Why Using AI to Interpret Mammogram Results Can Cause Unnecessary Panic

You know it, I know it: our patients are using AI to interpret medical results.
Last week my friend Annie called me in a panic late at night.

She had had a screening mammogram earlier in the day, the first one in 3 years. She already felt guilty about the delay. Life had gotten busy. There were kids, work, appointments, obligations. The mammogram had slipped down the list of things she knew she needed to do.

Then, less than 24 hours after her appointment, her patient portal notified her that she needed additional imaging. She panicked. And before she could talk to her doctor, a nurse, a radiologist, or anyone else who could put the report into context, she did what millions of patients are doing every day. She asked AI.

Annie, my smart and capable small business owner friend who once worked as a CPA and handled huge accounts for a national company, copied and pasted her mammogram report into an AI tool and asked the algorithm to explain what it meant. The AI tool gave her an answer -- and it scared the hell out of her. Based on the language in the report and the explanation of the BI-RADS categories, the AI led Annie to believe she probably had cancer and might need surgery.

Except that wasn't what her report said - her report was BI-RADS 0.

For someone who works in breast cancer care, BI-RADS 0 is familiar territory. It simply means the imaging is incomplete, and the radiologist needs additional information before making a final assessment. It is not a cancer diagnosis. It is not a recommendation for surgery. It is not a crystal ball. In my friend's case, the radiologist wanted additional views of an area in the right breast that wasn't adequately characterized on the initial screening images.

That's it. She needed more pictures, that’s all, no need to panic, but Annie didn’t know that.

As a breast cancer nurse navigator who has spent more than 20 years helping patients navigate breast imaging, biopsies, diagnoses and treatment, I knew exactly what I was looking at.

I also knew something the AI didn't, a simple reality - we don’t know have enough information to make a diagnosis yet.

That distinction matters enormously in healthcare. Additional imaging after a screening mammogram is common. Often, the additional views clarify what was seen on the original images and the patient goes right back to routine screening. Sometimes additional imaging leads to a biopsy. Sometimes a biopsy leads to a cancer diagnosis.

But you don't get to skip those steps simply because an algorithm has generated a frightening possibility.

And this is where I have a problem with the way we're talking about AI in healthcare. We're spending an enormous amount of energy asking what AI can do. We need to spend a lot more time asking what patients think it can do. Because AI doesn't know what it doesn't know.

AI hasn't looked at my friend's mammogram images. It doesn't know what the radiologist saw. It doesn't know whether the finding was new or stable compared with her prior studies. It doesn't know her complete clinical history - her family history or that she breastfed for four years. It doesn't know the nuances behind the radiologist's recommendation.

And it certainly doesn't know my friend. It doesn’t know that she is anxious, that she recently lost her mom and a close friend to cancer, or that this response she got from her AI tool sent her into a tailspin.

AI knows a lot but not enough. Yet we have created a healthcare environment perfectly designed to make patients turn to it: We release pathology reports before anyone calls. We release radiology reports after hours. We send lab results directly to patients without explanation. Then we tell them to wait three business days for someone to return their message.

So of course they're Googling. Of course they're asking ChatGPT. Of course they're looking for someone - or something - to tell them what is happening inside their own body.

We created an information vacuum and now AI is filling it.

That doesn't mean we should tell patients to stop using AI. It means we need to get much more serious about teaching patients how to use it. I tell my patients to use it to generate questions and translate medical jargon. I have used it myself to prepare an outline for an appointment with a specialist.

But, please, please for goodness sake, please don't let an AI-generated interpretation become the voice of your medical record! Because a medical report is not a conversation. A recommendation is not a diagnosis. A possibility is not a probability, and an algorithm is not a clinician.

My friend Annie didn't need an AI tool to tell her the worst thing that could possibly be true. She needed someone to tell her what the report actually meant, and what it didn't mean. She needed a reassuring voice on the other end of the phone to say: “Wait a minute, let me explain and let you take a breath. You don't have a cancer diagnosis, you have an incomplete work up, the radiologist just needs another picture. Let's take this one step at a time.”

That is navigation. That is clinical judgment. That is human connection. And in an increasingly AI-driven healthcare system, I would argue that those things aren't becoming less important - they're becoming more important than ever.

Emily M. Beard, BSN, RN, OCN, CBCN, BHCN, is an oncology survivorship navigator at the Winship Cancer Institute at Emory University.