These instructions are for end users. With these instructions you can install Apache cTAKES, configure it, and use it to process text (typically text associated with a medical record). If you were planning to expand, change, or modify the code within cTAKES, refer to the cTAKES 3.1 Developer Install Guide.
These instructions will cover installation and a test of the main product including trained models for sentence detection and tagging parts of speech, dictionaries from a subset of the UMLS, the LVG resource, etc. Optional components are described in the Component Use Guide.
Once you have finished installing cTAKES and its separately-bundled resources, you will be able to see what cTAKES is capable of.
1. Make sure you have Java 1.6 or higher. Most systems come with Java already installed.
1. Navigate to the cTAKES downloads page on the Apache site and download the binary package. Select a mirror site and press the Change button to modify the URL to your desired mirror location before doing the download or accept the default.
2. (Optional but recommended) Verify the downloaded files against a file signature to ensure you have the proper and complete file.
3. Unzip the file you downloaded into a directory that you want to be the cTAKES install location. The compressed files contain a single directory at the top level. This folder we will call <cTAKES_HOME>. You will need to refer to this directory later.
5. Copy (or move) the resources to cTAKES_HOME.
(Optional) Add UMLS access rights
In the initial setup cTAKES will recognize only few sample concepts in text. If you wish to perform named entity recognition or concept identification for anything other than these few words, you will need to 1) obtain the rights to use UMLS resources 2) add those credentials to cTAKES, and 3) use an aggregate that makes use of those UMLS resources. If you don't, cTAKES will work but won't recognize much.
1. If you do not have a UMLS username and password, you may request one at UMLS Terminology Services.
2. Edit the following files. Find the line in each script that runs java and add the ctakes.umlsuser and ctakes.umlspw parameters to the java command with your credentials. Make sure you substitute your actual ID and password if you cut and paste the example.
Process documents using cTAKES
This version allows you to test most components bundled in cTAKES in two different ways:
- Using the bundled UIMA CAS Visual Debugger (CVD) to view the results stored as XCAS files or run the annotators
- Using the bundled UIMA Collection Processing Engine (CPE) to process documents in cTAKES_HOME/testdata directory
You will need a windowing environment on Linux to run these tools.
CAS Visual Debugger (CVD)
1. Open a command prompt and change to the cTAKES_HOME directory.
2. Start the CAS Visual Debugger by running this command:
3. Copy the example text from the next cell in this table and paste the contents into the Text section of CVD, replacing the text that is already there.
4. An analysis engine (AE) needs to be loaded in order to process text.
in this step.
5. From the menu bar, click Run -> Run AggregatePlaintextProcessor or "Run AggregatePlaintextUMLSProcessor".
6. You'll get a list of all the annotations for this clinical document in the Analysis Results frame. Annotations such as named entities, division by sentence, etc from the pipeline are viewable. To see one, in the Analysis Results frame, click on the key in front of:
This will show an AnnotationIndex in the lower frame. Select any annotation in that lower frame and you will see the text discovered in
Now select items in the lower frame to see the text being annotated.
Collection Processing Engine (CPE)
1. Open a command prompt and change to the cTAKES_HOME directory:
2. Create a directory for some test data.
3. Download this sample file and place it into the testdata directory.
4. Start the collection processing engine by running this command:
5. This will bring up the Collection Processing Engine Configurator. In the Menu bar click File >Open CPE Descriptor
6. Navigate to the following file, which uses the AggregateCdaProcessor
7. Change the Collection reader input directory to testdata, which contains a CDA file(s).
8. Click the Play button (green/blue play arrow near the bottom).
9. You should see that one document was processed. You did process a collection of documents. In this case the collection only contained one just to show how to do it. Close the results window.
10. Close the CPE application. You may be prompted to save changes. Since this was just a test you may click the No button.
Using the same CVD and CPE programs in the manner described above, you can test all the other components. The analysis engines and collection processing engines shipped with cTAKES for some of the annotators are described in the following table.
cTAKES 3.1 binary distributions did not include test data. Loading the CPE descriptors into the CPE tool will require resetting the input and output directories. Test files could be obtained from the cTAKES 2.5 release binary distribution. Look for a testdata directory in cTAKES_HOME.
Example Aggregate Analysis Engine (AE)
Example Collection processing Engine (CPE)
Clinical Document Pipeline
The complete cTAKES pipeline to obtain majority of cTAKES annotations
Obtain cTAKES chunk annotations
Obtain dependency parsing tree
The annotator to obtain drug annotations
Mapping cTAKES annotations to dictionaries (e.g., SNOMED_CT or RxNorm
PAD Term Spotter
Identifying terms related to PAD
Annotate certain relations between certain Event, Entity, and Modifier annotations
The annotator to obtain document or patient-level smoking status
The annotator to find side effect mentions and sentences from clinical documents
The cTAKES 3.1 Component Use Guide will help you to understand, in great detail, each of the cTAKES components that have been installed. In some cases you can learn how to improve the components.
Also, before you go on to process text in production, you will want to consider dictionaries and models. If you did not obtain the rights yet to the UMLS resources and models, you will want to do so. Be aware, the models have been trained on data that may not match your data well enough to be effective. In some cases you might want to modify the dictionaries and train models using your own data.