Since breast cancer is the most common in women, its early detection is a global priority because it can considerably improve patient care. This screening traditionally relies on a double reading by two radiologists, which is a long and expensive process. The idea of ​​using an artificial intelligence capable of automatically interpreting exams in order to save time seems quite natural. But does it work in practice? Does AI really make the radiologist’s job easier? Doesn’t she risk seeing cancers where there are none and, conversely, missing out on certain diagnoses?

A Swedish study, published in the journal The Lancet Oncology, today presents the very promising results obtained thanks to an AI specially designed for breast cancer screening. “We could be witnessing a small revolution”, comments Professor Jean-Emmanuel Bibault, radiation oncologist, researcher specializing in artificial intelligence and author of the book 2041: the odyssey of medicine (Éditions des Equateurs).

The AI ​​in question, developed by ScreenPoint Medical, was used to diagnose more than 80,000 Swedish women aged 40 to 80 with cancer in a clinical trial that ran between April 2021 and July 2022. Based on the mammogram provided to it, the algorithm establishes a score between 1 and 10 to assess the risk of the presence of a tumour. He is not satisfied with a general opinion, but is also able to indicate the area which he thinks would be affected, a major asset in guiding the search for suspicious nodules. A radiologist then uses this information to make their diagnosis.

” READ ALSO – How artificial intelligence is revolutionizing medicine

The scientists thus compared the diagnoses made according to the usual procedure (a double reading by two radiologists) with those made by an AI-assisted radiologist. The results are more than encouraging: not only was the use of AI with rereading by a radiologist as effective as conventional double reading, but it reduced the time it takes to interpret mammograms by 44%.

Also, the algorithm did not lead to an increase in “false positives” (i.e. cases where a patient is diagnosed with cancer that does not exist), which is always a risk with an automated system. whose objective is above all not to miss a case. However, this overdiagnosis is a real problem because it generally leads to unnecessary additional examinations. The only downside to the study is that it does not fully validate the functioning of AI for women of different ethnic origins.

“It’s impressive,” comments Michelle Ariche-Cohen, a radiologist specializing in female imaging at the French Breast Institute in Paris. “To my knowledge, they are the first authors to demonstrate the interest of their algorithm on a prospective study”, adds Professor Bibault. In the context of a shortage of specialists, this time saving could make it possible to reduce waiting times for appointments for patients or to increase the frequency of screenings.

” READ ALSO – Doctors, lawyers, journalists… Will artificial intelligence replace us?

In France, the national breast cancer screening program is aimed at patients aged 50 to 74 and consists of performing a mammogram every two years. However, cancer can still be detected between these two control mammograms: this is interval cancer. “These cancers, which represent up to 20% of breast cancers according to studies, escape the usual screening and have a poor prognosis”, explains Professor Bibault.

Another question arises on this subject: could AI not simply be able to anticipate the occurrence of these cancers by detecting early signals? The results will not be available for several years, but then this would represent enormous potential for improving the chances of survival for thousands of women. “AI could fundamentally transform our approach to screening”, comment Michelle Ariche-Cohen. “We didn’t see this type of tool as a threat to our business, as long as it remained a helper and was effective. “AIs could be powerful allies in saving lives while reducing the workload,” also believes Professor Bibault. However, further studies will need to confirm these promising results.