| language: en | |
| library_name: transformers | |
| pipeline_tag: token-classification | |
| base_model: StanfordAIMI/stanford-deidentifier-base | |
| tags: | |
| - clinical | |
| - phi | |
| - ner | |
| # Synthetic PHI Detector | |
| This token-classification model was fine-tuned from `StanfordAIMI/stanford-deidentifier-base` using | |
| Synthetic Synthea-derived clinical templates. It recognizes three entity types | |
| with the BIO labels stored in `config.json`. | |
| Training examples: 5997 | |
| | Entity | Strict seqeval F1 | | |
| |---|---:| | |
| | MEDICAL_RECORD | 1.0000 | | |
| | DIAGNOSIS | 1.0000 | | |
| | MEDICATION | 1.0000 | | |
| Overall strict seqeval F1: 1.0000 | |
| ## Limitations | |
| Synthetic templates do not represent every institution, document style, or | |
| identifier format. False negatives and false positives are expected. This | |
| model assists PHI detection and is not a compliance guarantee or substitute | |
| for human review and layered detection controls. | |