few records, but over time as patterns are used the permutations would become corrupted. This caused documents that were fed through cTAKES more than once to have less codes returned than the first time. For example, if a permutation of 4,2,3,1 was found, the permutation would be corrupted to be 1,2,3,4. It would no longer be possible to detect permutations of 4,2,3,1 until cTAKES was restarted. We got the fix in after the cTAKES 3.2.0 release. https://issues.apache.org/jira/browse/CTAKES-310 Depending upon the corpus size, I could see the permutation engine eventually only have a single permutation of 1,2,3,4. Typically though, this isn't very easily detected in the first 100 or so documents. We discovered this issue when we made cTAKES have consistent output of codes in our system. [IMAT Solutions]<http://imatsolutions.com><http://imatsolutions.com> Kim Ebert Software Engineer [Office:] 801.669.7342 kim.eb...@imatsolutions.com<mailto:kim.eb...@imatsolutions.com><mailto:greg.hub...@imatsolutions.com><mailto:greg.hub...@imatsolutions.com> On 12/19/2014 07:05 AM, Savova, Guergana wrote: We are doing a similar kind of evaluation and will report the results. Before we released the Fast lookup, we did a systematic evaluation across three gold standard sets. We did not see the trend that Bruce reported below. The P, R and F1 results from the old dictionary look up and the fast one were similar. Thank you everyone! --Guergana -----Original Message----- From: David Kincaid [mailto:kincaid.d...@gmail.com] Sent: Friday, December 19, 2014 9:02 AM To: dev@ctakes.apache.org<mailto:dev@ctakes.apache.org><mailto:dev@ctakes.apache.org><mailto:dev@ctakes.apache.org> Subject: Re: cTakes Annotation Comparison Thanks for this, Bruce! Very interesting work. It confirms what I've seen in my small tests that I've done in a non-systematic way. Did you happen to capture the number of false positives yet (annotations made by cTAKES that are not in the human adjudicated standard)? I've seen a lot of dictionary hits that are not actually entity mentions, but I haven't had a chance to do a systematic analysis (we're working on our annotated gold standard now). One great example is the antibiotic "Today". Every time the word today appears in any text it is annotated as a medication mention when it almost never is being used in that sense. These results by themselves are quite disappointing to me. Both the UMLSProcessor and especially the FastUMLSProcessor seem to have pretty poor recall. It seems like the trade off for more speed is a ten-fold (or more) decrease in entity recognition. Thanks again for sharing your results with us. I think they are very useful to the project. - Dave On Thu, Dec 18, 2014 at 5:06 PM, Bruce Tietjen < bruce.tiet...@perfectsearchcorp.com<mailto:bruce.tiet...@perfectsearchcorp.com><mailto:<mailto:bruce.tiet...@perfectsearchcorp.com> bruce.tiet...@perfectsearchcorp.com><mailto:bruce.tiet...@perfectsearchcorp.com>> wrote: Actually, we are working on a similar tool to compare it to the human adjudicated standard for the set we tested against. I didn't mention it before because the tool isn't complete yet, but initial results for the set (excluding those marked as "CUI-less") was as follows: Human adjudicated annotations: 4591 (excluding CUI-less) Annotations found matching the human adjudicated standard UMLSProcessor 2245 FastUMLSProcessor 215 [image: IMAT Solutions] <http://imatsolutions.com><http://imatsolutions.com>< <http://imatsolutions.com> http://imatsolutions.com><http://imatsolutions.com> Bruce Tietjen Senior Software Engineer [image: Mobile:] 801.634.1547 bruce.tiet...@imatsolutions.com<mailto:bruce.tiet...@imatsolutions.com><mailto:bruce.tiet...@imatsolutions.com On Thu, Dec 18, 2014 at 3:37 PM, Chen, Pei <pei.c...@childrens.harvard.edu<mailto:pei.c...@childrens.harvard.edu><mailto:pei.c...@childrens.harvard.edu><mailto:pei.c...@childrens.harvard.edu> wrote: Bruce, Thanks for this-- very useful. Perhaps Sean Finan comment more- but it's also probably worth it to compare to an adjudicated human annotated gold standard. --Pei -----Original Message----- From: Bruce Tietjen [mailto:bruce.tiet...@perfectsearchcorp.com] Sent: Thursday, December 18, 2014 1:45 PM To: dev@ctakes.apache.org<mailto:dev@ctakes.apache.org><mailto:dev@ctakes.apache.org><mailto:dev@ctakes.apache.org> Subject: cTakes Annotation Comparison With the recent release of cTakes 3.2.1, we were very interested in checking for any differences in annotations between using the AggregatePlaintextUMLSProcessor pipeline and the AggregatePlanetextFastUMLSProcessor pipeline within this release of cTakes with its associated set of UMLS resources. We chose to use the SHARE 14-a-b Training data that consists of 199 documents (Discharge 61, ECG 54, Echo 42 and Radiology 42) as the basis for the comparison. We decided to share a summary of the results with the development community. Documents Processed: 199 Processing Time: UMLSProcessor 2,439 seconds FastUMLSProcessor 1,837 seconds Total Annotations Reported: UMLSProcessor 20,365 annotations FastUMLSProcessor 8,284 annotations Annotation Comparisons: Annotations common to both sets: 3,940 Annotations reported only by the UMLSProcessor: 16,425 Annotations reported only by the FastUMLSProcessor: 4,344 If anyone is interested, following was our test procedure: We used the UIMA CPE to process the document set twice, once using the AggregatePlaintextUMLSProcessor pipeline and once using the AggregatePlaintextFastUMLSProcessor pipeline. We used the WriteCAStoFile CAS consumer to write the results to output files. We used a tool we recently developed to analyze and compare the annotations generated by the two pipelines. The tool compares the two outputs for each file and reports any differences in the annotations (MedicationMention, SignSymptomMention, ProcedureMention, AnatomicalSiteMention, and DiseaseDisorderMention) between the two output sets. The tool reports the number of 'matches' and 'misses' between each annotation set. A 'match' is defined as the presence of an identified source text interval with its associated CUI appearing in both annotation sets. A 'miss' is defined as the presence of an identified source text interval and its associated CUI in one annotation set, but no matching identified source text interval and CUI in the other. The tool also reports the total number of annotations (source text intervals with associated CUIs) reported in each annotation set. The compare tool is in our GitHub repository at https://github.com/perfectsearch/cTAKES-compare
