Table 4 Performances for each PHI category achieved by the customized LSTM-CRFs model using fine-tuning
From: A study of deep learning methods for de-identification of clinical notes in cross-institute settings
Entity Type | Performance on UF test set | |||||
|---|---|---|---|---|---|---|
Strict | Relax | |||||
Precision | Recall | F1 score | Precision | Recall | F1 score | |
DATE | 0.9807 | 0.977 | 0.9789 | 0.985 | 0.9813 | 0.9831 |
AGE | 0.9861 | 0.8659 | 0.9221 | 0.9861 | 0.8659 | 0.9221 |
ID | 0.9173 | 0.8905 | 0.9037 | 0.9624 | 0.9343 | 0.9481 |
NAME | 0.9029 | 0.8807 | 0.8917 | 0.9694 | 0.9455 | 0.9573 |
PHONE | 0.9048 | 0.8085 | 0.8539 | 0.9762 | 0.8723 | 0.9213 |
ZIP | 0.75 | 0.75 | 0.75 | 0.9 | 0.9 | 0.9 |
INSTITUTE | 0.75 | 0.5042 | 0.603 | 0.9375 | 0.6303 | 0.7538 |
CITY | 0.9048 | 0.4222 | 0.5758 | 1 | 0.4667 | 0.6364 |
STREET | 0.55 | 0.5238 | 0.5366 | 0.85 | 0.8095 | 0.8293 |
WEB | 0 | 0 | 0 | 0 | 0 | 0 |