AI breast cancer screening is safe and detects 20% more cancers

Research conducted at Lund University indicates that AI breast cancer screening is not only a viable alternative to conventional screening but also a more efficient one, detecting more cases of cancer. The findings from the Mammography Screening with Artificial Intelligence (MASAI) trial showcased that AI-enabled screening diagnosed 20% more instances of cancer while reducing radiologists' burden by over 40%. This technological innovation could drastically reduce wait times for millions of global breast cancer patients and enhance treatment outcomes.

Conventional diagnostic procedures are time-consuming Approximately one million women undergo mammography screening annually in Sweden, with each examination requiring review by two breast radiologists to maintain a high sensitivity 'double reading.' However, the scarcity of breast radiologists and the increasing backlog can pose a risk to patients. Therefore, specialists are considering the potential of AI breast cancer screening.

The MASAI trial is a pioneering randomized controlled study assessing the impact of AI-supported screening.

The MASAI trial's methodology The MASAI trial assigned 80,033 women randomly into two groups: 40,003 women participated in AI breast cancer screening (intervention group), and 40,030 women underwent standard double reading without AI assistance (control group).

Kristina Lång, the study leader and an associate professor in diagnostic radiology at Lund University, detailed the trial's promising results.

Lång stated, “The application of AI led to the identification of 20% (41) more cancers compared to traditional screening, without impacting false positives. A false positive during screening happens when a woman is recalled but cleared of cancer suspicion after workup.”

Screen readings with AI-supported screening stood at 46,345, compared to 83,231 with traditional screening, reducing the screen-reading workload of radiologists by 44%.

This substantial improvement in screening efficiency is critical as radiologists typically read 50 screening examinations per hour. Predictions suggest that AI breast cancer screening could trim down screen reading times by five months.

Future plans for the breast cancer project The team intends to examine the types of cancer detected with and without AI, with the trial's main outcome being the interval cancer rate.

Interval cancer rate refers to cancer diagnosed between screenings and often has a worse prognosis than screen-detected cancers.

The trial's women will have their interval cancer rate evaluated once they have a minimum of a two-year follow-up.

Lång concluded: “Screening is intricate. The balance between benefit and harm must always be weighed. The fact that a screening method detects more cancers doesn't necessarily mean it's superior.

“The key is to find a method that can detect clinically significant cancers at an early stage. However, this needs to be counterbalanced with the harm caused by false positives and the overdiagnosis of slow-growing cancers.

“The initial analysis results indicate that AI-supported screening is safe as the cancer detection rate didn't drop despite a significant decrease in the screen-reading workload.

“The planned analysis of interval cancers will determine whether AI-supported screening also contributes to a more precise and effective screening programme.”