A randomized study of more than 80,000 women in Sweden shows artificial intelligence (AI)-assisted breast screening is safe and effective and can significantly reduce radiologist workload.
The results are reported in The Lancet Oncology and are the first, interim, read out of a study that is aiming to follow 100,000 women over two years to assess if using AI to help interpret mammography images can reduce the number of cancers detected between screenings.
This study was carried out to assess safety and found slightly better results were obtained with AI-assisted screening compared with double reading without AI and the screen reading workload for radiologists was reduced by almost 50%.
“These promising interim safety results should be used to inform new trials and programme-based evaluations to address the pronounced radiologist shortage in many countries. But they are not enough on their own to confirm that AI is ready to be implemented in mammography screening,” said lead author Kristina Lång, from Lund University, Sweden, in a press statement.
“We still need to understand the implications on patients’ outcomes, especially whether combining radiologists’ expertise with AI can help detect interval cancers that are often missed by traditional screening, as well as the cost-effectiveness of the technology.”
The study included 80,022 women in Sweden between the age of 40 and 80 years who were eligible for either standard screening (mammography every 1.5–2 years) or at-risk screening (mammography once a year for those at higher risk).
Half the participants were randomly assigned to AI-supported screening and the other half to “gold standard” double radiologist reading without AI assistance. The AI system analyzed each image and predicted cancer risk on a 10-point scale, with 10 being the highest risk score. For cases with a risk of less than 10, one radiologist checked the image and if the score was 10 then two radiologists checked the image.
“European guidelines recommend double reading of screening mammograms to ensure high sensitivity. A meta-analysis suggested that double reading resulted in 0.44 more cancers being detected per 1000 people screened than with single reading; however, this comes at the expense of a large screen-reading workload and can potentially increase false positives,” write the authors. “Double reading can also be difficult to sustain because of a shortage of breast radiologists in many countries.”
The mammography results showed that the AI-assisted group had 244 detected cancer cases, 861 recalls, and 46,345 screen readings in total. The control group, without AI-support, had 203 detected cancer cases, 817 recalls and 83,231 screen readings. The cancer detection rates were 6.1 per 1000 screened participants in the AI group, meeting the safety requirement of more than three cancers per 1,000 screened women, and 5.1 per 1000 in the control group.
Of the women who were recalled, 244 (28%) from the AI group were found to have cancer compared with 203 (25%) in the control group. This resulted in 41 more cancers being detected with the support of AI. The false-positive rate was the same in both groups at 1.5%.
The researchers will continue the study to try and assess if AI-assistance can help reduce the number of cancer cases that are missed on screening and only picked up between screening sessions, so-called ‘interval cancers’ usually leading to a worse prognosis.
“The greatest potential of AI right now is that it could allow radiologists to be less burdened by the excessive amount of reading,” said Lång. “While our AI-supported screening system requires at least one radiologist in charge of detection, it could potentially do away with the need for double reading of the majority of mammograms easing the pressure on workloads and enabling radiologists to focus on more advanced diagnostics while shortening waiting times for patients.”
