Novel Image Analysis IDs Women for MRI After Negative Mammography

An artificial intelligence tool to select individuals for supplemental MRI scans after negative mammography led to detection of many cancers that would otherwise have been missed, according to a randomized trial.

Use of the tool, called AISmartDensity, to identify women at risk of undetected cancer following negative mammography screening resulted in a four-fold higher supplemental cancer detection rate compared with traditional breast density measures, reported Frederik Strand, MD, PhD, of Karolinska University Hospital in Stockholm, Sweden, and colleagues.

Moreover, “most additional cancers detected were invasive and several were multifocal, suggesting that their detection was timely,” wrote Strand and colleagues in Nature Medicine.

In explaining the rationale behind the ScreenTrustMRI trial, the authors pointed out that although the early detection of cancer with screening mammography has decreased breast cancer mortality, many breast tumors in screened women are found as interval cancers: those detected before the next planned screening, usually because they have become symptomatic.

“[S]ensitivity of mammography is reduced for women with extremely dense breasts, while the sensitivity of MRI” — which is better than with mammography — “is unaffected,” Strand and colleagues wrote. “The increased detection, however, comes at a substantially increased per examination cost, which results in lower cost-effectiveness compared with mammography and hinders widespread implementation,” the group explained.

They hypothesized that an AI-based imaging analysis could be more efficient, in terms of the number of MRI examinations that lead to a cancer diagnosis.

The primary endpoint of the ScreenTrustMRI trial, slated to complete in August 2025, is to evaluate the incidence of advanced cancer at 27-month follow-up after the initial screening of individuals randomized to MRI compared with those randomized to no MRI. The current report reflects an interim analysis focused on additional cancers detected in those women randomized to supplementary MRI.

ScreenTrustMRI enrolled 59,354 women whose mammograms were screened with AISmartDensity. Of these, 3,821 were eligible for randomization due to a “very high” AISmartDensity score (the top 6.9%), and 1,315 were eventually randomized, with half (663) undergoing MRI.

Among those assigned to MRI, 104 women (16%) chose not to undergo the examination, while 559 (84%) completed MRI and formed the final cohort for the current analysis.

Median age was 56; 22 women (4%) had a previous history of breast cancer, and 104 (19%) reported having a family member with a history of breast cancer.

Lesions found on MRI were classified according to BI-RADS in order to assess the degree of malignancy suspicion of imaging findings. Of the 559 examinations, 95 were graded at BI-RADS 3-5 and flagged for further work-up. Biopsies were obtained for 71 of these 95.

Cancerous lesions were detected in 36 participants, corresponding to a detection rate of 64.4 per 1,000 MRI examinations, which was about 3.8 times greater than that found with the traditional density method used in the DENSE trial (which evaluated the use of supplemental MRI screening in women with extremely dense breast tissue).

A positive predictive value (PPV) of 37.9% was calculated for women recalled after MRI. For participants assessed as BI-RADS 3, 4, and 5, the PPV was 13.0%, 63.0%, and 85.7%, respectively.

Among all the diagnosed cancers, seven of 36 (19%) presented with multiple mass lesions on MRI while histopathological analysis confirmed multifocality for four (11%). In the histopathological analysis of the surgical specimens, 61% were a combination of invasive and ductal cancer in situ, with five (14%) being in situ only.

Strand and colleagues noted that a cost-effectiveness study based on results from the DENSE study estimated the cost per quality-adjusted life year (QALY) gained for supplemental MRI every 3 years would be €37,181 (approximately $40,200).

“Our results suggest that using AISmartDensity, the cost per QALY would probably be markedly lower given the close to four times higher supplemental cancer detection rate,” they observed.

Strand and colleagues acknowledged the current analysis had several limitations — with a key one being that this report compared the cancer detection rate with results using traditional density in other studies, and not within this study.

Please enable JavaScript to view the

comments powered by Disqus.