Hi MedCAT-team! I just wanted to let you know our paper in which we used MedCAT for negation detection in Dutch medical text has been published:
Thanks for your help in the process!
1 Like
Great work @sandertan and team. I’ll add to our news items in the MedCAT project
Congrats @sandertan and team!! Nice breakdown across the different clinical document types. I think its an important fact that is often overlooked. Differences between documents within a hospital let alone across different hospitals!
A bit of an off topic question:
I was wondering if anyone has had a brief look or a discussion of the Value of BiLSTM vs RoBERTA models?
I.e. Although performance of RoBERTA is > BiLSTM, BiLSTM requires less resources to train. So maybe the trade off is worth it? Would love to hear your thoughts on this.
@anthony.shek With value, you mean value for usage in clinical practice? We briefly discussed this in the “Additional aspects for model comparison” section. Because the biLSTM is easier to train and implement (computationally less resources and we can use the MedCAT pipeline) and it works good enough (our users are not complaining (yet…) about missed negations), we’re currently using the biLSTM in production.
1 Like