OpenMed 1.8 Released: On-Device Clinical Data De-identification

OpenMed 1.8, an open-source clinical NLP toolkit, has been released. This toolkit prioritizes data privacy, with all data processing occurring fully locally on the user's hardware without any cloud calls.

Key Updates:

• Multi-Platform Support: Supports Android (Kotlin, ONNX Runtime Mobile), iOS (Swift), React Native bridges, and browser environments (Transformers.js, WebGPU/Wasm).
• Enhanced De-identification: Includes functionality to remove names, MRNs, dates, and other identifiers from documents, along with a verify-pdf feature that addresses vulnerabilities in PDFs with residual text layers. It also offers a feature that uses OCR to read and de-identify text embedded in pixels within DICOM files.
• Model Performance: Supports over 1,500 Apache 2.0-licensed models, including models that ranked at the top in PII (Personally Identifiable Information) masking benchmarks.
• Lightweight Model: A compact model with 44M parameters enables smooth operation on mobile devices.