MIMIC-IV-Ext-SEQ is a benchmark for foundation models in Intensive Care, representing the journey of an intensive care patient as a sequence of event tokens with optional event intensity markers designed to make it easy to train sequence models (Transformers, etc.). See the paper for more details.
- Obtain access to MIMIC IV dataset if you haven’t already. You will have to sign up on PhysioNet, accept the data use agreement and take a small course.
- Go to your profile settings on PhysioNet, select the “Cloud” tab and specify your Google account.
- Download the data from here.
See example.py
reproduce directory contains the Google BigQuery SQL requests that we ran on MIMIC IV database to create our dataset. It is not required to use MIMIC-IV-Ext-SEQ, but included for transparency and reproducibility.