A new AI image tool that selects high-risk women for magnetic resonance imaging (MRI) after negative mammography detects many missed breast cancers, interim findings from a clinical trial suggest.
The AI-based system could help offset the extra costs of MRI that have prevented it from being included in national screening programs, despite its effectiveness in spotting malignancies.
It was nearly four times more efficient as a selection tool than traditional breast density measures evaluated in an earlier study in terms of the proportion of MRI examinations leading to a cancer diagnosis.
“Using the AISmartDensity method would make the detection cost per cancer similar to the cost in population-wide screening mammography and contribute to earlier detection of invasive cancer,” maintained Fredrik Strand, MD, PhD, from Karolinska University Hospital in Stockholm, and colleagues in the journal Nature Medicine.
Although mammography is standard for breast-cancer screening in populations, just under a third of screened women have so-called interval cancers, which become symptomatic after a negative screening and before the next scheduled one.
These cancers can be fast-growing and not present or undetectable on mammograms at the time of screening, or they may have been missed by the radiologist reader.
Contrast-enhanced MRI is more sensitive at detecting early breast cancer than mammography and does not have the latter’s reduced sensitivity for women with extremely dense breasts.
But the cost of MRI has meant it is not included in any national screening program, despite women with dense breast tissue have up to double the risk of breast cancer of others.
The ScreenTrustMRI trial assessed the value of the AISmartDensity tool to select individuals for supplemental MRI using three component models that assess underlying risk, potential masking, and suspicious cancer signs.
Follow up is planned until August 2025, and the primary endpoint is advanced cancer at 27-months after initial screening in women randomly assigned to MRI or no MRI.
Secondary endpoints include cancer detection at supplemental MRI, participant engagement, AI score distribution, tumor characteristics, radiological process measures and questionnaire responses.
The study included 59,354 women whose mammograms were screened using the AI tool, of whom 6.9% were eligible to participate due to a having a “very high” AISmartDensity score.
After removing those who did not participate due to an invitation error or lack of consent, 652 were not randomly assigned to MRI. A further 663 were randomly assigned to MRI, of whom 559 completed this imaging.
There were 64.4 cancers detected per 1000 MRI examinations, with a positive predictive value of 38% for individuals recalled after MRI and 50.7% for individuals who were biopsied.
The authors note that this cancer detection rate is nearly four times more than with traditional breast density measures used in a previous clinical trial called DENSE, which identified 16.5 per 1,000 MRI examinations. Most additional cancers detected were invasive and several were multifocal, suggesting that their detection was timely.
“The detected 36 cancers in the selected individuals correspond to 63% of the 57 expected future cancers in the entire screening population,” noted the researchers.
“The potential to pre-emptively detect most cancers by offering MRI to a small proportion of individuals represents an important healthcare value proposition.”
