Medical NLP Stack OpenMed 2.0 Released

OpenMed 2.0, a medical NLP stack, has been released under the Apache-2.0 license. The core of this update is PHI (Protected Health Information) de-identification functionality that can be performed in a local environment without sending data to cloud APIs.

Key features include:

• Lightweight models: Uses encoders with 279M and 560M parameters to perform efficient de-identification tasks, offering higher security than cloud-based APIs.
• Expanded multilingual support: Language coverage has expanded from 23 to 56, with newly added PII models for Bengali (279M), Chinese (560M), and Tamil (279M).
• Browser execution support: Provides artifacts that can run directly in browser environments via Transformers.js.
• Identifier validation: Includes validation for 51 National Identifiers, supporting sophisticated validation that uses actual checksums rather than simple regular expressions.
• Data anonymization risk management: Beyond simple identifier removal, it introduces a structured release risk workflow that assesses re-identification risk through data linkage and helps prevent it.