Medical Data
De-identification
Automatically de-identify structured data, unstructured data, documents, PDF files, and images in compliance with HIPAA, GDPR, or custom needs
Medical Data
>99% Accuracy on real-world document
How Providence Health De-Identified 700 Million Patient Notes with Spark NLP
Accuracy:
99.19that is, above human accuracy
Performance:
2.46hours
to de-identify 500K patient notes.
Explore a full range or features
| John Snow Labs’ De-identification solutions | AWS Medical Comprehend | Microsoft Presidio | Google DLP | |
|---|---|---|---|---|
| De-dentification tool | ||||
| End-to-end service | ||||
| Available also as a standalone library | ||||
| Established new state of the art accuracy in peer reviewed publication | ||||
| Real world reference with >99% correctly recognized PHI | ||||
| Scanned PDF | Integrated | Separate service | Separate service | |
| DICOM | Integrated | Separate service | Separate service | |
| Obfuscation | ||||
| Multilingual support | ||||
| Built on big data framework | ||||
| Possible to fine tune standard pre-trained models | ||||
| Data does not leave your premise | ||||
| Works in air gap insulated server with no internet access |
- Entities available out of box:
ACCOUNT, AGE, BIOID, CITY, CONTACT, COUNTRY, DATE. DEVICE, DLN, DOCTOR, EMAIL, FAX, HEALTHPLAN, HOSPITAL, ID, IDNUM, IPADDR, LICENSE, LOCATION, LOCATION-OTHER, MEDICALRECORD, NAME, ORGANIZATION, PATIENT, PHONE, PLATE, PROFESSION, SSN, STREET, STATE, URL, USERNAME, VIN, ZIP
- Easy to add other entities.
- Works with virtually any input – text, scanned PDF, DICOM, docx, pptx.
The De-identification Service
1
Analyze
- Risk analysis​
- Legal requirements review
- HIPAA Safe Harbor, HIPAA Expert Determination​
- CCPA​
- GDPR pseudoanonymization, GDPR anonymization
- Quality assurance strategy & process
2
Identify
- ID, name, email, patient ID, SSN, credit card, address, birthday, phone, URL, license number
- Physician name, hospital name, profession, employer, affiliation
- Racial or ethnic origin, religion, political or union affiliation, biometric or genetic data, sexual practice or orientation
3
Measure
- Cleanroom AI Platform (on-site)
- Annotation tool
- Active learning
- Accuracy Measurement & agreement processes
- Correct sampling
- Multi-lingual
4
De-identify
We support:
- Tabular (headers, values)
- Text (NER, text matching)
- PDF: Text or Scanned
- Images (OCR & metadata)
- DICOM (OCR & metadata)
So you can:
- Replace (or delete a field)
- Mask (hash identifiers or shift dates)
- Obfuscate (name, locations, organizations)
- Generalize (disease codes, dates, addresses)
5
Monitor
- Ongoing measurement & model improvement
- Missed sensitive data
- Incident response
- GDPR & CCPA requests
- Emergency unblinding
- Audits
