DICOM Anonymization & HIPAA De-identification

Overview

dicomtk implements HIPAA Safe Harbor and the DICOM PS3.15 Basic Application Level Confidentiality Profile via dicom_anonymize(). Unlike legacy wrappers that invoke external dcmodify subprocesses, dicomtk performs de-identification natively in C++17 memory.

library(dicomtk)
library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union

# Create a synthetic DICOM file containing PHI tags
mat <- matrix(seq(0, 255), nrow = 16, ncol = 16)
raw_dcm <- tempfile(fileext = ".dcm")
write_dicom(
  mat,
  raw_dcm,
  metadata = list(
    PatientName = "DOE^JOHN",
    PatientID = "MRN-987654",
    PatientBirthDate = "19800415",
    StudyDate = "20251016",
    InstitutionName = "GENERAL HOSPITAL",
    Modality = "CT"
  )
)

# Inspect original PHI tags
dicom_tag_value(
  raw_dcm,
  c("PatientName", "PatientID", "PatientBirthDate", "StudyDate", "InstitutionName")
)
#>        PatientName          PatientID   PatientBirthDate          StudyDate 
#>         "DOE^JOHN"       "MRN-987654"         "19800415"         "20251016" 
#>    InstitutionName 
#> "GENERAL HOSPITAL"

# Anonymize with longitudinal date shifting (-100 days), UID regeneration, and private tag removal
anon_dcm <- dicom_anonymize(
  raw_dcm,
  patient_name = "TRIAL^SUBJECT_001",
  patient_id = "SUBJ_001",
  shift_days = -100L,
  erase_private = TRUE,
  regenerate_uids = TRUE
)

# Verify de-identification
dicom_tag_value(
  anon_dcm,
  c("PatientName", "PatientID", "StudyDate", "InstitutionName", "PatientIdentityRemoved")
)
#>            PatientName              PatientID              StudyDate 
#>    "TRIAL^SUBJECT_001"             "SUBJ_001"             "20250708" 
#>        InstitutionName PatientIdentityRemoved 
#>                     NA                  "YES"