Identifying Terms of Interest in Radiology Reports

Synopsis

This query finds all terms labelled as disease, findings, and morphologic abnormality SNOMED entity types in the IMPRESSION section of Radiology reports. Entity type searches are less specific than concept search but are useful for finding entities for further analysis and filtering.

Processing Requirements

This query assumes that at least the following emtellipro-db-client processing options were enabled:

--feature snomed-ontology
--store-sections-and-sentences

For this query, Radiology imaging studies were processed with the following document type:

--category Radiology
--subcategory CT

Clinical Context

Medical ontologies like SNOMED classify their concepts into semantic types such as disorder, morphologic abnormality, findings, substance, procedure, etc. The NLP API reports this information separately from the concept itself as the entity_type field in its output, which allows users to filter for concepts based on this information retrospectively.

A common data science or clinical application scenario is one where a clinician wants to know, very generically, what medical problems a patient has had in their past. One simple way is to just search an Output Database for any concepts that are asserted, which have a SNOMED semantic type of disorder, morphologic abnormality, or finding. While this is a somewhat broad definition of medical problems (as many findings may be benign), it can be used with high-recall to identify almost any medical issue of concern that NLP API has coded with SNOMED CT. The results return can be reduced by searching for disorder only, or disorder and morphologic abnormality if higher precision is required.

SQL query

SQL

Microsoft SQL Server

Snowflake

-------------------------------------------------------------------------------------------------------------
-- Retrieve disease/findings/morphologic abnormality in IMPRESSION section of Radiology reports
-- that are 'asserted' and 'certain'
-------------------------------------------------------------------------------------------------------------
SELECT d.category AS document_category,
    c.concept_id,
    c.description,
    fe.section_name AS section,
    fe.text AS annotated_term,
    fe.polarity,
    fe.uncertainty,
    sl.text AS sentence
FROM foundentity fe
        JOIN foundentityconcept fec ON fe.id = fec.found_entity_id
        JOIN concept c ON fec.concept_id = c.concept_id AND fec.concept_ontology = c.ontology
        JOIN foundentitytype fet on fe.id = fet.found_entity_id AND fet.ontology = 'snomed'
        JOIN foundentitylocation fel ON fe.id = fel.found_entity_id
        JOIN location l ON fel.location_id = l.id AND l.type_ = 'sentence'
        JOIN sentencelocation sl ON sl.id = l.location_id
        JOIN document d ON l.document_id = d.id
WHERE c.ontology = 'snomed'
AND fet.type_name IN ('disorder', 'finding', 'morphologic abnormality') -- Snomed CT entity types
AND d.category = 'Radiology'
AND fe.section_name = 'IMPRESSION'                                      -- REPLACE: with section of interest e.g. FINDINGS, INDICATIONS
AND fe.polarity = 'asserted'
AND fe.uncertainty = 'certain';
-- LIMIT 1000; -- RECOMMENDED: for very large data sets

document_category	concept_id	description	section	annotated_term	polarity	uncertainty	sentence
Radiology	40199007	Supine body position (finding)	IMPRESSION	supine	asserted	certain	No obvious pneumoperitoneum or bowel obstruction, but evaluation is limited on these supine views.
Radiology	64613007	Inflammation of small intestine (disorder)	IMPRESSION	enteritis	asserted	certain	Several fluid-filled loops of jejunum are present compatible with mild enteritis.
Radiology	396347007	Umbilical hernia (disorder)	IMPRESSION	umbilical hernia	asserted	certain	Fat containing umbilical hernia.
Radiology	197321007	Steatosis of liver (disorder)	IMPRESSION	Fatty liver	asserted	certain	Fatty liver.
Radiology	89977008	Increased thickness (finding)	IMPRESSION	thickening	asserted	certain	Subtle area of thickening of the lower rectal wall.
Radiology	300312006	Lesion of rectum (disorder)	IMPRESSION	rectal wall thickening	asserted	certain	Subtle area of thickening of the lower rectal wall.
Radiology	76197007	Hyperplasia (morphologic abnormality)	IMPRESSION	hyperplasia	asserted	certain	Multiple visualized nodes of the bilateral deep lateral cervical nodal chain, within normal size and morphology, most compatible with mild hyperplasia.
Radiology	4147007	Mass (morphologic abnormality)	IMPRESSION	mass lesion	asserted	certain	Large, well demarcated mass lesion of the deep lobe of the left parotid gland, with probable involvement of the left facial nerve.
Radiology	53461003	Normal size (finding)	IMPRESSION	normal size	asserted	certain	Multiple visualized nodes of the bilateral deep lateral cervical nodal chain, within normal size and morphology, most compatible with mild hyperplasia.
Radiology	312894000	Osteopenia (disorder)	IMPRESSION	osteopenia	asserted	certain	Severe osteopenia.

Next steps

To refine the query, one practical strategy that may be explored is to limit the search to specific report sections only. The section of the report where a diseased is mentioned typically provides important context that alters the interpretation and significance of the disease mention.

The NLP API normalizes a large set of recognized section headings into standardized section names. To see a list of normalized section names present in your data, you can use the provided here. Select one or more section names to use as a query constraint.

To search for these entity types in a specific section, simply relace the fe.section_name WHERE clause with appropriate section name.