Measured Imaging

Quantitative medical imaging, imaging data quality, DICOM metadata, and the validation of imaging AI.

Medical imaging is treated as a measurement instrument long before anyone establishes that it behaves like one. A pixel value is not a quantity until something ties it to an acquisition, a calibration and a reconstruction, and most of the metadata that would let you make that tie is optional, vendor- specific, or quietly wrong.

This site is about that gap: what imaging data actually supports, what has to be true of the acquisition before a biomarker means anything, and what it takes to validate an imaging model on data whose provenance you can defend.

  1. 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.

    DICOM & Metadata QualityScanner & Multi-site Variability

  2. 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.

    DICOM & Metadata QualityQuantitative Imaging Biomarkers

  3. 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.

    DICOM & Metadata QualityImaging AI Validation

  • 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.