Developer uses Claude Code Opus 4.8 to analyze personal MRI and disputes clinical diagnosis

Antoine, a software engineer, documented an experiment using Anthropic’s Opus 4.8 via Claude Code to analyze a personal shoulder MRI and generate a second-opinion report. The MRI package was a standard DICOM export containing a few hundred files totaling around 266 MB.

Opus 4.8 was instructed to review the MRI and produce a detailed report based on the context “right shoulder pain for 2–3 weeks.” After about an hour, it returned a PDF report concluding the subscapularis tendon was intact, contradicting the clinic’s diagnosis of a Grade III (>50%-width) partial-thickness tear at the apical insertion.

To adjudicate, the developer had Opus 4.8 compare the human radiology report with prior AI-generated movement tests and produce a second arbitration report. This report concluded with moderate-to-high confidence that there was mild insertional tendinosis and no discrete partial- or full-thickness tear, including at the apical insertion.

Separately, the author noted that the clinic had administered shockwave therapy despite a clinical practice guideline advising against it for rotator-cuff tendinopathy without calcification, and had injected Traumeel, a product registered in Germany as a homeopathic medicine without a therapeutic indication.

The developer emphasized the experiment is not medical advice and acknowledged that both the AI and human interpretations could be wrong, urging readers to interpret the results cautiously.