Jay's Brief

The Saturday Explainer ·

Some cancer AI works like a highlighter, and the doctor makes the call

A doctor in a white coat holding a clipboard studies a chest scan on a glowing wall lightbox, while a floating yellow highlighter draws a dashed circle around one small spot on the lung.

Saturdays are different: no news, just one big AI idea, explained simply.

The short version

Some AI tools used in cancer care work like a highlighter. They look at a scan or a thin slice of tissue on a glass slide and mark spots that may need a closer look. Then the doctor decides. For example, the US Food and Drug Administration has authorized the marketing of software that helps pathologists find areas of prostate biopsy images that may contain cancer. Pathologists are the doctors who study tissue samples, and a biopsy is a small piece of tissue taken to test for disease.

How it actually works

Start with something you may have had: a mammogram, the X-ray picture of the breast. A radiologist, a doctor trained to read medical images, looks it over for anything odd. AI can help process these images quickly, a bit like a helper with a highlighter. Through machine learning, a way of teaching by example, these tools learned from lots of past images what trouble tends to look like. The National Cancer Institute, the US government's main cancer research agency, says this lets radiologists spend more time on work that needs their judgment. So who makes the call?

The doctor does. The highlighter only points, and it can point at the wrong thing or skip something that matters. Here's why that teamwork matters. One study looked at whether breast cancer had spread to the lymph nodes, small glands that can carry cancer around the body. Doctors helped by a computer program reading tissue slides were more accurate than the doctors alone. They were also more accurate than the program alone.

AI isn't only for pictures, either. Doctors' notes are messy and written in everyday language. The NCI says researchers built tools that pull tumor details out of those notes automatically, saving thousands of hours of work people used to do by hand. That is meant to help the NCI's cancer statistics program, so researchers can better understand how new treatments affect patients.

So why not just trust big claims? Because other researchers said one of them couldn't be checked. In 2020, Google DeepMind, an AI lab, announced a program it said beat human experts at spotting breast cancer in screening scans. Other researchers replied that the study left out key details, which they said made it impossible for others to confirm. The steadier progress looks a lot more like the highlighter.

The word you'll hear

Machine learning

A way of building AI by showing it lots of examples, like thousands of past scans, so it learns the patterns itself instead of following rules a person wrote.

Why it matters to you

If you get a mammogram or a biopsy in the next few years, software might help read it. You likely won't see it happen. The doctor still signs off, but a program may have marked the spots they looked at first.

Be careful about

These tools only know what they were shown. The NCI warns that if the data used to train these tools don't reflect everyone, the tools can repeat medical bias, meaning they may work less well for some groups of patients. Researchers studying skin cancer tools, for example, have noted fewer pictures of darker skin to learn from, which may make them less reliable for some people. The NCI also says more clinical trials, careful tests with real patients, are needed to show these tools help in everyday care.

See it for yourself

There's nothing to try at home here. The small thing: at your next screening, it's fair to ask whether software helped read your images and who made the final call.

Go deeper

AI and Cancer
cancer.gov

Also checked against: en.wikipedia.org