Using AI to interpret medical tests and diagnostics

AI is transforming how patients interact with their own medical results. Joanna Stern, a technology journalist with over a decade at the Wall Street Journal, spent an entire year integrating AI tools into every aspect of her life, including health. In a conversation with Dr. Eric Topol, a physician and medical technology researcher, Stern shares concrete experiences with blood tests, mammograms, breast ultrasounds, and dental care, while Topol provides the clinical context needed to understand what is actually changing.

Blood tests: from a 30-second voicemail to an AI-generated podcast

It started when Stern received her lab results with a concerning flag: elevated LDL cholesterol. Her doctor's communication was a 30-second voicemail from a nurse: "Avoid fatty foods. Exercise more." No further details.

Rather than accepting that response, Stern uploaded the PDF of her results to Google NotebookLM, a free tool that lets you load documents and receive summaries, quizzes, or even an automatically generated podcast with two AI voices. The system explained what elevated LDL meant, though the final advice replicated almost exactly what the nurse had already said.

The experience illustrates a point Dr. Topol emphasizes repeatedly: many patients already use chatbots to interpret their lab work because doctors rarely have time to review results in depth. Using AI to understand results before a medical appointment is an increasingly common and valid practice, as long as it does not replace clinical supervision.

AI in mammography and breast ultrasound: a new standard taking hold

Stern has a family history of breast cancer and very dense breast tissue, which makes imaging harder to read. In the most relevant chapter for female preventive health in her book, she describes her experience with two AI tools during her annual mammogram and breast ultrasound:

• ScreenPoint (Transpera algorithm): validated in a Swedish study involving over 100,000 women. It detected nearly 30% more breast cancers than conventional review without AI.
• KIOS: an AI tool for breast ultrasound. It flagged three suspicious areas, which her radiologist then carefully compared against studies from six months earlier.

Her radiologist, Dr. Lori Margoles at Mount Sinai, did not defer her judgment to the AI. She used it as a visual second opinion, ruling out two of the three flagged areas based on the absence of change from the prior ultrasound. The third, which appeared new, led to an additional breast MRI.

Dr. Topol notes that both tools have FDA clearance and that some radiology networks around the country already offer the service, though in certain centers it carries an out-of-pocket cost of up to $40. At Mount Sinai, it is free.

AI in dentistry: get a second opinion before accepting costly treatment

One of the most practical chapters in the book deals with dentistry. Stern investigated how AI tools like Pearl, Overjet, and Rydia are being integrated into dental diagnostic systems, and she warns about the risk of overdiagnosis. Dr. Topol himself admitted falling victim to this before reading about AI in dentistry: he underwent a deep cleaning across two quadrants that turned out to be unnecessary in retrospect.

The takeaway is clear: if a dentist recommends expensive or extensive treatment, seeking a second opinion or choosing a practice that uses AI-assisted diagnostic tools can prevent unnecessary procedures and unjustified spending.

ChatGPT for interpreting complex medical reports

When Stern received her breast MRI results indicating she needed two biopsies, the anxiety was immediate. She uploaded the report to ChatGPT, after removing any identifying information, and received a clear explanation of a technical document that even her own gynecologist had not fully grasped.

This is perhaps the most direct application of large language models in healthcare: turning complex medical text into understandable information for the patient, without replacing the clinical consultation. Stern gave this the highest score in her AI health report card: 10 out of 10.

Three rules for evaluating any AI health tool

Dr. Topol proposes three criteria for assessing any AI tool applied to health:

1. Test it rigorously: do not accept results without verifying them.
2. Benchmark against the human standard: is it better, equivalent, or worse than the specialist?
3. Evaluate the cost: financial, computational, and energetic.

Conclusion

AI is not going to replace the doctor, but it is already changing the patient experience. Uploading a lab report to NotebookLM, asking ChatGPT to explain a complex medical report, or choosing a mammography center that uses validated algorithms are concrete and accessible decisions. The greatest risk is not using these tools, but doing so without critical thinking or in place of consulting a healthcare professional.