Demos
Discover how John Snow Labs enables secure, scalable DICOM de-identification using AWS HealthImaging and SageMaker.
What is the most secure way to de-identify DICOM files in AWS?
To share medical images safely for research or AI model training, DICOM de-identification must meet high regulatory standards. John Snow Labs provides a turnkey solution that integrates seamlessly with AWS HealthImaging (AHI), enabling high-volume, compliant image processing—without your data ever leaving your secure cloud perimeter.
Why is accurate de-identification essential in healthcare AI?
In healthcare, 80% or 90% de-identification isn’t enough. To meet regulatory and ethical standards, de-identification must be precise. This includes removing Protected Health Information (PHI) embedded in both the image pixels—like handwriting or burnt-in text, and the thousands of metadata fields that vary across radiology modalities.
How does the integration with AWS HealthImaging work?
John Snow Labs’ solution runs entirely within your AWS environment using Amazon SageMaker. It follows three streamlined steps:
-
Pull DICOM files from S3 using AWS libraries.
-
Process files in SageMaker, removing PHI from metadata and image content.
-
Push de-identified files back to S3, ready for research or model development.
No data leaves AWS. No manual intervention is needed. Just scalable, secure, and compliant de-identification.
FAQs
How many images can the solution process?
It’s built for scale, supporting hundreds of millions to billions of images, optimized for both speed and cost.
Does it redact all visual content in DICOM images?
Only PHI is redacted. Clinical indicators like laterality are preserved to retain diagnostic value.
What tools are used in the implementation?
Amazon SageMaker handles processing, and AWS HealthImaging manages data storage and retrieval.
Is the de-identification customizable?
Yes. You can configure what types of PHI to remove depending on your research or compliance needs.
Does the solution maintain compliance with HIPAA?
Absolutely. The design supports regulatory-grade de-identification, ensuring compliance and auditability.
Supplementary Q&A
How does this compare to traditional DICOM de-identification tools?
Most tools require complex setup and may not scale efficiently. John Snow Labs’ solution is plug-and-play with AWS, automates the entire pipeline, and meets the rigorous standards required by healthcare providers and researchers.
Can this be used in academic research settings?
Yes. Many academic medical centers already rely on John Snow Labs’ models. This solution helps researchers quickly create compliant, shareable datasets from existing imaging archives.
What happens if PHI is missed in de-identification?
That’s why accuracy is non-negotiable. The model is trained specifically for medical imaging use cases to ensure complete PHI removal, offering a high degree of confidence for secondary data use.
