Using AI to Find Breast Cancer on Mammograms

According to a recent article in the journal Nature, artificial intelligence can outperform humans at finding breast cancer on mammograms.

Every year, health professionals in the United States perform about 33 million screening mammograms. According to the American Cancer Society, not only does the test miss about 20 percent of breast cancers, but false positives are also not uncommon.

As you may have guessed, doctors wanted to increase the accuracy of mammography. So in 1998, the FDA approved the computer-aided detection or CAD, a technology for spotting cancerous cells in mammograms

At the time, some hospital administrators reportedly charged extra for CAD. That meant radiologists had to use the technology whether they liked it or not.

However, later studies found that CAD did not improve health professionals’ accuracy at detecting breast cancer. Instead, it made them even worse.

Speaking to NY Times, the director of breast imaging at the Massachusetts General Hospital in Boston, Dr. Constance Lehman, said:

“We can learn from the mistakes with CAD and do it better… Using computers to enhance human performance is long overdue.”

It seems a team of Northwestern University researchers are close to doing just that.

In a recent paper, the researchers reported a new method of finding breast cancer on mammograms using artificial intelligence. The Google-funded research also involved two British medical centers — Cancer Research Imperial Centre and Royal Surrey County Hospital.

Here’s how the researchers are using the A.I. system to find cancer in breast tissue.

Training Computers to Find Breast Cancer on Mammograms

For the task, the researchers collected data from about 15,000 women in the U.S. and 76,000 in Britain.

The team trained their system to recognize cancer using the mammograms whose diagnoses were already known. Then, they tested the system’s performance using x-rays from about 25,000 other women in Britain and 3,000 in the U.S.

For this part, the Northwestern team used mammograms with known outcomes. That way, they could compare the system’s readings with the radiologists’ diagnosis.

“We took mammograms that already happened, showed them to radiologists and asked, ‘Cancer or no?’ and then showed them to A.I., and asked, ‘Cancer, or no?” said Dr. Mozziyar Etemadi, an author of the study from Northwestern University.

Findings from the study suggest that artificial intelligence is more accurate at detecting breast cancer on mammograms than radiologists.

On scans from the U.S., the system reduced the false positives by 5.7 percent and false negatives by 9.4 percent. Similarly, the A.I. beat mammograms perform in Britain too, reducing false positives by 1.2 percent and false negatives by 2.7 percent.

With that said, the system was far from perfect.

While the A.I. outperformed six humans in a test, it missed cancer that all six radiologists spotted — and vice versa. Despite the challenges, the results are still promising.

Dr. Etemadi explained:

“There’s no denying that in some cases our A.I. tool totally gets it wrong and they totally get it right. Purely from that perspective, it opens up an entirely new area of inquiry and study. Why is it that they missed it? Why is it that we missed it?”

The system for reading mammograms is one of Google’s current ventures into medicine, and it’s not yet available for widespread use.

Read More: Regulating AI and Machine Learning in Medical Technologies