I have looked into the medcat paper and I would still like to ask, apart from reported better F-score, what makes medcat a better solution than ctakes?
hi @bkakke - we’re obviously a bit biased here - but MedCAT is superior not only in terms of F-Score also:
- MedCAT is terminology flexible. UMLS, SNOMED CT, ICD, OPCS, RadLex, HPO and so on. cTakes only supports UMLS.
- You can build your own custom ontology with MedCAT - you can build from scratch, or you can extend an existing ontology via the MedCATtrainer interface.
- MedCAT is language agnostic - if you can represent it in unicode - great MedCAT can annotate it. We’ve had users annotate Dutch, french, spanish, korean and chinese, alongside english. This is because MedCAT doesn’t rely on UMLS and american focused ontology.
- MedCAT is easily installable and integrated into a clinical NLP pipeline via MedCATService, and small lightweight library.
- self-supervised and supervised training for model fine-tuning. Annotation collection and viewing is possible via the Trainer interface
- model pipelines are transferable via a couple lines of code and completely self-contained .zip file.
- configurable meta-annotations (cTakes nomenclature - property)) - its possible to define meta-annotations in the Trainer and collect any annotations for any definable task.
- its an actively used and maintained project - we have a growing team of developers and users improving the tools on a daily / weekly basis.