Mammographic density defined by higher than conventional brightness threshold better predicts breast cancer risk for full-field digital mammograms

Abstract

IntroductionWhen measured using the computer-assisted method CUMULUS, mammographic density adjusted for age and body mass index predicts breast cancer risk. We asked if new mammographic density measures defined by higher brightness thresholds gave better risk predictions.MethodsThe Korean Breast Cancer Study included 213 women diagnosed with invasive breast cancer and 630 controls matched for age at full-field digital mammogram and menopausal status. Mammographic density was measured using CUMULUS at the conventional threshold (Cumulus), and in effect at two increasingly higher thresholds, which we call Altocumulus and Cirrocumulus, respectively. All measures were Box-Cox transformed and adjusted for age, body mass index, menopausal status and machine. We used conditional logistic regression to estimate the change in Odds PER standard deviation of transformed and Adjusted density measures (OPERA). The area under the receiver operating characteristic curve (AUC) was estimated.ResultsCorresponding Altocumulus and Cirrocumulus density measures were correlated with Cumulus measures (r approximately 0.8 and 0.6, respectively). Altocumulus and Cirrocumulus measures were on average 25 % and 80 % less, respectively, than the Cumulus measure. For dense area, the OPERA was 1.18 (95 % confidence interval: 1.01−1.39, P = 0.03) for Cumulus; 1.36 (1.15−1.62, P < 0.001) for Altocumulus; and 1.23 (1.04−1.45, P = 0.01) for Cirrocumulus. After fitting the Altocumulus measure, the Cumulus measure was no longer associated with risk. After fitting the Cumulus measure, the Altocumulus measure was still associated with risk (P = 0.001). The AUCs for dense area was 0.59 for the Altocumulus measure, greater than 0.55 and 0.57 for the Cumulus and Cirrocumulus measures, respectively (P = 0.001). Similar results were found for percentage dense area measures.ConclusionsAltocumulus measures perform better than Cumulus measures in predicting breast cancer risk, and Cumulus measures are confounded by Altocumulus measures. The mammographically bright regions might be more aetiologically important for breast cancer, with implications for biological, molecular, genetic and epidemiological research and clinical translation.

Going Beyond Conventional Mammographic Density to Discover Novel Mammogram-Based Predictors of Breast Cancer Risk

“…The risk discrimination from conventional mammographic density compares favourably with that achieved by the latest polygenic risk scores for which the OPERA is about 1.6 [9]. Our new mammogram-based risk measures have OPERAs of 1.6 or higher [8,[10][11][12][13][14] and we will now discuss how this came about.…”

Breast Density and Risk of Breast Cancer in Asian Women: A Meta-analysis of Observational Studies

“…The study subjects of Lee et al [30] were participants of the Singapore Breast Cancer Screening Programme (SBCSP), and two duplicates were removed [23,35]. The study subjects of Kim et al [20] were patients at the Samsung Medical Cancer in Korea, and one duplicate was removed [36]. …”

Genome‐wide association study of peripheral blood DNA methylation and conventional mammographic density measures

“…In our study, mammographic density was defined by the conventional brightness threshold using the CUMULUS software and has been referred to as Cumulus . New mammographic density measures defined by higher pixel brightness thresholds, Cirrocumulus and Altocumulus , have been found to predict breast cancer risk better than the Cumulus measures, and multivariate analyses suggest that the Cumulus measures are only surrogates for the causal aspects of a mammographic image that predict risk . A new breast cancer risk‐predicting measure called Cirrus has also been developed using machine learning .…”

“…New mammographic density measures defined by higher pixel brightness thresholds, Cirrocumulus and Altocumulus, have been found to predict breast cancer risk better than the Cumulus measures, and multivariate analyses suggest that the Cumulus measures are only surrogates for the causal aspects of a mammographic image that predict risk. 25 A new breast cancer risk-predicting measure called Cirrus has also been developed using machine learning. 26 The Cirrocumulus, Altocumulus and Cirrus measures might be more etiologically relevant to breast cancer risk.…”