DICOM & Metadata Quality
What imaging headers actually establish, where they mislead, and what has to be captured at ingestion because it cannot be recovered afterwards.
DICOM is a filing standard that acquired a second career as a measurement record, and it carries the marks of that history. The attributes protecting identity and geometry are strongly obliged and largely reliable. The attributes describing acquisition physics — the ones a quantitative pipeline depends on — are frequently optional, often vendor-specific, and never validated against the pixel data they describe.
Work in this area is mostly about knowing which tier a given field sits in, and building ingestion that records the answer while the answer still exists.
Articles in this topic
What DICOM Metadata Guarantees About a Series, and What It Only Suggests
Type 1, 2 and 3 attributes carry very different obligations, and almost every field a quantitative pipeline depends on sits in the weakest tier. A working taxonomy of which header values you can rely on and which you must verify against the pixel data.
The Header Fields That Break Quantitative Analysis Across Sites
Multi-site quantitative studies rarely fail on the obvious parameters. They fail on rescale handling, reconstruction kernels, units declarations and private diffusion tags. A field-by-field account of where cross-site comparability is actually lost.
Metadata Provenance in Imaging AI Validation Sets
A validation cohort is a claim about independence. De-identification, format conversion and re-curation routinely destroy the evidence needed to support that claim, and the destruction is invisible in every metric the model reports.