Table 4 Performance of an i2b2-2014 model with custom embedding tested on 4 different datasets
From: Customization scenarios for de-identification of clinical notes
Test on | PHI type | Embedding | ||
|---|---|---|---|---|
GloVe | embed-mimic | Matching embedding | ||
physionet | All | 76.2/61.3/67.9 | 81.8/64.1/71.8 | 76.9/62.4/68.9 - embed-mimic-nursing |
physionet | Name | 92.9/73.1/81.8 | 95.5/81.8/88.1 | 91.0/73.4/81.3 - embed-mimic-nursing |
mimic-radiology | Name | 92.9/85.7/89.1 | 97.2/85.9/91 | 92.0/87.2/89.4 - embed-mimic-radiology |
mimic-discharge | Name | 92.5/85.4/88.8 | 93.4/89.8/91.6 | 92.1/86.3/89.1 - embed-mimic-discharge |
mimic-echo | Name | 95.5/61.4/74.7 | 98.7/64.4/77.3 | too small to build embedding |