Skip to main content
Search

# Clinical Text De-identification

1 related articles

Subscribe to updates
  1. arXiv

    TIDE 2.0 separates an exchangeable recognizer from a keyed anonymizer, reaching span-level recall 0.88 and 0.77 on two institutions' corpora

    TIDE 2.0 is an MIT-licensed, institution-hosted engine for clinical-note de-identification that separates an interchangeable PHI recognizer from a keyed, table-free anonymizer: the anonymizer applies a per-patient interval-preserving date shift, format-preserving encryption and deterministic keyed selection so that the same value receives the same surrogate across a patient's notes under one key and releases under a new key cannot be linked to earlier ones; the default configuration, a distilled recognizer plus a regex layer, reached span-level recall 0.88 at precision 0.88 on Stanford's SHIELD corpus and 0.77 at 0.87 on a second institution's i2b2 2014 corpus.

You’re all caught up