metamap parser

A python script for parsing the JSON output of the NLM MetaMap

What is MetaMap?

It takes free text and identifies UMLS concepts in that text. These concepts are identified using CUIs - concept unique identifiers. For more, see below:

MetaMap - A Tool For Recognizing UMLS Concepts in Text

UMLS: Unified Medical Language System

What does the script do?

It finds mentions of CUIs in the JSON output of MetaMap, and inserts them back into the original document, replacing the parts of the string that MetaMap identified as mapping to that CUI.

Examples

Input sentence: It does not impede her lifestyle at this point --> It does C1518422 impede her C0023676 at this C1552961

Negation:

If MetaMap has flagged that the CUI is negated, it will be replace as NOTCUI, e.g. No rashes -- > No NOTC0015230

Remapping:

The script also creates a "remapped" version, where the CUI is translated back into its preferred string form, e.g.: Sclerae white --> C0036410 C0007457 --> SCLERA WHITE

This is largely for sanity-checking the parsing and occasionally breaks sentence structure, but could in principle be used to reduce sentence variability.

Usage

Assuming a python(3) interactive session, run process_document(path_in, path_out), where path_in is the path to the MetaMap JSON output file, and path_out is optionally the desired path to write to. If you don't provide path_out it will just write to path_in + ".parsed"

Assumptions

I hacked this together using somewhat limited documentation, so I make the following assumptions:

We take the first utterance in the list of Utterances
We take the first mapping in the list of Mappings
If a concept maps to multiple places in a string, we put the CUI replacement at the first part (e.g. infection of the lung --> LUNGINFECTION of the because infection and lung are both part of the CUI, so we just replace infection with the whole CUI (which I am pretending is LUNGINFECTION)

corcra/metamap_parser