Zheng Yuan, Chuanqi Tan, Songfang Huang
Automatic ICD coding is defined as assigning disease codes to electronic medical records (EMRs). Existing methods usually apply label attention with code representations to match related text snippets. Unlike these works that model the label with the code hierarchy or description, we argue that the code synonyms can provide more comprehensive knowledge based on the observation that the code expressions in EMRs vary from their descriptions in ICD. By aligning codes to concepts in UMLS, we collect synonyms of every code. Then, we propose a multiple synonyms matching network to leverage synonyms for better code representation learning, and finally help the code classification. Experiments on the MIMIC-III dataset show that our proposed method outperforms previous state-of-the-art methods.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Medical Code Prediction | MIMIC-III | Macro-AUC | 95 | MSMN |
| Medical Code Prediction | MIMIC-III | Macro-F1 | 10.3 | MSMN |
| Medical Code Prediction | MIMIC-III | Micro-AUC | 99.2 | MSMN |
| Medical Code Prediction | MIMIC-III | Micro-F1 | 58.4 | MSMN |
| Medical Code Prediction | MIMIC-III | Precision@15 | 59.9 | MSMN |
| Medical Code Prediction | MIMIC-III | Precision@8 | 75.2 | MSMN |
| Multi-Label Classification | MIMIC-III | Macro-AUC | 95 | MSMN |
| Multi-Label Classification | MIMIC-III | Macro-F1 | 10.3 | MSMN |
| Multi-Label Classification | MIMIC-III | Micro-AUC | 99.2 | MSMN |
| Multi-Label Classification | MIMIC-III | Micro-F1 | 58.4 | MSMN |
| Multi-Label Classification | MIMIC-III | Precision@15 | 59.9 | MSMN |
| Multi-Label Classification | MIMIC-III | Precision@8 | 75.2 | MSMN |