Computer-aided detection evaluation methods are not created equal.
Computer-aided detection evaluation methods are not created equal.
复制标题
计算机辅助检测评估方法并不相同。
DOI：
10.1148/radiol.2513081130
复制
发表时间：
2009
期刊：
影响因子：
19.7
通讯作者：
Lorenzo L. Pesce
中科院分区：
文献类型：
--
作者：
R. Nishikawa;Lorenzo L. Pesce
虽然乳房X光检查在降低乳腺癌死亡率方面是有效的，但对数千万妇女进行这种发病率相对较低的复杂疾病的检查是极其困难的。计算机辅助检测（CAD）（特别是用于检测疾病的CADe）被引入以促进这项任务。由于不正确的临床指南和其他政策错误可能会影响成千上万人的死亡率和发病率，因此放射科医生在使用CAD时的诊断性能一直是广泛研究的主题。此外，CAD的临床评价结果可能在CAD器械的报销和CAD研究本身的资金方面产生深远的影响。对临床CAD研究（1-3）和所有CAD临床研究（4-15）的综述暗示或明确地假设，迄今为止用于测量诊断性能的不同方法基本等同。第一种方法是纵向设计（4-7），在4项研究中使用，比较了CAD实施前和CAD实施后的癌症检出率。这些研究的研究人员使用癌症检出率的变化来评估CAD的诊断性能。有时这种方法被称为历史控制方法，因为过去的诊断决策被重新分析。第二种方法是横断面设计，已用于7项研究（8-14）。(We排除了Khoo等人[15]的研究，他们结合双重阅读来研究CAD。）对于每一个病例，放射科医生在不了解其诊断的情况下做出并记录他或她的解释。
While mammography screening is effective in reducing breast cancer mortality, the task of screening tens of millions of women for a complex disease with a relatively low prevalence is extremely difficult. Computer-aided detection (CAD)(specifically, CADe for detection of disease) was introduced to facilitate this task. Because incorrect clinical guidelines and other policy errors could affect tens of thousands of lives with respect to mortality and morbidity, the diagnostic performance of the radiologist when using CAD has been the subject of extensive research. In addition, the results of clinical evaluation of CAD can have farreaching consequences in terms of reimbursement of CAD devices and the funding of CAD research itself. Reviews of clinical CAD studies (1–3) and all the clinical studies of CAD (4–15) result in, either implicitly or explicitly, the assumption that the different approaches used so far to measure diagnostic performance are essentially equivalent. We will argue that this is an erroneous assumption, with the focus on two methods employed.The first method, used in four studies, is longitudinal in design (4–7): The cancer detection rate for a time period before CAD was implemented is compared with that for a time period when CAD was used. Investigators of these studies used the change in cancer detection rate to assess the diagnostic performance of CAD. Sometimes this method is referred to as the historical control method because past diagnostic decisions are reanalyzed. The second method is cross-sectional in design and has been used in seven studies (8–14).(We are excluding the study by Khoo et al [15], who looked at CAD when combined with double reading.) For each case, the radiologist makes and records his or her interpretation without knowledge of the
DOI：
10.1093/jnci/djh067
发表时间：
2004-02-04
期刊：
JOURNAL OF THE NATIONAL CANCER INSTITUTE
影响因子：
--
作者：
Gur, D;Sumkin, JH;Wallace, L
通讯作者：
Wallace, L