Author: Yile Fang, Leslie Lamb, Nathaniel David Mercaldo, Kai Yang 👨🔬
Affiliation: Massachusetts General Hospital 🌍
Purpose: To quantitatively evaluate power-law exponent β as a potential image-based breast cancer risk factor.
Methods: Two groups of breast cancer screening cohorts (target vs. control, 20 subjects in each group) were included by matching race, age, and breast density. The target group had breast cancer diagnosed within one year of the mammographic exam. A custom algorithm was developed to decompress DICOM files, automatically extract breast tissue regions, and perform power spectrum analysis within these regions. The power-law exponent β, derived as the slope of the linear relationship between log-frequency and log-power, was used as a quantitative measure of breast texture. β was evaluated across four parameters: modality (2D vs. 3D), breast cancer status (cancer detected within one year vs. no cancer within one year), race (White vs. Black), and breast density (scattered fibroglandular vs. heterogeneously dense). Multivariate regression and paired t-tests were applied, with statistical significance defined as p < 0.05.
Results: A total of 9,198 3D slices or 2D images from 389 exams in 40 women were analyzed. β values showed strong power-law behavior across all images. Significant differences in β were observed for cancer status (p = 0.00941), modality (p = 4.94 × 10-35), and breast density (p = 3.96 × 10-19), but not for race (p = 0.844). β was positively correlated with cancer presence (coefficient = 0.0641) and negatively correlated with 2D modality (coefficient = -0.309) and heterogeneously dense breasts (coefficient = -0.209). The correlation with race was weak and non-significant (coefficient = -0.00437 for white patients).
Conclusion: The value β effectively captures quantitative differences in breast texture associated with cancer presence, imaging modality, and breast density. These findings highlight the potential for β as an image-based biomarker for breast cancer risk prediction.