- Talked about DICOM anonymization tools
- CTP is of course a well known one, but not super user friendly
- DicomCleaner is from Dr. David Clunie, very capable, can also do pixel redaction and has a GUI so it is user friendly.
- Orthanc Can do anonymization via GUI and also API at any level (patient, study, series, instance), but no pixel redaction
- Horos Can also do it, but Mac only
- Sante Have viewers and tools for batch anonymization
- Mirth Connect is another option for scripting
- DCMTK is another option for command-line options that can be batched
- Typical pixel redaction (e.g. off Ultrasounds) vs. preventing the reconstruction of a patient's face from something like a head CT scan.
- Paper on deidentification of faces: https://pubmed.ncbi.nlm.nih.gov/17295313/
- Could AWS Rekognition be used to find text on a DICOM image then redact it?
- Patient names showing up in the wrong places (random DICOM tags, radiology reports)
- Was a popular practice before the advent of structured reporting
- HL7 defines what fields to anonymize, and points out jurisdictional differences.
- Non-obvious use cases like health card numbers in Ontario
- DICOM also defines "profiles" for de-identification
- Difference between de-identification vs. anonymization
- Ethics rules that require a mapping between original identifiers and anonymized ones, in case the research study uncovers new findings that were not spotted when the original study was performed
- Talked about SIIM
- Free membership to students and those in training
- So many different groups doing different things: migrations, podcast, machine learning, etc