Mammography is a simple medical test capable of detecting very small lesions suspicious of breast cancer, which makes it possible to diagnose the disease in its early stages and thus facilitate treatment and improve the prognosis. Now, a trial led by researchers at Lund University in Sweden has shown that incorporate artificial intelligence (AI) into the reading of mammograms significantly improves the detection of breast cancer cases.
Breast radiologists currently perform a double mammography reading to guarantee a high sensitivity of the test and the use of AI constitutes a safe alternative to this work and would help to reduce the workload of these specialists, according to the results of the study that have been published in The Lancet Oncology.
To know exactly what happens when radiologists work with AI support, studies are needed in which women are randomized to either AI-supported screening or standard screening. The Mammography Screening With Artificial Intelligence (MASAI) trial is the first randomized controlled trial to evaluates the effect of detection supported by artificial intelligence.
“We found that the use of AI resulted in the detection of 20% more cancers compared to standard detection, without affecting false positives”
“In our trial, we used AI to identify screening exams with a high risk of breast cancer, which were double read by radiologists. The rest of the scans were classified as low risk and were read only by a radiologist. In the screen reading, the radiologists used the AI to support detection, in which it highlighted suspicious findings in the images”, he explained. Kristina Långresearcher and associate professor of diagnostic radiology at Lund University and consultant at Skåne University Hospital, who led the study.
An advance in the early diagnosis of breast cancer
The research involved 80,033 women who were randomly assigned to two groups: 40,003 women in the intervention group who underwent AI-assisted screening and 40,030 in the control group who underwent standard double reading without support. of AI. “We found that the use of AI resulted in the detection of 20% (41) more cancers compared to standard detection, without affecting false positives. There is a false positive in the screening when a woman is called back, but the suspicion of cancer is cleared up after the study”, has commented Kristina Lång.
The additional advantage of this screening is that the Screen reading workload for radiologists decreased by 44%. The number of screen reads with AI-assisted detection was 46,345 compared to 83,231 with standard detection. This reduction in time is important, as Kristina Lång explains, noting that on average a radiologist reads 50 screening exams per hour and using AI they estimate the time it takes a radiologist to read the approximately 40,000 screening exams. detection in the AI group was reduced by about five months.
The researcher has highlighted that the results are promising, but that they must find out if they are maintained when changing conditions “for example, with other radiologists or other AI algorithms.” In addition, they are provisional results and definitive conclusions will not be available until a few years from now because they need to verify if the use of AI to interpret the images translates into a reduction in interval cancers, which are tumors that are detected between check-ups and usually have a worse prognosis than those seen in check-ups. The researchers plan to evaluate it in a study of 100,000 women, with at least two years of follow-up.
“Detection is complex. The balance between benefit and harm should always be taken into account. Just because a screening method finds more cancers does not necessarily mean that it is a better method. The important is find a method that can identify clinically significant cancers at an early stage. However, this must be balanced against the harm of false positives and overdiagnosis of indolent cancers. The results of our first analysis show that AI-assisted detection is safe, as the cancer detection rate did not decrease despite a substantial reduction in screen reading workload. The planned analysis of interval cancers will show whether AI-assisted detection also leads to a more accurate and effective screening program”, concludes Kristina Lång.
Stephen Duffy, Professor of Cancer Screening at the Wolfson Institute for Population Health, Queen Mary University of London (QMUL) finds the results very interesting, although he warns of the danger of overdiagnosing harmless tumors: “There may be concerns that these increases in detection driven by by technology include excessive detection of relatively harmless lesions. For example, the results of this work include increased detection of ductal carcinoma in situ, which is considered potentially overdiagnosed. The authors plan to address this question by estimating the effect of the incorporation of artificial intelligence on the rate of symptomatic cancers that arise in future years in people who were negative in the screening”, he told SMC Spain.