Release notes for NLP API 2.22

Summary

Fixed

Fixed some issues with complex overlapping entity spans.
Fixed bugs related to mentions that are ambiguous but disambiguated wrt document category, subcategory, etc.
Bug fix: sentence ending for a double period bug fixed.
Bug fix: text spans in section and sub-section presents.
Bug fix: qualifer relation configuration bug
Bug fix: do not make sections on multi-word BIRADS labels in radiology reports.
Fixed bug that caused a crash in followup relations in some cases.
Fix for missing annotations due to disambiguation bug in entitylink.
Bug fix: Missing annotations attributed to UTF8 handling.
Bug fix: Fixed an issue where some biomarkers were linked to incorrect concepts.
Bug fix: Corrected the concept used for some unitless measurements.
Fixed bug in qualifier relations related to config file loading.
Fixed some entity text span bugs in engine output JSON.
Bug fix: correct disambiguation in entitylink case sensitive matching.
Double period sentence ending bugfix
Force OCR worker to use only one thread, avoiding overload.
Fixes to PDF handling to handle more parsing cases.
Fixed a bug that caused a rare crash in the NLP processing.
Proper raw spans for section locations.
Bug fix: imagelinks were being associated with procedure mentions.
Bug fix: prevent spurious imagelinks relations.

Added

Added support for SNOMED CT 2020-2021 version.
Better section finding for Cardiology/Echocardiology.
Relations are now reported for cardiology echocardiogram report types as well.
Added feature support for spellchecking but it is disabled by default.
Support for new timeout parameter on status endpoint.
Handle time expressions that have underscores in them, e.g. __4__ weeks.
Added functionality to unitless measurements.
Add spellchecker support (it is currently turned off by default).
Add new semantic types “observable entity” and “regime/therapy” to SNOMED linking.
Improved concept linking for the HGNC ontology.
Tokenizer updated to handle combined route and frequency in medication relations.
Support PATCH as a verb on the cancel endpoint.

Changed

Better support for unitless measurements.
Improvements to sentence boundary detection.
Improvements to negation detection.
Improvements to medication relations.
Large number of SNOMED, RadLex, Medcin, HGNC accuracy updates.
Allow only one to one trigger and subject in stage/grade type of qualifier relations.

Compatibility

API version 2 Release 22
Python SDK and client v4.3.0 to v5.5.0
Java SDK and client v3.0.2 to v5.1.0
JSON result output schema v2.14 for emtellipro-json-2

The latest version of the SDK is required for the new features in this release. The JSON result output schemas are backward compatible for all v2.x versions of emtellipro-json-2.

New timeout parameter on status endpoint

The new timeout parameter lets any NLP API client to add a timeout to the status request so that the NLP API service will wait until the timeout to return back the status to the client. This will often allow the client to get a completed status much more quickly than the usual poll status and sleep for one second that was the previous default behaviour of the client.

Support PATCH as a verb on cancel endpoint

We now support both verbs PATCH and GET for the cancel endpoint in the NLP API service. At a later date we may deprecate the use of GET as it is more idiomatic to use PATCH in REST APIs for such an endpoint.

Python SDK updates

The new NLP API Python SDK is version 5.5.0. Refer to the changelog for the Python SDK for a full explanation of the updates made since the last public release which was version 5.2.0 released with the NLP API v2.21.

Contents