Bessho Yuichi, Yamaguchi Michihiro, Fujita Hideki, Azuma Masami
Department of Radiology, Osaka Prefectural Medical Center for Respiratory and Allergic Disease, Osaka Prefectural Hospital Organization, 3-7-1 Habikino, Habikino, Osaka, Japan.
Acad Radiol. 2009 Aug;16(8):940-6. doi: 10.1016/j.acra.2009.03.006. Epub 2009 May 14.
The purposes of the present study were to investigate the impact of viewing size on soft-copy diagnosis for detecting abnormalities on digital chest radiographs and to verify the usefulness of reduced digital chest radiography.
Receiver-operating characteristic (ROC) and localized ROC (LROC) analysis of clinical images was performed using the standard digital image database of the Japanese Radiation Technology Society. A total of 30 images with and 20 images without nodule samples were extracted randomly from the database and used for ROC analysis. A total of 100 images were prepared for observation of reduced and nonreduced images. Observers' viewing images were adjusted as nonreduced size (large, 30 x 30 cm) and reduced size (small, 15 x 15 cm).
The estimated mean areas under the LROC curves were 0.665 +/- 0.071 for large-size images and 0.669 +/- 0.087 for small-size images. However, no statistically significant difference was found between the two groups (P = .823). The viewing time of small-size images (1201.4 seconds) was significantly shorter than that of large-size images (1719.7 seconds). A statistically significant difference was also found in viewing times between reduced and nonreduced digital images with Wilcoxon's signed-rank test (P < .05).
This study provided important information that there was a statistically significant difference in viewing times between reduced and nonreduced digital chest radiographic images, whereas no significant difference was found in areas under the LROC curves for the accuracy of diagnosis between the two groups. However, the double-check method for chest x-ray screening would be done efficiently by changing the viewing size to improve the specificity of diagnosis.
本研究的目的是调查观察尺寸对数字胸部X光片检测异常的软拷贝诊断的影响，并验证缩小尺寸的数字胸部X光片的实用性。
使用日本放射技术协会的标准数字图像数据库对临床图像进行接收者操作特征（ROC）和局部ROC（LROC）分析。从数据库中随机提取30张有结节样本的图像和20张无结节样本的图像用于ROC分析。共准备了100张图像用于观察缩小和未缩小的图像。观察者的观察图像调整为未缩小尺寸（大，30×30厘米）和缩小尺寸（小，15×15厘米）。
大尺寸图像的LROC曲线下估计平均面积为0.665±0.071，小尺寸图像为0.669±0.087。然而，两组之间未发现统计学上的显著差异（P = 0.823）。小尺寸图像的观察时间（1201.4秒）明显短于大尺寸图像（1719.7秒）。使用Wilcoxon符号秩检验，缩小和未缩小的数字图像之间的观察时间也存在统计学上的显著差异（P < 0.05）。
本研究提供了重要信息，即缩小和未缩小的数字胸部X光片之间的观察时间存在统计学上的显著差异，而两组之间LROC曲线下面积在诊断准确性方面没有显著差异。然而，通过改变观察尺寸以提高诊断特异性，可以有效地进行胸部X光筛查的复查方法。
