Until now, the prevailing opinion was that AI could detect early signs of breast cancer in mammograms at least as well and reliably as a trained radiologist. Now, however, a recent study of Hungarian physicians, conducted by AI software company Kheiron Medical Technologies, has shown that in multiple cases physicians only identified cancerous areas after they had first been flagged by AI. The detection technology, which is partly based on probability calculations, was able to mark potential cancerous cells on the scans even before the cancer became visible or when it only appeared several years later.
Hardly any other country has as much experience using AI to screen patients as Hungary. Accompanied by scientific studies, more than 35,000 screenings a year have been performed at five clinics over the past two years. The aim is to use a step-by-step approach to optimize the detection technology.
Test programs are running in the US and the EU. The US National Cancer Institute estimates that around 20 percent of breast cancer cases are overlooked in mammograms. In Sweden, Laszlo Tabár, a professor emeritus at Uppsala University, has been studying the radiology of mammograms for a number of years. “I was really surprised how good the AI system functions.” AI could become a lifesaver for patients, he adds, and become invaluable for public health systems. “I dream of the day when women come to a breast cancer center and their first question will be: Do you work with AI or not?”
“I dream of the day when women come to a breast cancer center and their first question will be: Do you work with AI or not?”
However, before AI can take over the early detection of cancers on its own, the systems must become safer and more reliable. For instance, the technology must show that it can deliver accurate results for women of every age, ethnicity and body type. Beyond this, it must also be demonstrated that false positives—areas that appear cancerous but in fact are not—can be ruled out.
The fact remains that AI will not replace human radiologists over the long term, but rather support them in their work. This combination, of AI with trained and experienced physicians, is already delivering excellent results.
