Concept Detection

The concept detection service detects medical concepts from unstructured data. The service provides concepts based on the Unified Medical Language System (UMLS). Consumers can elect to have a set of medical codes associated with the UMLS concepts by specifying the optional configuration parameter to return the medical codes. When medical codes are requested, the UMLS concept annotations from concept detection will include the applicable medical codes as metadata within the annotations.

Expanded Concepts

Concept detection provides an expanded option that allows you to go beyond exact matching of surface forms in a dictionary. With expanded set to true, concept detection will learn from the dictionary entries you have specified and generalize to other ways that concepts can be expressed in text. This allows you to define a core set of surface forms without exhaustively listing every surface form for every concept. This becomes particularly important with compound ideas. Consider the following text:

The patient broke his leg and hip when he fell outside his home.

There are two different injuries expressed in this text that we want to capture - broken leg and broken hip. Neither one will have a dictionary entry that covers the way the concept is expressed. In this example, concept detection would return the following concepts with expanded set true.

{
    "cui": "C0159852",
    "preferredName": "Fracture of tibia and fibula",
    "semanticType": "inpo",
    "source": "umls-expanded",
    "sourceVersion": "2018AA",
    "type": "umls.InjuryOrPoisoning",
    "begin": 12,
    "end": 25,

Expanded detection will look for diseases, conditions, abnormalities, injuries, and procedures defined in the UMLS dictionary that ships with Annotator for Clinical Data. In addition, expanded detection will look for all of the concepts defined in custom dictionaries that are enabled for expanded detection in the cartridge configuration.

Configurations

The following table lists parameters of the concept_detection service.

Configuration	Values	Description
Libraries	umls.latest umls.2022AA umls.2021AA umls.2020AA (deprecated - will be removed in 2023)	Defines the version of the UMLS library that is used when analyzing unstructured data.*
inference_rules		The name of a derived concept rule set that will be used for deriving additional concepts based on the concepts discovered by the libraries specified.
filters		The name of a concept filter that is used to remove unwanted concepts.
expanded	true/false	When true, the concept detection annotator will attempt to expand concept coverage beyond the surface forms explicitly listed in the specified library. For example - if broken collarbone is a surface from for C0159658 (Fracture of clavicle), the expanded option would match textual representations of that concept like broke my collarbone. This option is false by default.
include_optional_fields	medical_codes / source_vocabularies	Optional fields that should also be returned for each concept. If not specified, only the libary’s default fields will be returned.
longest_span	true/false	When true (default), only the concept with the longest text span will be returned if there are multiple concepts overlapping the same span of text.

*The value umls.latest will reference the latest available version of UMLS within the service. As newer versions of UMLS are made available in the service, umls.latest library configurations will automatically leverage the latest available version of UMLS in the service once available. Declaration of a specific version of UMLS is recommended to avoid undesirable changes in output as newer versions of UMLS are made available within the service. Through declaration of a specific version of UMLS, newer versions of UMLS may be evaluated prior to use in production.

Annotation Types

Concept

Concept

Fields	Description
cui	Concept Unique ID (CUI). CUIs are used to uniquely identify concepts.
preferredName	Normalized name for the concept.
semanticType	Shorthand version of the UMLS semantic type.
source	The library source used for the detection of the concepts.
sourceVersion	he version of the library source used for the detection of the concepts.
type	The semantic type associated with the concept.
begin	The start position of the annotation as a character offset into the text. The smallest possible start position is 0.
end	The end position of the annotation as character offset into the text. The end position points at the first character after the annotation, such that end-begin equals the length of the coveredText.
coveredText	The text covered by an annotation as a string.
cptCode	This code represents the type of procedure that is performed. CPT stands for Current Procedural Terminology. This code a standard terminology used by different members of medical society such as physicians, financial administrators, coders, and other organizations. This value is only available when a CPT Codes file is referenced in the profile.
sourceVocabularies	This provides the list of UMLS source vocabularies that contained the concept. This is provided when the include_optional_fields parameter is specified with a value of `source_vocabularies`.

The following optional response fields are provided when the include_optional_fields parameter is specified with a value of medical_codes.

Fields	Description
loincId	LOINC stands for Logical Observations Identifiers, Names, Codes. The value for this feature comes from UMLS.
NCI Code	The NCI Thesaurus covers vocabulary for cancer-related clinical care, translational and basic research, and public information and administrative activities. The value for this feature comes from UMLS.
snomedConceptId	Numerical code provided by the SNOMED dictionaries that represents the cancer.
MeSHId	The MeSH thesaurus is a controlled vocabulary used for indexing, cataloging, and searching for biomedical and health-related information and documents. The value for this feature comes from UMLS.
icd9Code	ICD stands for International Classification of Diseases. The number 9 is a revision number for this code set.
icd10Code	ICD stands for International Classification of Diseases. The number 10 is a revision number for this code set.
rxNormID	Also called the RXCUI which is a normalized id that is defined in the RxNorm standard and commonly used amongst different organizations.

Sample Response

Sample response from the concept detection annotator for the text: She is taking Metformin for her type 2 diabetes.

This example provides the optional medical codes and source vocabularies.

{
  "unstructured": [
    {
      "text": "She is taking Metformin for her type 2 diabetes.",
      "data": {
        "concepts": [
          {
            "cui": "C0025598",
            "preferredName": "Metformin",

Annotators: Cancer (Deprecated)

Annotators: Concept Disambiguation