NA-MIC Project Weeks

Back to Projects List

Fan mask and OCR bound box editing for OHIF-based DICOM De-Identification verification mode

Key Investigators

Dave Dinh (BWH, Brigham and Women's Hospital, USA)
Deepa Krishnaswamy (BWH, USA)
Tina Kapur (BWH, USA)
Andras Lasso (Queen's, Canada)
Steve Pieper (Isomics, USA)

Project Description

Optimal DICOM de-identification for downstream research requires balancing removal and retention. We must scrub embedded PHI in both tags and pixels while preserving valuable non-PHI metadata (e.g., scan settings like depth). Even with both addressed, human review is still needed to catch misses.

We’re building a closed-loop system with two components:

an automated pipeline that scrubs DICOM tags (PS3.15 with configurable overrides), masks pixel PHI by manufacturer, extracts non-PHI pixel data, and produces deterministic, reproducible outputs
an OHIF-based reviewer for human verification. Each review records passes, failures, and false positives/misses, which feed back as rule updates so errors don’t recur.

The pipeline handles most de-identification automatically, but reviewers still face significant cognitive load when inspecting every metadata diff, OCR-flagged region, and fan crop. The OHIF mode currently surfaces areas in the DICOM metadata that require verification based on predefined rules but reviewers also need tools to edit PHI masks, OCR bounding boxes for non-PHI extraction, and fan geometry for fan-only pixel extraction.

Objective

The OHIF-based mode should allow users to edit PHI / fan masks and OCR bounding boxes.
The pipeline uses edits to self-update and re-run flagged DICOMs for the reviewer to re-verify.