AI Can Flag Mammograms for Supplemental MRI

To achieve the highest detection accuracy, international guidelines recommend combining mammography and MRI screening for women with a lifetime breast cancer risk of 20% or higher based on family history. However, in the Netherlands, women with a breast cancer risk ranging from 20% to 50% typically do not have access to additional MRI screening due to limited MRI capacity, high costs of implementation, and inconsistent application of eligibility criteria in clinical practice. Several recent studies have shown the potential of artificial intelligence (AI) to enhance cancer detection in mammography screenings, including detecting cancers that may not be visible through standard mammogram interpretations by radiologists. AI could therefore be used to triage mammograms and identify women who might benefit from supplemental MRI after a negative result according to radiologist interpretation. A new study indicates that AI can effectively identify women at higher breast cancer risk within a select Dutch population. The study, published in Radiology, suggests that using AI in mammogram analysis could improve breast cancer detection by identifying patients who are most likely to benefit from breast MRI scans.

In this retrospective study, conducted by researchers at Radboud University Medical Center (Radboudumc, Nijmegen, Netherlands), women with a personal history of breast cancer, dense breasts, a history of high-risk lesions at biopsy, or those with an increased risk due to family history (but no genetic mutations) were classified as "intermediate risk." The researchers utilized a commercially available AI system to analyze the 2D screening mammograms of women they identified as intermediate risk to detect patients most likely to have cancers that were not visible on mammograms (mammographically occult cancers), indicating the need for supplemental MRI. The study cohort included 1,833 consecutive women who underwent at least one screening MRI in combination or alternated with a mammogram between 2003 and 2020, sourced from the patient breast MRI database at Radboudumc. Women with a lifetime breast cancer risk greater than 50% were excluded.

A total of 3,358 mammography exams were performed on 875 women. Of these, 2,819 (84%) exams from 760 women (with an average age of 48.9 years) were processed by the AI system and assigned a case-based suspicion score (ranging from 0 to 10) that ranked the likelihood of malignancy. Combined screening detected 37 (1.3%) breast cancers. In 19 (51%) of these cases, the cancer was not visible on mammography. Using a threshold score of 5 (which allowed supplemental MRI screening for 50% of the women), the AI system selected 31 (84%) of the breast cancer-positive exams for additional MRI screening, including 68% of exams with occult breast cancer that had been missed in the radiologists' initial reading.

"AI could potentially triage mammograms performed in the subgroup and select women that could potentially benefit from supplemental MRI after a negative mammogram," said the study's lead author, Suzanne van Winkel, R.N., M.Sc. "Using AI to triage the mammograms of populations who are not yet eligible for MRI may improve screening results while simultaneously reducing unnecessary costs."

Related Links:
Radboud University Medical Center