Applying an AISmartDensity, a recently developed artificial intelligence (AI) tool to score each mammogram to identify women at risk of undetected breast cancer following negative mammography screening, as a selection method for supplemental magnetic resonance imaging (MRI), in a randomised ScreenTrustMRI study is around four times more efficient in terms of cancer detection than second-best approaches based on traditional density measures and risk models.

Most additional cancers detected were invasive and several were multifocal, suggesting that their detection was timely according to Dr. Fredrik Strand of the Department of Oncology-Pathology and Breast Radiology Unit, Karolinska University Hospital in Stockholm, Sweden and colleagues, who published the findings on 8 July 2024 in the Nature Medicine.

Early detection achieved through screening mammography leads to decreased breast cancer mortality. However, around 30% of cancers in screened women are diagnosed as interval cancers. Interval cancers may escape screening as they may be fast-growing and not present on mammograms at the time of screening, were missed by the radiologist reader or were undetectable by mammography at the time. Interval cancers show aggressive biology and have relatively poor prognosis.

The sensitivity of mammography is reduced for women with extremely dense breasts, while the sensitivity of MRI is unaffected. Previous studies have shown that reductions in the number of interval cancers can be achieved by using MRI for supplemental screening of women with high mammographic density. However, as qualified MRI staff are lacking, the equipment is expensive to purchase and cost-effectiveness for screening may not be convincing, the utilisation of MRI is currently limited. An effective method for triaging individuals to supplemental MRI screening is therefore needed.

AISmartDensity has a modular structure with three component models assessing underlying risk, potential masking and suspicious cancer signs. The primary endpoint of the ScreenTrustMRI study is to examine the incidence of advanced cancer at 27-month follow-up after the initial screening in the group of individuals randomised to MRI compared with those randomised to no MRI. Secondary endpoints include cancer detection at supplemental MRI, participant engagement, AI score distribution, tumour characteristics, radiological process measures and questionnaire responses.

Advanced cancer was defined as any of the following: interval cancer, invasive component larger than 15 mm and lymph node positive cancer, or screen-detected cancers meeting specific criteria such as interval cancers, cancers having an invasive component larger than 15 mm or lymph node metastasis.

The study started inclusions on 1 April 2021 and ended inclusions on 7 April 2023, and the follow-up period will end in August of 2025 for the last included patient. The report in the Nature Medicine focuses on the detection of additional cancers for individuals randomised to, and undergoing, supplemental MRI. It was prespecified in the study protocol that secondary endpoints could be reported before primary endpoints. The study hypothesis was that AI-based image analysis will provide a more effective selection tool than traditional density in terms of the proportion of MRI examinations leading to a cancer diagnosis. 

The investigators offered study participation to individuals with a negative screening mammogram and a high AI score (top 6.9%). Upon agreeing to participate, individuals were assigned randomly to one of two groups: those receiving supplemental MRI and those not receiving MRI. Compared with traditional breast density measures used in a previous clinical trial, the current AI method was nearly four times more efficient in terms of cancers detected per 1,000 MRI examinations (64 versus 16.5).

Positive predictive value was 38% for individuals recalled after MRI and 50.7% for individuals who were biopsied. Most of these malignant lesions had invasive components. The invasive cancers had a median size of 13 mm on pathology analysis, which is smaller than the average size of 15.8 mm and 19.6 mm for mammography screen-detected cancer and interval cancer previously reported for a similar breast centre in the Stockholm area. It is notable that most of the cancers detected in the population selected for supplementary MRI exhibited invasive features. This could be attributed to a combination of the AI tool and the use of MRI.

The authors commented that even if AISmartDensity is likely to catch around four times as many cancers as using BI-RADS D, there could be some cancers identified by BI-RADS D that AISmartDensity very high might miss.

As an alternative approach with potential to reduce MRI utilisation and further increase cost-effectiveness, it may be suggested that radiologists should re-review the mammograms of all women with a very high AISmartDensity score. An assessment of this approach is planned as a post hoc reader study.

Furthermore, subsequent follow-up research will determine primary outcomes of the effect on prognostically important cancer characteristics.

Funding was provided by Region Stockholm, Medtechlabs, the Swedish Breast Cancer Association, the Swedish Cancer Society grant, and the Swedish Research Council grants. Open access funding was provided by Karolinska Institute.

Reference

Salim M, Liu Y, Sorkhei M, et al. AI-based selection of individuals for supplemental MRI in population-based breast cancer screening: the randomized ScreenTrustMRI trial. Nature Medicine; Published online 8 July 2024. DOI: https://doi.org/10.1038/s41591-024-03093-5

Source