Department of Radiology, University of Pittsburgh, Magee Womens Hospital, 3362 Fifth Ave, Pittsburgh, PA 15213, USA.
Eur J Radiol. 2012 Nov;81(11):3222-8. doi: 10.1016/j.ejrad.2012.04.018. Epub 2012 May 12.
To improve efficacy of breast cancer screening and prevention programs, it requires a risk assessment model with high discriminatory power. This study aimed to assess classification performance of using computed bilateral mammographic density asymmetry to predict risk of individual women developing breast cancer in near-term. The database includes 451 cases with multiple screening mammography examinations. The first (baseline) examinations of all case were interpreted negative. In the next sequential examinations, 187 cases developed cancer or surgically excised high-risk lesions, 155 remained negative (not-recalled), and 109 were recalled benign cases. From each of two bilateral cranio-caudal view images acquired from the baseline examination, we computed two features of average pixel value and local pixel value fluctuation. We then computed mean and difference of each feature computed from two images. When applying the computed features and other two risk factors (woman's age and subjectively rated mammographic density) to predict risk of cancer development, areas under receiver operating characteristic curves (AUC) were computed to evaluate the discriminatory/classification performance. The AUCs are 0.633±0.030, 0.535±0.036, 0.567±0.031, and 0.719±0.027 when using woman's age, subjectively rated, computed mean and asymmetry of mammographic density, to classify between two groups of cancer-verified and negative cases, respectively. When using an equal-weighted fusion method to combine woman's age and computed density asymmetry, AUC increased to 0.761±0.025 (p<0.05). The study demonstrated that bilateral mammographic density asymmetry could be a significantly stronger risk factor associated to the risk of women developing breast cancer in near-term than woman's age and assessed mean mammographic density.
为了提高乳腺癌筛查和预防计划的效果，需要一种具有高区分能力的风险评估模型。本研究旨在评估使用计算机双侧乳房密度不对称性来预测个体女性在近期内发生乳腺癌风险的分类性能。该数据库包括 451 例具有多次筛查乳房 X 线照片检查的病例。所有病例的第一次（基线）检查均为阴性。在下一次连续检查中，187 例发生癌症或手术切除高危病变，155 例仍为阴性（未召回），109 例为召回良性病例。从基线检查中获得的每对双侧头尾视图图像中，我们计算了平均像素值和局部像素值波动的两个特征。然后，我们计算了从两个图像计算的每个特征的平均值和差异。当将计算得出的特征和其他两个风险因素（女性年龄和主观评估的乳房 X 线密度）应用于预测癌症发展风险时，计算了接收器操作特征曲线（AUC）下的面积，以评估区分/分类性能。当使用女性年龄、主观评估、计算的平均值和乳房 X 线密度的不对称性来区分癌症证实病例和阴性病例时，AUC 分别为 0.633±0.030、0.535±0.036、0.567±0.031 和 0.719±0.027。当使用等权重融合方法将女性年龄和计算的密度不对称性结合使用时，AUC 增加到 0.761±0.025（p<0.05）。该研究表明，双侧乳房 X 线密度不对称性可能是与女性近期发生乳腺癌风险相关的比女性年龄和评估的乳房 X 线密度更显著的风险因素。
