Semiautomated magnetic resonance imaging-based breast density measurement from routine sequences: comparison with...

High breast density increases breast cancer risk and decreases mammographic sensitivity, yet prior magnetic resonance imaging (MRI)-based density assessments have relied on specialized sequences or vendor-dependent software. This study aimed to evaluate a semiautomated workflow for MRI-based breast density measurement using routinely acquired MRI sequences and compare its diagnostic performance and concordance with visual and software-based mammographic classifications.

In this single-center retrospective study, we analyzed 113 women (mean age: 60.8 ± 13.6 years) who underwent both breast MRI and mammography. The contralateral, normal breast was evaluated. MRI was performed using a 3.0-T scanner with a 3D T1-weighted gradient-echo sequence. Breast and fibroglandular tissues were segmented using a deep learning-assisted automated method with manual correction. Statistical analyses included correlation analysis, Bland-Altman analysis, and receiver operating characteristic (ROC) analysis.

MRI-based breast density showed a strong correlation with mammographic density (r = 0.781, P < 0.001), although MRI values were on average 27.7% lower. The area under the ROC curve for detecting high-density breasts was 0.869 for MRI and 0.799 for mammography (P = 0.085).

Semiautomated MRI-based breast density assessment using routine MRI sequences allowed for quantitative evaluation of breast composition without using ionizing radiation. MRI-based density measurements exhibited classification performance comparable to mammographic assessment. These findings support the potential utility of MRI for quantitative breast density evaluation, although further validation and workflow standardization are required before broader clinical application.