Compare an MRI report with the scan itself

A person used Claude Code to review a shoulder MRI and then compare its read with the doctor’s report.

Done with Claude Code· Advanced

What they did
They gave Claude Code a shoulder MRI DICOM export of about 266 MB, allowed it to install packages, and shared only the symptom: 2–3 weeks of pain in the right shoulder. The agent spent about an hour planning and examining the images. They then ran a second pass with subagents to weigh the agent’s report against the clinician’s version, adding more context for the comparison.
How it went
Claude Code judged the tendon to be intact, while the doctor wrote up a Grade III partial tear. The arbitration leaned toward Claude Code with moderate-to-high confidence, but the person still did not feel sure either way and kept going with physical therapy. Each run took roughly an hour, and the result remained unverified.
Tips
Treat this kind of output as a second opinion prompt, not a diagnosis. If the AI and the radiologist disagree this strongly, a third human opinion is still important.

Try it yourself with Claude Code

I have MRI files in [folder path] and a radiology report in [report file]. Read the images and the report, explain the report in plain language, and list the findings that matter most for [symptom or concern]. Then compare the scan-based impression with the written report and say where they agree or differ. Include questions I should ask my doctor. Do not give a diagnosis, and clearly state any limits of what you can tell from the images alone. Before any irreversible action, pause and ask me to confirm.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.