S

Stanford Deidentifier Only Radiology Reports

Developed by StanfordAIMI
Automated radiology report de-identification system based on transformer and rule-based methods, capable of detecting PHI entities and replacing them with realistic values
Downloads 26
Release Time : 6/9/2022

Model Overview

This model is specifically designed for de-identification processing of medical radiology reports. By combining the PubMedBERT transformer model with rule-based methods, it automatically detects Protected Health Information (PHI) and performs secure replacements in compliance with HIPAA privacy standards.

Model Features

Hybrid Architecture
Combines PubMedBERT transformer model with 'Hidden in Plain Sight' rule-based methods to achieve high-precision PHI detection and replacement
Multi-institutional Validation
Trained on 6,193 cross-institutional medical documents, including chest X-rays, CT reports, and medical records
Production-grade Accuracy
Achieves 97.9 F1-score on known institutional radiology reports and 99.6 F1-score on new institutional test sets

Model Capabilities

Medical Entity Recognition
Protected Health Information Detection
Realistic Value Replacement
Radiology Report Processing
Cross-institutional Generalization

Use Cases

Medical Data Privacy Protection
Radiology Report De-identification
Automatically identifies and replaces PHI such as patient names, doctor names, and contact information
Outperforms manual annotation performance on the i2b2 2014 dataset
Multi-center Research Data Sharing
Securely processes cross-institutional medical documents to comply with privacy regulations
Supports medical data transmission systems like MedClinical
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase