Fourier multiscale analysis links bilateral breast asymmetry to breast cancer risk
A matched case-control study of 368 pairs found that bilateral breast asymmetry measured in the Fourier domain was associated with breast cancer across mammography formats, with the strongest signal emerging from raw full-field digital mammography images. The work is built on a simple clinical premise: the two breasts usually share similar parenchymal architecture, so a measurable departure from symmetry may reflect localized pathology or broader tissue changes linked to malignancy. In practical terms, the study asks whether a mathematically derived asymmetry measure can help separate cancer cases from controls when the images are analyzed at diagnosis or near that time.
The investigators used contralateral breast images and summarized image power in concentric radial bands across the Fourier plane, a multiscale approach that compresses texture information into a bilateral asymmetry measure, or ASM. This is not a visual impression of asymmetry but a quantitative readout of structural divergence across spatial frequencies. In breast imaging, that matters because malignant lesions and cancer-associated parenchymal remodeling do not appear at a single scale; they may alter coarse patterns, fine texture, or both. A method that captures those shifts across scales could prove more informative than a single global comparison of left and right breast appearance.
Raw mammograms carried the strongest association
Raw FFDM images with 70 µm pixel spacing produced the most pronounced association with breast cancer. For each standard deviation increase in ASM, the odds ratio was 1.90, with a 95% confidence interval of 1.58 to 2.29, and the area under the receiver operating characteristic curve reached 0.72, with a 95% confidence interval of 0.67 to 0.76. In a screening or risk-stratification context, those numbers suggest a meaningful discriminatory signal, though not one that would stand alone as a diagnostic test. A metric with an AUC in this range can contribute to a broader model, especially when combined with mammographic density, clinical history, and other imaging features.
The biologic logic is straightforward. Cancer can distort glandular architecture, create focal density changes, and remodel surrounding tissue in ways that break the expected bilateral pattern. Fourier analysis is well suited to this problem because it detects differences in texture energy across spatial frequencies rather than relying only on localized visual contrast. A lesion need not be large to perturb the symmetry of a breast if it induces broader parenchymal change around it. The study’s strongest result came from the image type with the finest spatial sampling, suggesting that small-scale detail still carries meaningful diagnostic information.
Clinical FFDM and DBT were weaker but still informative
When the same analytical framework was applied to clinical FFDM images and to DBT synthetic two-dimensional images, the association persisted but weakened. Clinical FFDM yielded an odds ratio of 1.31 per standard deviation increase in ASM, with a 95% confidence interval of 1.11 to 1.54, and an AUC of 0.63, with a 95% confidence interval of 0.58 to 0.67. DBT images showed an odds ratio of 1.48, with a 95% confidence interval of 1.25 to 1.76, and an AUC of 0.65, with a 95% confidence interval of 0.60 to 0.70. Both remained statistically significant, but both fell short of the performance seen with raw FFDM images.
The difference between imaging platforms is clinically meaningful. Clinical FFDM and DBT had lower effective spatial resolution than raw FFDM, and that reduction likely blunted the asymmetry signal. Even so, DBT performed better than its resolution alone might have suggested. That pattern hints that the asymmetry metric is not merely extracting noise from pixel-level detail; it may be robust to some degree of image degradation while still benefitting from the sharper information preserved in raw acquisition data. For algorithm development, this matters because many real-world breast imaging systems differ in processing pipelines, reconstruction methods, and pixel spacing.
A single sentence captures the main tension in the data: better spatial detail produced better breast cancer discrimination.
Why bilateral asymmetry may matter in early detection modeling
This study does not claim that bilateral asymmetry replaces established screening tools. It does support the idea that a quantitative asymmetry measure could become one component of a multivariable early detection model. That is a more realistic target than standalone diagnosis. Breast cancer screening already depends on layered interpretation: image appearance, patient risk, tissue density, and radiologist judgment all shape the final assessment. A multiscale asymmetry metric could add a machine-readable feature that captures anatomy in a way the human eye may not quantify consistently.
The matched case-control design also frames the findings carefully. Women had mammograms acquired approximately at the time of diagnosis, which makes this an association study rather than a prospective prediction study. The analysis adjusted for body mass index and ethnicity, both of which can influence breast tissue composition and mammographic interpretation. Even with those controls, the central result remained: bilateral asymmetry was associated with cancer, and the magnitude of association depended on how the image was acquired and processed.
The clinical promise here lies in modeling, not in immediate deployment. Any next step will need to test whether ASM improves risk prediction outside the case-control setting, where prevalence is more balanced and performance estimates are harder to inflate. Prospective validation will also need to determine whether the signal is stable across scanners, populations, and clinical workflows. If it is, bilateral asymmetry could become a useful feature in software intended to flag subtle changes before cancer becomes obvious on conventional review.
The current evidence leaves one practical question unanswered: whether this signal will remain strong enough in real screening populations to change referral or follow-up decisions. For now, the finding is narrower and more concrete. Fourier-based bilateral asymmetry tracks with breast cancer, and raw FFDM preserves that signal best.