Artificial intelligence (AI) integrated into Sweden’s national breast cancer screening programme has been found to outperform standard mammography, according to the full results of a 2023 trial published in The Lancet.
The findings show that AI-supported breast cancer screening detected more women with clinically relevant cancers without increasing false positives. Researchers also observed that women who underwent AI-assisted screening were less likely to be diagnosed with aggressive or advanced breast cancer over the subsequent two years.
The study, led by researchers from Lund University along with institutes in Sweden, Norway, Denmark and the Netherlands, builds on interim results released in August 2023 from the Mammography Screening with Artificial Intelligence (MASAI) trial. Those interim findings had shown that AI use led to the detection of 20 per cent more cancers compared with standard screening, while reducing radiologists’ screen-reading workload by 44 per cent.
The complete results now indicate that AI-supported mammography reduced cancer diagnoses in the years following a screening appointment by 12 per cent, which researchers described as a key measure of screening programme effectiveness.
"Our study is the first randomised controlled trial investigating the use of AI in breast cancer screening and the largest to date looking at AI use in cancer screening in general," lead author Dr Kristina Lang, a breast radiologist and clinical researcher from Lund University, Sweden, said.
She added, "AI-supported screening improves the early detection of clinically relevant breast cancers which led to fewer aggressive or advanced cancers diagnosed in between screenings."
Between April 2021 and December 2022, more than 1,05,900 women were randomly assigned to either AI-supported mammography screening or standard double reading by radiologists without AI. The AI system used in the trial was trained, validated and tested using more than two lakh examinations from multiple institutions across over ten countries.
Over a two-year follow-up period, 1.55 interval cancers per 1,000 women (82 out of 53,043) were detected in the AI-supported group, compared with 1.76 interval cancers per 1,000 women (93 out of 52,872) in the standard screening group — reflecting a 12 per cent reduction in interval cancer diagnoses in the AI arm.
An interval cancer refers to a malignancy detected between scheduled screening tests, after a prior negative result and before the next routine examination.
The study also found that 81 per cent of cancer cases (338 of 420) in the AI-supported mammography group were detected during screening, compared with 74 per cent (262 of 355) in the standard reading group — a nine per cent increase. False-positive rates were similar in both groups, at 1.5 per cent in the AI-assisted screening arm and 1.4 per cent in the standard reading arm.
"Widely rolling out AI-supported mammography in breast cancer screening programmes could help reduce workload pressures amongst radiologists, as well as helping to detect more cancers at an early stage, including those with aggressive subtypes," Lang said.
Commenting on the findings, Kalyan Sivasailam, Founder and CEO of Bengaluru-based radiology AI firm 5C Network, who was not involved in the study, told PTI, "The MASAI study provides strong evidence that AI can improve mammography screening in a well-organised, well-resourced Swedish program with standardised equipment and experienced radiologists."
However, he cautioned that replicating these outcomes in India would be challenging due to major structural differences.
"India's mammography landscape is fundamentally different from Sweden's. India has fragmented equipment, and there is no organised population-based screening at a national scale. Image quality varies dramatically between a tertiary centre in Mumbai and a district hospital in rural Bihar. An AI system would need validation across the equipment mix actually present in Indian facilities," he said.
Sivasailam also highlighted India’s shortage of radiologists, particularly in district hospitals, and the absence of a robust cancer registry system.
"Interval cancers (in the study) were identified through linkage with the Regional Cancer Registry using participants’ national personal identification number. India lacks this infrastructure. Without robust follow-up systems, we wouldn't even know if an AI system was missing cancers that later presented as interval cancers. We'd only see the detection rate, not the full picture," Sivasailam said.
He added that AI deployment strategies in India might need to differ from those used in the MASAI trial.
"Rather than 'AI supporting double reading' (as in the MASAI study), it might be 'AI as primary reader with radiologist confirmation of positive cases'. This is a more aggressive deployment than what MASAI studied. Before deployment, we would need prospective studies of AI as standalone reader in resource-limited settings, not just AI as decision support for experienced radiologists," Sivasailam said.
The evidence base for AI-powered mammography screening in Indian healthcare settings does not yet exist, he added.
