Cockmartin Lesley, Marshall Nicholas W, Van Ongeval Chantal, Aerts Gwen, Stalmans Davina, Zanca Federica, Shaheen Eman, De Keyzer Frederik, Dance David R, Young Kenneth C, Bosmans Hilde
Department of Imaging and Pathology, Division of Medical Physics & Quality Assessment, KU Leuven, Herestraat 49 Box 7003, 3000 Leuven, Belgium. Department of Radiology, University Hospitals Leuven, Herestraat 49, 3000 Leuven, Belgium.
Phys Med Biol. 2015 May 21;60(10):3939-58. doi: 10.1088/0031-9155/60/10/3939. Epub 2015 Apr 24.
This paper introduces a hybrid method for performing detection studies in projection image based modalities, based on image acquisitions of target objects and patients. The method was used to compare 2D mammography and digital breast tomosynthesis (DBT) in terms of the detection performance of spherical densities and microcalcifications. The method starts with the acquisition of spheres of different glandular equivalent densities and microcalcifications of different sizes immersed in a homogeneous breast tissue simulating medium. These target objects are then segmented and the subsequent templates are fused in projection images of patients and processed or reconstructed. This results in hybrid images with true mammographic anatomy and clinically relevant target objects, ready for use in observer studies. The detection study of spherical densities used 108 normal and 178 hybrid 2D and DBT images; 156 normal and 321 hybrid images were used for the microcalcifications. Seven observers scored the presence/absence of the spheres/microcalcifications in a square region via a 5-point confidence rating scale. Detection performance in 2D and DBT was compared via ROC analysis with sub-analyses for the density of the spheres, microcalcification size, breast thickness and z-position. The study was performed on a Siemens Inspiration tomosynthesis system using patient acquisitions with an average age of 58 years and an average breast thickness of 53 mm providing mean glandular doses of 1.06 mGy (2D) and 2.39 mGy (DBT). Study results showed that breast tomosynthesis (AUC = 0.973) outperformed 2D (AUC = 0.831) for the detection of spheres (p < 0.0001) and this applied for all spherical densities and breast thicknesses. By way of contrast, DBT was worse than 2D for microcalcification detection (AUC2D = 0.974, AUCDBT = 0.838, p < 0.0001), with significant differences found for all sizes (150-354 µm), for breast thicknesses above 40 mm and for heights above the detector of 20 mm and above. In conclusion, the hybrid method was successfully used to produce images for a detection study; results showed breast tomosynthesis outperformed 2D for spherical densities while further optimization of DBT for microcalcifications is suggested.
本文介绍了一种基于目标物体和患者图像采集，在基于投影图像的模态中进行检测研究的混合方法。该方法用于比较二维乳腺钼靶摄影和数字乳腺断层合成（DBT）在检测球形密度和微钙化方面的性能。该方法首先采集不同腺体等效密度的球体以及不同大小的微钙化，将它们浸入均匀的乳腺组织模拟介质中。然后对这些目标物体进行分割，并将后续模板融合到患者的投影图像中进行处理或重建。这就产生了具有真实乳腺钼靶解剖结构和临床相关目标物体的混合图像，可用于观察者研究。球形密度的检测研究使用了108幅正常的以及178幅二维和DBT混合图像；微钙化检测使用了156幅正常图像和321幅混合图像。7名观察者通过5分置信度评分量表对正方形区域内球体/微钙化的有无进行评分。通过ROC分析比较二维和DBT的检测性能，并对球体密度、微钙化大小、乳腺厚度和z位置进行子分析。该研究在西门子Inspiration断层合成系统上进行，使用平均年龄58岁、平均乳腺厚度53毫米的患者图像，二维平均腺体剂量为1.06毫戈瑞，DBT为2.39毫戈瑞。研究结果表明，在检测球体方面，乳腺断层合成（AUC = 0.973）优于二维（AUC = 0.831）（p < 0.0001），这适用于所有球形密度和乳腺厚度。相比之下，在微钙化检测方面，DBT比二维差（二维AUC = 0.974，DBT AUC = 0.838，p < 0.0001），在所有大小（150 - 354微米）、乳腺厚度超过40毫米以及探测器上方高度20毫米及以上时均存在显著差异。总之，该混合方法成功用于生成检测研究的图像；结果表明乳腺断层合成在检测球形密度方面优于二维，同时建议对DBT检测微钙化进行进一步优化。
