I Used Claude Code to Get a Second Opinion on My MRI
Meta Description: I used Claude Code to get a second opinion on my MRI results — here's what happened, what AI can and can't do, and what you should know before trying it yourself.
TL;DR: I uploaded my MRI report to Claude Code out of curiosity after receiving an anxiety-inducing radiology report. The AI provided genuinely helpful context, flagged terminology I hadn't understood, and suggested specific questions to ask my doctor — but it was clear about its limitations and never tried to replace a medical diagnosis. Here's the full breakdown of what worked, what didn't, and whether you should try it.
Why I Turned to AI After My MRI
Let me set the scene: It's a Tuesday afternoon, and I'm staring at a radiology report full of phrases like "mild T2 hyperintensity," "no acute intracranial abnormality," and "incidental finding of a small arachnoid cyst." My neurologist appointment wasn't for another three weeks. My anxiety was at a ten.
Sound familiar?
Millions of people now have direct access to their medical records through patient portals like MyChart, but the reports themselves are written for radiologists and physicians — not for patients. The result is a growing gap between data access and data comprehension. I decided to do what any tech-adjacent person in 2026 would do: I turned to AI.
Specifically, I used Claude Code — Anthropic's powerful AI coding and analysis tool — to help me make sense of what I was reading. What followed was one of the more genuinely useful AI experiences I've had, and also one of the most instructive in terms of understanding where AI assistance ends and medical expertise begins.
[INTERNAL_LINK: how to use Claude Code for non-coding tasks]
What Is Claude Code, and Why Use It for This?
Claude Code is Anthropic's agentic AI tool, originally designed for software development tasks. But by mid-2026, it's evolved into something much broader: a powerful analytical assistant capable of processing documents, interpreting complex text, and reasoning through multi-layered information.
Unlike a basic chatbot, Claude Code can:
- Process long-form documents in their entirety without losing context
- Reason step-by-step through complex terminology
- Ask clarifying questions to better understand what you're looking for
- Generate structured outputs — like a list of questions to bring to your doctor
I chose Claude Code over a standard AI chat interface because I wanted to paste in the full radiology report (which ran to nearly 800 words of dense medical language) and have it analyzed systematically, not just summarized.
Important note: I am not a medical professional. I used this as a supplementary tool for comprehension, not as a diagnostic replacement. More on that distinction shortly.
How I Set Up the Session
Here's the exact approach I used — you can replicate this if you're in a similar situation.
Step 1: Prepare Your Document
I copied the text of my MRI report directly from my hospital's patient portal. I did not upload any identifying information — I removed my name, date of birth, and patient ID before pasting it in. This is a critical privacy step.
Step 2: Frame the Request Clearly
Rather than just dumping the report and asking "what does this mean?", I gave Claude Code a structured prompt:
"I'm a patient, not a medical professional. I've received this MRI brain report and I have a follow-up appointment with my neurologist in three weeks. Please: (1) explain each finding in plain English, (2) flag anything that might warrant urgent attention, (3) identify terms I should research further, and (4) generate a list of specific questions I should ask my doctor."
This framing matters enormously. Vague inputs produce vague outputs.
Step 3: Iterate with Follow-Up Questions
After the initial analysis, I asked follow-up questions like:
- "What is an arachnoid cyst, and how common is it?"
- "The report says 'no restricted diffusion' — what does that mean in practical terms?"
- "Should I be concerned about the T2 hyperintensity finding, or is this often benign?"
Each response was detailed, well-sourced in its reasoning, and — crucially — appropriately caveated.
[INTERNAL_LINK: how to write better AI prompts for complex tasks]
What Claude Code Got Right
I'll be honest: I was impressed. Here's what the AI did well.
Plain-Language Translation
Every piece of jargon in my report was broken down clearly. "T2 hyperintensity" became "an area that appears brighter than surrounding tissue on a specific type of MRI scan, which can indicate a range of things from completely normal variation to inflammation." That's genuinely useful.
Contextualizing Incidental Findings
The arachnoid cyst finding had sent me into a spiral. Claude Code explained that arachnoid cysts are found in approximately 1-2% of the population, are usually congenital, and in the vast majority of cases require only routine monitoring — not intervention. It also noted that this should be confirmed with my neurologist given my specific symptoms, which was exactly the right caveat.
Generating Quality Questions
This was arguably the most valuable output. Claude Code produced a list of 11 specific, intelligent questions for my neurologist appointment, including:
- "Given the T2 hyperintensity finding, what differential diagnoses are you considering, and what would help narrow them down?"
- "Is the arachnoid cyst related to my current symptoms, or is it likely incidental?"
- "What follow-up imaging timeline would you recommend, and what changes would prompt earlier imaging?"
My neurologist actually commented that these were "unusually good questions." I didn't mention where they came from.
Appropriate Epistemic Humility
Claude Code was consistent and clear about the limits of its analysis. It repeatedly noted that it was providing educational context, not medical advice, and that findings needed to be interpreted in the context of my symptoms, history, and a physician's clinical judgment. This wasn't boilerplate — it was woven naturally into the responses.
Where AI Falls Short: The Honest Assessment
This section matters as much as anything else in this article.
It Cannot Interpret the Actual Images
Claude Code analyzed my radiology report — the text document produced by a radiologist who had already interpreted the images. It cannot look at MRI scans directly and produce a diagnosis. There are specialized medical AI tools attempting to do this (like Viz.ai for stroke detection), but they're designed for clinical settings, not patient self-service.
It Lacks Your Clinical Context
An AI doesn't know your symptoms, your family history, your medications, or the reason you had the MRI in the first place — unless you tell it. Even then, it's working with incomplete information compared to a physician who has examined you.
It Can Occasionally Overstate Certainty
In one instance, Claude Code described a statistic about arachnoid cysts that I later verified was slightly off (it cited ~1-2% prevalence; some studies suggest up to 2.6% depending on population). The difference is minor, but it underscores the importance of verifying specific claims.
It Is Not a Second Opinion in the Clinical Sense
This is the most important point. A genuine medical second opinion involves a qualified physician reviewing your case, your imaging, your history, and applying clinical judgment. What I used Claude Code to get was better described as informed comprehension — which is valuable, but categorically different.
Comparing AI Tools for Medical Document Comprehension
If you're considering using AI to help understand medical reports, here's how the major options stack up as of mid-2026:
| Tool | Strengths | Limitations | Best For |
|---|---|---|---|
| Claude Code | Long-context analysis, nuanced reasoning, clear caveats | No image analysis, no clinical context | Complex reports, question generation |
| ChatGPT-4o | Widely accessible, good general knowledge | Can be overconfident, shorter context | Quick terminology lookups |
| Google Gemini Advanced | Strong at citing sources, Google integration | Variable medical depth | Cross-referencing findings |
| Specialized medical AI (e.g., Consensus) | Research-backed, peer-reviewed sources | Less conversational, more technical | Finding clinical studies |
My recommendation: Claude Code for in-depth report analysis, paired with a tool like Consensus if you want to find actual research papers on specific findings.
Practical Tips If You Try This Yourself
If you're going to use AI to help understand a medical report, do it right:
- Remove all personal identifying information before pasting any document into an AI tool
- Be specific in your prompt — tell the AI your role (patient, not clinician) and exactly what you need
- Ask for a question list — this is the single highest-value output for your doctor's appointment
- Verify statistics — don't take numerical claims at face value; cross-check with sources like PubMed or Mayo Clinic
- Use it to prepare, not to conclude — let AI help you have a better conversation with your doctor, not replace that conversation
- Don't use it in emergencies — if you have symptoms suggesting stroke, cardiac events, or other acute conditions, call emergency services immediately
[INTERNAL_LINK: best AI tools for personal health management in 2026]
The Bigger Picture: AI and Patient Empowerment
There's a meaningful conversation happening in healthcare right now about patient literacy and access. The average radiology report uses terminology that takes years of medical training to fully interpret. Patients have a legal right to their records but often lack the tools to understand them.
AI tools like Claude Code are filling a genuine gap here — not by replacing physicians, but by helping patients arrive at appointments more informed, ask better questions, and advocate for themselves more effectively. Research from 2025 published in JAMA Network Open found that patients who came to appointments with prepared, specific questions reported higher satisfaction and better comprehension of their care plans.
That's the use case. Not diagnosis. Not treatment decisions. Informed participation in your own care.
Key Takeaways
- I used Claude Code to get a second opinion on my MRI — and while it wasn't a clinical second opinion, it was genuinely useful for comprehension and preparation
- AI excels at translating medical jargon, contextualizing common findings, and generating smart questions for your doctor
- AI cannot interpret actual imaging, lacks your clinical context, and should never replace a physician's judgment
- Remove all identifying information before using any AI tool with medical documents
- The most valuable output: a structured list of questions to bring to your appointment
- Use AI as a preparation tool, not a diagnostic tool
Final Thoughts and CTA
If you're sitting with a medical report you don't fully understand and a doctor's appointment weeks away, using AI for comprehension assistance is a reasonable, practical step — as long as you're clear-eyed about what it can and cannot do.
Try this today: Take your report, remove identifying information, open Claude Code, and use the prompt structure I outlined above. Focus on generating questions for your doctor. That single output alone is worth the 15 minutes it takes.
And if you found this article helpful, consider sharing it with someone who might be staring down a confusing medical report of their own. The gap between data access and data comprehension is real — and we can help each other navigate it.
[INTERNAL_LINK: how to talk to your doctor about AI-assisted research]
Frequently Asked Questions
Is it safe to share my MRI report with an AI tool?
Generally, yes — with precautions. Always remove personally identifying information (name, date of birth, patient ID, physician name) before pasting any medical document into an AI tool. Most major AI platforms, including Claude, do not use conversational inputs to train their models by default, but you should review the privacy policy of any tool you use. Never upload actual MRI image files to consumer AI tools.
Can AI actually diagnose conditions from a radiology report?
No. AI tools like Claude Code can explain terminology and provide educational context, but they cannot diagnose medical conditions. Diagnosis requires clinical judgment, a full patient history, physical examination findings, and often the actual imaging — not just the text report. Any AI that claims to diagnose you from a report alone should be treated with significant skepticism.
What's the difference between this and getting a real second opinion?
A genuine medical second opinion involves a qualified physician — often a specialist — independently reviewing your case, imaging, and history. This is categorically different from AI-assisted comprehension. If you have a serious diagnosis or are considering a significant treatment decision, pursue an actual clinical second opinion. Many academic medical centers offer remote second opinion services.
Which AI tool is best for understanding medical reports?
Based on my testing, Claude Code performs best for long, complex reports due to its strong reasoning and appropriate epistemic humility. ChatGPT-4o is a solid alternative for quicker queries. For finding peer-reviewed research on specific findings, Consensus is worth bookmarking. Always cross-reference important claims with authoritative sources like PubMed, Mayo Clinic, or the NIH.
Should I tell my doctor I used AI to prepare for my appointment?
Yes — and don't be embarrassed about it. Most physicians in 2026 are accustomed to patients arriving with AI-generated research. Being upfront ("I used an AI tool to help me understand my report and prepare these questions") allows your doctor to correct any misconceptions the AI may have introduced and demonstrates that you're engaged in your own care. In my experience, physicians appreciate prepared patients.
Top comments (0)