I have been experimenting with relationships between concepts, and have started trying out annotating the following semantic relationships:
This is aimed at capturing anatomical localisation, chronological change and adjectives describing a concept with more features. Works reasonably well, but Relations doesn’t seem to save. Bug report
Also, while one can see which concept is being annotated when one clicks on the concept, it gets very confusing when there is more than one named the same way, see below:
Feature request This would be alleviated with a colour scheme, but this would conflict with the main MedCat function (unless there is a toggle which turns on/off Relationship displays)
Looks like an interesting use case of the Relation annotation collection in MedCATtrainer.
A given relation if selected on the right hand side adds little ‘start’ / ‘end’ labels on the top right of the relation to help tell one set of relations from another, although this still isn’t as clear as it could be…
Looks interesting… Other than relationships between two identified concepts. Is it possible to link a concept with a numerical or date value? e.g. fever, cough on the 01/04/2020. Where there are links between each symptom concept and the date?
One could imagine doing this through a relationship with this
ConceptB is a numeric value of ConceptA
ConceptB is a date value of ConceptA
But this means ConceptB needs to be created as a custom non-snomed concept of a number/date. These non-snomed concepts won’t work very well since they would have infinite synonyms.
In light of the demand for this functionality, I have been thinking some more.
SNOMED has the concept domain of a “qualifier”, concepts which act as attributes to existing concepts. CUI 362981000
Perhaps instead of inventing a whole relational annotation system, one would only annotate the Qualifier and the main concept and then annotate the Relation as a “Snomed Qualifier”.
In this way, future RelCat annotations might even begin to resemble elements of post-coordinated Snomed expressions.
The “fly in the ointment” to this is that it may not work so well if the attributed concept isn’t a Qualifier but an anatomical term or something else.
