AI boosts accuracy in detecting hard-to-spot breast cancers

A study shows that artificial intelligence (AI) can help doctors spot hard-to-detect types of breast cancer more accurately. This breakthrough could mean more patients get the right treatment sooner and avoid being misdiagnosed. A study shows that artificial intelligence (AI) can help doctors spot hard-to-detect types of breast cancer more accurately. This breakthrough could mean more patients get the right treatment sooner and avoid being misdiagnosed.

Being diagnosed with breast cancer can often change lives beyond the imaginable sorrow and loss. Modern cancer detection is capable of early diagnosis, which enables more effective treatment compared with cancers that are detected later. New tests show that AI integration helps to better detect small breast tumors and avoid false negatives.

Here’s how, speaking very broadly, breast cancer detection works:

To know what kind of treatment will work best for the patient, doctors need to know exactly what kind of breast cancer a person has. One thing they look at is a protein called HER2 that may show up on cancer cells. If there’s a lot of HER2, that’s called HER2-positive, and there are special medicines that work well for those patients.

But scientists now know that many breast cancers that were once thought to have no HER2 protein (so they were HER2-negative) in reality have just a little bit of it. These are now called HER2-low or HER2-ultralow. About 65% of breast cancers fall into these groups.

Now, new medicines can treat them if doctors are able to detect the tiny amounts of HER2 present.

Here is the problem: it's hard to see low or ultralow levels of HER2 using regular lab tests. About 1 in 3 ultralow cases get labeled wrong. That means some patients who could get a helpful treatment might miss out because the cancer looked HER2-negative.

That’s where AI comes in.

In a study funded by AstraZeneca and published by the American Society of Clinical Oncology, 105 doctors from 10 countries looked at cancer samples to see how much HER2 they had. The doctors did this with and without help from an AI tool that gave them feedback. The AI was only added during the last test.

The results speak for themselves:

• Doctors were more accurate when using AI.

• Their agreement with expert results went from 76% to almost 90%.

• The number of mistakes on HER2-ultralow cases dropped from about 30% without AI to 4% with AI.

• They correctly identified the HER2 level almost 89% of the time with AI, compared to just 67% without it.

This is important because getting the HER2 level right means more people can get the latest treatments that might help them live longer.

One of the study's experts said AI isn’t replacing doctors; it’s just helping them make better, faster decisions so that patients get personalized care.

In the future, the researchers want to start using this AI tool in real hospitals to see how much it helps with actual patients, especially how quickly they can start treatment.

Other studies

A 2024 study published in Breast Cancer Research developed deep learning models to predict HER2-low status directly from hematoxylin and eosin (H&E) stained slides. These slides contain pink and purple dyes that show the shape and size of the cell, its outline, the surrounding tissue, and other properties.

By analyzing over 1,400 images, the researchers demonstrated that AI could effectively distinguish between HER2-negative, HER2-low, and HER2-positive tumors, offering a cost-effective alternative to traditional cancer tests.