DICOM Desk

Drop the .dcm files of one DICOM series into your browser, or paste a dcmdump, a gdcmdump or a pydicom print(ds), and work the series to usable in one sitting. A real DICOM Part 10 reader runs free in the browser: it handles the 128-byte preamble and the DICM magic, reads the file meta group as Explicit VR Little Endian regardless of what the dataset then declares, decodes implicit VR, explicit VR little-endian and explicit VR big-endian, walks sequences so a PersonName nested in a content sequence is still found, recognises a raw dataset with no preamble, refuses a deflated one honestly, and stops at (7FE0,0010) so no pixel data is ever decoded. Then it does the arithmetic no header field can give you: the slice normal as the cross product of the two direction cosines in (0020,0037), each slice's position as the dot product of (0020,0032) with that normal, and the slice spacing as the difference between consecutive projections - so a stack whose steps read 1.25, 1.25, 2.50, 1.25 is reported as having a missing slice even though SliceThickness and SpacingBetweenSlices both say 1.25. Gaps, duplicate positions, orientation drift in degrees, InstanceNumber order that disagrees with the geometry, localizers mixed into an acquisition, lossy compression and CT with no rescale are all found and quantified. It also audits every tag on the DICOM PS3.15 Annex E confidentiality profile list, separating direct identifiers from quasi-identifiers from UIDs that need remapping rather than removal, and it shows the values in your browser while withholding them from the model by construction. A series manifest, an instance sort order, a de-identification plan and the parsed header all export with no account and no run. Four AI lanes then run over the same header: preflight the series across eight areas, plan the de-identification tag by tag, plan the volume with its affine and its pitfalls, or design the research manifest and its quality columns - each ending in a runnable pydicom script whose every DICOM tag is checked back against the header you pasted. A derived work inspired by four published agent skills: @nvidia/dicom-series-preflight, @nvidia/dicom-metadata-extract and @nvidia/dicom-series-to-volume (github.com/nvidia/skills), and @k-dense-ai/pydicom. Not affiliated with those authors, with NVIDIA, with the pydicom project or with DICOM's standards body. DICOM Desk is NOT a de-identifier, it never rewrites a file, it cannot see identifiers burned into pixels, and its profile action codes are a transcription of one published baseline rather than legal advice - your data-use agreement governs. Not a medical device; nothing it produces is for clinical use.

Every public app is built from a security-scanned skill and must pass a clean scan — skill and frontend — before it can be listed. Have a skill of your own? Turn it into an app — or read the step-by-step walkthrough.