Claude Code Called 'Grade III Tear' an Intact Tendon - Then Proved Itself Right

I gave Claude Code a raw DICOM export of my shoulder MRI and told it nothing except “right shoulder pain for 2–3 weeks.” After an hour of analysis, it reported that my subscapularis tendon was intact — directly contradicting the orthopedist’s Grade III (>50%-width) partial-thickness tear diagnosis.

Why I Grabbed the DICOM Export Instead of Accepting Treatment

The orthopedist didn’t just give me a diagnosis; they started shockwave therapy and injected Traumeel, a homeopathic medicine “without a therapeutic indication” in Germany, within minutes of the MRI. I asked for a copy of the results before leaving. Back home, GPT 5.5 Pro flagged two things: the clinical practice guideline says no shockwave therapy for rotator-cuff tendinopathy without calcification, and the ultrasound showed no calcification. That killed my confidence in the treatment plan.

Claude Code’s Analysis: Intact Tendon vs. Grade III Tear

The MRI package was a standard DICOM export — a few hundred files, 266 MB, no extensions. I used Claude Code with Opus 4.8 (xhigh) to give it the ability to run code and install packages. The difference between Claude Code and the chat interface is enormous, even with the same model underneath. After an hour of autonomous work — no further prompting besides the one-line symptom — it returned a report: my subscapularis tendon was intact. No tear at the apical insertion.

Arbitration Confirmed the AI — But Who Do You Trust?

To settle the disagreement, I had Claude run an arbitration using multiple subagents to avoid bias. I fed it the human report and the earlier discussion with GPT 5.5 Pro where I’d tested movement patterns. After another hour, the arbiter delivered a moderate-to-high confidence verdict: “Mild insertional tendinosis; NO discrete partial- or full-thickness tear identified, including at the apical insertion.” The AI explicitly noted disputes it couldn’t resolve — but on this one, it was decisive.

Where does that leave me? The diagnosis and treatment now look premature and more intervention-heavy than the facts seem to justify. But I don’t know if I can fully trust AI either. My hope is that in a couple of model generations, we’ll trust AI to review MRIs the way we trust it to proofread our emails.

Source: I used Claude Code to get a second opinion on my MRI
Domain: antoine.fi