I used Claude Code to get a second opinion on my MRI
Points and comments are a snapshot, not live.
AI analysis of an MRI found no tear, contradicting a doctor's diagnosis of a Grade III partial-thickness tear.
The author, experiencing right shoulder pain, received an MRI and a diagnosis of a Grade III partial-thickness tear of the subscapularis tendon, accompanied by shockwave therapy and a homeopathic injection. Suspicious of the aggressive treatment, the author fed the DICOM MRI data to Opus 4.8 via Claude Code. The AI reported an intact tendon instead of a tear. An arbitration process with more context concluded with moderate-to-high confidence that there was no discrete tear, only mild tendinosis. The author remains uncertain whether to trust the AI or seek another doctor.
What commenters are saying
Commenters heavily cautioned against using LLMs for medical image analysis, especially 3D modalities like MRI. A radiologist noted that ultrasound is poor at detecting calcification, and shockwave therapy without calcification is merely not helpful, not harmful. Another commenter emphasized that LLMs are weak at spatial reasoning and microscopic analysis, though they excel at text extraction from images. The thread split between those who view AI as dangerously unfit for this task and those who found it useful for simpler image interpretation.