Early detection
AI predicts breast cancer risk in the next 5 years
© Orawan / Adobe Stock
A US research team has presented an AI application that is said to be suitable for clinical use for the first time. Experts are celebrating this as the beginning of a new era in breast cancer prevention.
The sooner the better. This old truism is absolutely true when it comes to breast cancer: the earlier tissue changes are discovered, the higher the chances of recovery and the lower the risk of dying from it.
But by far not every woman examines her breasts regularly; some women do not see a gynecologist for years and only about half of the women who are invited for mammography over the age of 50 make these appointments.
6 pictures
Artificial eyes see more than human ones
But even if all of this were to happen in a perfect world, wouldn’t it be even better if a potential tumor could be detected even earlier, i.e. when it cannot yet be felt or visible to the doctor’s eye on the X-ray?
Approximately one in five women with breast cancer has a tumor that is missed by current mammography procedures. So-called CAD systems (computer-aided diagnostics) specialists can support the interpretation of the images or serve as a second diagnosis – quite successfully: in a British pilot project about which the BBC reportsaround 12 percent more tumors were found with this procedure than without. And this with a constant rate of false positive results, which can never be completely ruled out.
Predictions are amazingly accurate
But artificial intelligence can do even more, namely calculate the future cancer risk based on mammography data. The scientists who present their findings in the journal Radiology imagine, fill your mouth quite full: “We can predict with amazing accuracy whether a woman will develop cancer in the next one to five yearsbased solely on local differences between their left and right breast tissue,” says study author Jon Donnelly from Duke University, Durham/North Carolina. The accuracy of the 3-year prediction is 92 percent, compared to a 5-year one -Prediction still at 66 percent.
The AI algorithm called “AlgoMirai” is based on mammograms, i.e. X-rays of the breast tissue. While the predecessor “Mirai” already delivered similarly good results, this older model had weaknesses. So it only calculated the statistical risk based on personal and family history. “Mirai is a black box. Nobody knew how it made its decisions,” admits Jon Donnelly. It was precisely here, in the incomprehensible thinking process of the AI, that the catch lay. Because as the Canadian radiologist Prof. Dr. Vivianne Freitas, Joint Department of Medical Imaging, Toronto in one Accompanying article to the study emphasizes:
“The interpretability of artificial intelligence is an ethical prerequisite for its applicability in radiology.”
AlgoMirai now seems to make this possible; it is based on tissue differences between the two breasts. The algorithm was tested on more than 210,000 images from a total of 81,824 patients of different ethnicities from the USA, Sweden and Taiwan.
But Even the most sophisticated technology can only be of benefit if women undergo regular examinations. The impressive success figures may help people open up to this possibility if they had previously hesitated.