From the Department of Radiology (Y.Y., E.S., M.H.) and Department of Preventive Medicine and Public Health, Biostatistics Unit at Clinical and Translational Research Center (T.A.), Keio University School of Medicine, 35 Shinanomachi, Shinjuku-ku, Tokyo 160-8582, Japan; Department of Radiology, Nippon Koukan Hospital, Kawasaki, Japan (Y.Y., K.O.); and Department of Radiology, St. Luke's International Hospital, Tokyo, Japan (M.M., Y.S.).
Radiology. 2018 Apr;287(1):333-339. doi: 10.1148/radiol.2017170405. Epub 2017 Dec 4.
Purpose To assess the value of a computer-aided detection (CAD) system for the detection of pulmonary nodules on chest tomosynthesis images. Materials and Methods Fifty patients with and 50 without pulmonary nodules underwent both chest tomosynthesis and multidetector computed tomography (CT) on the same day. Fifteen observers (five interns and residents, five chest radiologists, and five abdominal radiologists) independently evaluated tomosynthesis images of 100 patients for the presence of pulmonary nodules in a blinded and randomized manner, first without CAD, then with the inclusion of CAD marks. Multidetector CT images served as the reference standard. Free-response receiver operating characteristic analysis was used for the statistical analysis. Results The pooled diagnostic performance of 15 observers was significantly better with CAD than without CAD (figure of merit [FOM], 0.74 vs 0.71, respectively; P = .02). The average true-positive fraction and false-positive rate per all cases with CAD were 0.56 and 0.26, respectively, whereas those without CAD were 0.47 and 0.20, respectively. Subanalysis showed that the diagnostic performance of interns and residents was significantly better with CAD than without CAD (FOM, 0.70 vs 0.62, respectively; P = .001), whereas for chest radiologists and abdominal radiologists, the FOM with CAD values were greater but not significantly: 0.80 versus 0.78 (P = .38) and 0.74 versus 0.73 (P = .65), respectively. Conclusion CAD significantly improved diagnostic performance in the detection of pulmonary nodules on chest tomosynthesis images for interns and residents, but provided minimal benefit for chest radiologists and abdominal radiologists. RSNA, 2017 Online supplemental material is available for this article.
目的 评估计算机辅助检测 (CAD) 系统在胸部断层合成图像中检测肺结节的价值。
材料与方法 当日对 50 例有和 50 例无肺结节的患者分别进行胸部断层合成和多排螺旋 CT 检查。15 名观察者（5 名实习医生和住院医师、5 名胸部放射科医生和 5 名腹部放射科医生）以盲法和随机的方式独立评估 100 例患者的断层合成图像，首先不使用 CAD，然后使用 CAD 标记。多排螺旋 CT 图像作为参考标准。采用自由响应接受者操作特征分析进行统计学分析。
结果 15 名观察者的总体诊断性能在使用 CAD 时明显优于不使用 CAD（受试者工作特征曲线下面积 [FOM]，分别为 0.74 和 0.71；P =.02）。所有病例中 CAD 的平均真阳性率和假阳性率分别为 0.56 和 0.26，而不使用 CAD 时分别为 0.47 和 0.20。亚分析显示，使用 CAD 时实习医生和住院医师的诊断性能明显优于不使用 CAD（FOM，分别为 0.70 和 0.62；P =.001），而对于胸部放射科医生和腹部放射科医生，CAD 的 FOM 值较高，但无统计学意义：0.80 比 0.78（P =.38）和 0.74 比 0.73（P =.65）。
结论 对于实习医生和住院医师，CAD 可显著提高胸部断层合成图像中肺结节的检测诊断性能，但对胸部放射科医生和腹部放射科医生的帮助较小。
RSNA，2017 在线补充材料可从本文获得。
