Qiao Mengyun, Hu Yuzhou, Guo Yi, Wang Yuanyuan, Yu Jinhua
Department of Electronic Engineering, Fudan University, Shanghai, China.
J Ultrasound Med. 2018 Feb;37(2):403-415. doi: 10.1002/jum.14350. Epub 2017 Aug 14.
This work focused on extracting novel and validated digital high-throughput features to present a detailed and comprehensive description of the American College of Radiology Breast Imaging Reporting and Data System (BI-RADS) with the goal of improving the accuracy of ultrasound breast cancer diagnosis.
First, the phase congruency approach was used to segment the tumors automatically. Second, high-throughput features were designed and extracted on the basis of each BI-RADS category. Then features were selected based on the basis of a Student t test and genetic algorithm. Finally, the AdaBoost classifier was used to differentiate benign tumors from malignant ones.
Experiments were conducted on a database of 138 pathologically proven breast tumors. The system was compared with 6 state-of-art BI-RADS feature extraction methods. By using leave-one-out cross-validation, our system achieved a highest overall accuracy of 93.48%, a sensitivity of 94.20%, a specificity of 92.75%, and an area under the receiver operating characteristic curve of 95.67%, respectively, which were superior to those of other methods.
The experiments demonstrated that our computerized BI-RADS feature system was capable of helping radiologists detect breast cancers more accurately and provided more guidance for final decisions.
本研究致力于提取新颖且经过验证的数字高通量特征，以对美国放射学会乳腺影像报告和数据系统（BI-RADS）进行详细全面的描述，旨在提高超声乳腺癌诊断的准确性。
首先，采用相位一致性方法自动分割肿瘤。其次，基于每个BI-RADS类别设计并提取高通量特征。然后，基于学生t检验和遗传算法选择特征。最后，使用AdaBoost分类器区分良性肿瘤和恶性肿瘤。
在一个包含138个经病理证实的乳腺肿瘤的数据库上进行了实验。该系统与6种先进的BI-RADS特征提取方法进行了比较。通过留一法交叉验证，我们的系统分别达到了93.48%的最高总体准确率、94.20%的灵敏度、92.75%的特异性以及95.67%的受试者操作特征曲线下面积，均优于其他方法。
实验表明，我们的计算机化BI-RADS特征系统能够帮助放射科医生更准确地检测乳腺癌，并为最终决策提供更多指导。
