What happens between dragging in a folder of DICOM files and reading a sentence you understand.

The difference between this and pasting a screenshot into a chat window is not the model — it is everything around the model. Read Your Scan opens the real study and works from the scan's own geometry, so it knows which plane a slice is and which side is the patient's right, instead of guessing from a flat picture.


1 · De-identify in your browser

Identifying DICOM tags — name, patient IDs, dates — are stripped on your device before anything is uploaded. The models never see who you are.

<aside> ℹ️

Limit worth knowing: text burned into the image pixels themselves (some ultrasound and X-ray exports do this) is not a DICOM tag and cannot be stripped this way.

</aside>

2 · Read every series

The exam is grouped into its acquisitions, and each one is read slice by slice by a frontier vision model. The anatomical plane comes from the volume's own geometry, not from the appearance of the image, and laterality comes from the file's metadata rather than being inferred from the picture.

3 · Merge, grade, synthesize

Findings seen on several series are merged into one. A separate pass grades severity from the reading's own wording. A final pass writes the plain-language report. Every finding keeps a confidence level and a clickable location — click it and the viewer jumps to the exact slice it came from.

4 · Cross-check — the second opinion

On request, a second independent model reads the same exam from scratch, and the two readings are shown side by side. Agreement and disagreement are both visible; neither is hidden.


Which models do the reading

No model here is our own. The roster is Anthropic's Claude Opus, Google's Gemini and Google's MedGemma 1.5 — compared head to head on real studies with a known diagnosis, with the one that wins on a given kind of study doing the work. The exact model version is stored alongside every analysis.

What is ours is the pipeline: opening the real DICOM study rather than a screenshot, choosing which series and slices are diagnostic, taking laterality from metadata, scoring each finding, placing it back on the slice it came from, and matching it against the radiologist's report.

Full write-up, including the limits of each step: readyourscan.com/methodology