dicom-anonymizer

Batch anonymize DICOM medical images by removing patient sensitive information (name, ID, birth date) while preserving image data for research use. Trigger when users need to de-identify medical imaging data, prepare DICOM files for research sharing, or remove PHI from radiology/scanned images.

1.51x

Quality

86%

Does it follow best practices?

Impact

100%

1.51x

Average score across 3 eval scenarios

Securityby

Passed

No known issues

Evaluation results

100%

62%

DICOM De-identification for Multi-Site Research Study

Batch anonymization with audit log

Criteria

Without context

With context

pydicom used

100%

Batch mode flag

100%

Audit log generated

100%

Audit log structure

80%

100%

PatientName is ANONYMOUS

100%

PatientID hashed not cleared

100%

AccessionNumber hashed

100%

PatientSex preserved

100%

PatientWeight preserved

100%

Private tags removed

100%

PatientBirthDate cleared

40%

100%

Without context: $0.4543 · 1m 52s · 23 turns · 23 in / 6,877 out tokens

With context: $0.7224 · 2m 19s · 24 turns · 29 in / 8,420 out tokens

100%

28%

Longitudinal MRI Study De-identification with Cross-Session Linkage

Study linkage preservation with UID hashing

Criteria

Without context

With context

preserve-studies enabled

66%

100%

StudyInstanceUID hashed not regenerated

100%

SeriesInstanceUID hashed

100%

Different patients different UIDs

100%

Pseudonym format ANON_XXXXXX

100%

Same patient same pseudonym

100%

PatientID hashed

50%

100%

Audit log present

70%

100%

Without context: $0.4930 · 1m 59s · 25 turns · 30 in / 7,574 out tokens

With context: $0.9336 · 3m 11s · 31 turns · 256 in / 11,549 out tokens

100%

11%

Metabolic Research DICOM Export with Selective Retention

Selective tag preservation and compliance warnings

Criteria

Without context

With context

keep-tags option used

100%

PatientWeight retained

100%

PatientSize retained

100%

PatientBirthDate cleared

60%

100%

SOPInstanceUID regenerated

100%

StationName removed

75%

100%

DeviceSerialNumber removed

75%

100%

Burned-in annotation warning

100%

Manual QA requirement noted

100%

InstitutionName removed

70%

100%

Without context: $0.6710 · 2m 51s · 28 turns · 33 in / 9,569 out tokens

With context: $1.1037 · 4m 15s · 34 turns · 649 in / 14,327 out tokens

Repository: aipoch/medical-research-skills
Commit: ca9aaa4

Evaluated: 17 days ago
Agent: Claude Code
Model: Claude Sonnet 4.6

Table of Contents

DICOM De-identification for Multi-Site Research Study Longitudinal MRI Study De-identification with Cross-Session Linkage Metabolic Research DICOM Export with Selective Retention

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.