4.2 Generate Pre-annotations Using Regular Expressions

Dictionaries are good for terms you can list. For things that follow a pattern — dates, identifiers, measurements, dosages — a regular expression is a better fit. eHOST can apply both in the same run.

Regular expressions come from two places:

Source	Where it lives	Configured in
Built-in patterns	Compiled into eHOST	The Using custom regular expressions box of the NLP Assisted panel
Your own patterns	customre.lib in eHOST’s working directory	Dictionary Setting → Custom Regular Expressions

Built-in patterns

In the NLP Assisted panel, tick Using custom regular expressions and then whichever built-in patterns you need:

• Date & Time
• SSN (US Social Security Numbers)

These are handy for de-identification work — the Pin Extractor toolbar button uses the same idea for personally identifying numbers.

Writing your own patterns

Click Dictionary Setting in the toolbar, then open the Custom Regular Expressions tab. The list shows the entries currently in customre.lib. Click New to add one.

An entry has three fields:

Field	Meaning
Regular Expression	A Java regular expression. Matching is case-insensitive.
Class Name	The annotation class assigned to every match. It should already exist in your schema — see 1.5.1 Define Classes
Comment	A free-text note describing the pattern; stored with the entry

Verify and Save Changes compiles the expression before storing it, so a typo is reported immediately instead of failing silently in the middle of a run. All three fields must be non-empty or the entry is dropped when the library is reloaded.

The customre.lib file

Entries are stored one per line, tab separated, in eHOST’s working directory:

<regular expression>\t<class name>\t<comment>

For example:

\d{1,3}\s?mg	Dosage	numeric dose in milligrams
(?:BP|blood pressure)\s*\d{2,3}/\d{2,3}	VitalSign	blood pressure reading

You can edit the file directly if you prefer, but going through the dialog gets you the syntax check. Because the file is shared by every project, patterns you add are available everywhere.

Applying the patterns

1. In the NLP Assisted panel tick Search Terms — Using custom regular expressions.
2. Optionally also tick a dictionary source and/or the built-in patterns; they all run in one pass.
3. Click Set Output Directory, then Start.

Tips

• Anchor patterns loosely. eHOST feeds sentences to the matcher, so ^/$ rarely behave the way you expect.
• Test on one document first — a greedy pattern can create thousands of annotations very quickly.
• Because matching is case-insensitive, do not add (?i) or separate upper/lower-case variants.
• Escape backslashes as Java expects them (\\d in Java source is \d in the dialog — type what you would type in a regex tester, not what you would type in Java code).