AI-Powered PHI Detection in Medical Images
Protecting patient privacy through advanced computer vision and language models
This research evaluates AI-based systems for detecting Protected Health Information (PHI) in medical imagery, addressing critical healthcare data privacy concerns.
Key Findings:
- Developed a three-component pipeline for PHI detection: text localization, extraction, and analysis
- Tested across 1,000 medical images spanning CT, X-ray, bone scan, and MRI modalities
- Evaluated multiple vision and language models for optimal PHI identification
- Created a simulation framework with 8 PHI and 8 non-PHI categories
This research is vital for healthcare institutions to maintain HIPAA compliance while sharing medical images for research or clinical purposes, mitigating privacy risks in an increasingly digital healthcare environment.