Mammograms interpreted by the skilled eye of a radiologist are highly effective and accurate when it comes to detecting cancer, but one in five women may have her cancer missed.
A local researcher says artificial intelligence shows real promise when it comes to improving outcomes. Northwestern Medicine researcher Dr. Mozzi Etemadi partnered with Google software engineers to develop the artificial intelligence model.
Using programming code, they taught a computer how to interpret a mammogram. As it identified cancers or missed lesions, it learned from its successes and failures and updated its algorithm.
“The interesting thing here is unlike a human that has a certain number of hours in a day, a certain number of years in a career. The computer has essentially no limit the more mammograms we can show it, it will keep learning and keep learning,” Dr. Etemadi said. “That’s why we think ultimately its able to find more cancer and have less false positives than a human radiologist.”
The researchers tested the program on thousands of images in an international study. In some cases, the AI outperformed radiologists. But other times, humans proved to be the better detectives.
“In some situations six radiologists miss a cancer that our tool finds, and in some situations six radiologists find a cancer that our tool doesn’t, and really that just highlights the early nature of most AI research we’re really just scratching the surface with what is possible,” Dr Etemadi said.
It’s not a perfect science. If a radiologist misses something on mammography, a false negative could delay life-saving treatment. On the other hand, false positives lead to unnecessary biopsies. That’s why Dr. Etemadi hopes the AI tool will be used like spell-check when it comes to helping radiologists find cancers on images.
“People ask me all the time, ‘is this going to replace all the doctors’ jobs?’ The human element is here to stay. If anything this is going to give us more time to spend with patients,” Dr. Etemadi said.
Over the next two to three years, researchers hope to test the AI tool in a clinical study with the goal of rolling it out to all patients.