Table 4 Evaluation metrics for each of the different neural networks tested
From: De-identifying Spanish medical texts - named entity recognition applied to radiology reports
Training set | Validation set | Test set | |||||||
|---|---|---|---|---|---|---|---|---|---|
Model | Precision | Recall | F1 | Precision | Recall | F1 | Precision | Recall | F1 |
LSTM-CRF | 90.39 | 81.93 | 85.95 | 87.09 | 77.11 | 81.79 | 81.35 | 61.37 | 69.96 |
LSTM-CRF with EMA | 91.19 | 84.15 | 87.53 | 87.05 | 78.49 | 82.55 | 71.48 | 59.65 | 64.96 |
LSTM-LSTM-CRF | 99.20 | 98.79 | 98.99 | 98.13 | 97.18 | 97.66 | 93.01 | 90.94 | 91.96 |
LSTM-LSTM-CRF with EMA | 99.06 | 98.96 | 99.01 | 98.00 | 97.34 | 97.67 | 94.20 | 91.10 | 92.63 |
Conv-LSTM-CRF | 99.31 | 99.05 | 99.18 | 98.11 | 97.29 | 97.70 | 94.49 | 90.43 | 92.41 |
Conv-LSTM-CRF with EMA | 99.17 | 99.05 | 99.11 | 98.08 | 97.36 | 97.72 | 93.72 | 90.64 | 92.15 |
Spacy | 99.87 | 99.28 | 99.58 | 98.06 | 96.10 | 97.07 | 93.23 | 89.39 | 91.31 |