Reproduce “De-Identification of Medical Imaging Data: A Comprehensive Tool for Ensuring Patient Privacy” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Moritz Rempe, Lukas Heine, Constantin Seibold, Fabian Hörst, Jens Kleesiek
An open-source Python and Docker framework consolidates metadata sanitization, skull-stripping, defacing, OCR-based text removal, and whole-slide image handling across DICOM MRI/CT, DICOM-WSI, NIfTI, and Siemens twix data. Its neural models achieved 80.62% defacing and 94.00% skull-stripping scores, while reducing processing to under a second per volume on GPU for the tested tasks.
Key insights
- Protecting medical imaging data requires addressing both structured metadata and information embedded in pixels; sanitizing only standard DICOM tags can leave identifiers in raw-data headers or image annotations.
- The defacing model was faster than comparison methods and performed better on the reported recognition-based metric, although its evaluation inherited limitations from pydeface-derived targets and face-recognition scoring.
- Speed came with trade-offs: the framework approached but did not surpass the strongest skull-stripping baselines, and residual skull structures remained in some infant scans.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.