How Clario technology detects PHI/PII in DICOM images using Amazon Bedrock

Summary

Clario developed an automated PHI and PII detection solution on AWS to process thousands of DICOM image slices and PDF files in clinical trials. Medical imaging workflows require strict privacy compliance under HIPAA, GDPR, and ICH E6 guidelines across standard tags, custom vendor metadata, and pixel-burned text. To address this, the pipeline retrieves files from Amazon S3 and runs memory-intensive detection workloads on Amazon EKS backed by Amazon RDS for PostgreSQL. Amazon Textract handles optical character recognition, while Anthropic's Claude Sonnet on Amazon Bedrock classifies sensitive data using domain-optimized prompt engineering. The system identifies precise coordinates and sensitive entity types rather than directly altering files, enabling human-in-the-loop review before downstream redaction occurs.

Context

Clinical trial DICOM image slices and PDFs often contain sensitive PHI and PII in standard tags, custom vendor metadata, or burned directly into pixels. Every slice must be de-identified in compliance with HIPAA, GDPR, and ICH E6 guidelines before moving downstream, but manual spot-checks risk compliance gaps and inconsistencies across globally distributed sites.

Approach / What changed

Clario built an AWS-native architecture using Amazon API Gateway, Amazon S3, and Amazon EKS backed by Amazon RDS for PostgreSQL. The pipeline combines Amazon Textract for text extraction with Anthropic's Claude Sonnet on Amazon Bedrock for PHI/PII classification, separating entity detection and coordinate generation from downstream pixel masking to permit human review.

Takeaways

• Separating PHI/PII detection and coordinate mapping from actual pixel masking allows human experts to review findings before irreversible changes are applied.
• The effective detection pipeline pairs Amazon Textract for text extraction with Claude Sonnet on Amazon Bedrock for PHI/PII classification using domain-optimized prompts.
• Investing in a manually validated ground truth dataset was essential for building a reliable, automated evaluation pipeline at production scale.

Related reading

• AWS · Aug 19, 2026

  AI-powered clinical trial eligibility and safety using Amazon Bedrock AgentCore

  Manual chart review across fragmented clinical sources slows clinical trial enrollment, contributing to timeline delays and high screen failure rates. To address this bottleneck, an architecture on AWS automates eligibility and safety assessments while preserving human clinician authority. AWS HealthLake ingests and normalizes records into FHIR R4 resources, while Amazon Bedrock AgentCore orchestrates pre-screening, detailed screening, and site enrollment agents. Amazon Bedrock AgentCore Evaluations scores screening determinations using an LLM-as-a-judge approach for clinical accuracy, operational effectiveness, and safety compliance. This human-in-the-loop workflow produces structured recommendations with source citations, reducing patient matching time from days to minutes while maintaining compliance audit trails.

  Sachin Jain

• AWS · Jul 22, 2026

  Automate custom PII detection at scale with Amazon Macie and Step Functions