Breast Density Histogram Analysis: The Role of Fat in Outcome Prediction After Whole Breast Radiation Therapy

Although the link between breast tissue density and cancer risk is well established, its influence on posttreatment outcomes remains unclear. Clarifying the role of breast density in these outcomes could enhance treatment personalization and patient stratification, potentially enabling clinicians to adapt radiation therapy plans to individual breast tissue composition. This study evaluated the role of pre-radiation therapy breast densitometric state in relation to local/distant progression, overall survival, and molecular subtypes.

A mono-institutional cohort of 1127 early-stage breast cancer patients treated with 40 Gy/15 fractions (2009-2017) was analyzed. Clinical target volume segmentations from planning computed tomography were used to extract hounsfield units (HU) histograms (range, -200 HU, +50 HU), excluding clips and artifacts. Fatty and fibroglandular tissues were quantified based on selected HU ranges. Extracted parameters included volume, mean/median HU, standard deviation, percentiles, and histogram shape indices. Densitometric, clinical, and combined predictive models were developed using multivariate Cox regression, minimizing redundancy. Internal validation involved 1000 bootstrap iterations. A prognostic index was calculated for each model, and Kaplan-Meier analysis stratified patients into risk groups. Densitometric prognostic indices were also tested for potential association with molecular subtypes (luminal A/B, Her2+, triple-negative breast cancer).

Median follow-up was 6 years (interquartile range, 4-8): local relapse/distant relapse/death rates were 2.3%/4.1%/7.0%, respectively. The combination of % fat volume (VFAT%) and HU percentiles was moderately associated with outcomes (densitometry models, C-index, 0.60-0.61): lower HU values and higher VFAT% were associated to better outcome. Clinical models showed higher predictive performance (C-index, 0.72-0.76), with key factors including tumor stage, nodal status, age, and triple-negative breast cancer subtype. Combined models (C-index, 0.71-0.79) improved the performances of the clinical model for distant progression-free survival. No significant association was found between densitometric models and molecular subtypes.

Clinical features are the strongest predictors, though fat-related metrics offered additional biological insights, improving the ability of local and distant relapse prediction.