Xing Xiping, Zhao Xinke, Wei Huiping, Li Yingdong
Affiliated hospital of Gansu University of Chinese Medicine.
Gansu University of Traditional Chinese Medicine, Lanzhou, China.
Medicine (Baltimore). 2021 Jan 22;100(3):e23817. doi: 10.1097/MD.0000000000023817.
Computer-aided detection (CAD) system for accurate and automated prostate cancer (PCa) diagnosis have been developed, however, the diagnostic test accuracy of different CAD systems is still controversial. This systematic review aimed to assess the diagnostic accuracy of CAD systems based on magnetic resonance imaging for PCa.
Cochrane library, PubMed, EMBASE and China Biology Medicine disc were systematically searched until March 2019 for original diagnostic studies. Two independent reviewers selected studies on CAD based on magnetic resonance imaging diagnosis of PCa and extracted the requisite data. Pooled sensitivity, specificity, and the area under the summary receiver operating characteristic curve were calculated to estimate the diagnostic accuracy of CAD system.
Fifteen studies involving 1945 patients were included in our analysis. The diagnostic meta-analysis showed that overall sensitivity of CAD system ranged from 0.47 to 1.00 and, specificity from 0.47 to 0.89. The pooled sensitivity of CAD system was 0.87 (95% CI: 0.76-0.94), pooled specificity 0.76 (95% CI: 0.62-0.85), and the area under curve (AUC) 0.89 (95% CI: 0.86-0.91). Subgroup analysis showed that the support vector machines produced the best AUC among the CAD classifiers, with sensitivity ranging from 0.87 to 0.92, and specificity from 0.47 to 0.95. Among different zones of prostate, CAD system produced the best AUC in the transitional zone than the peripheral zone and central gland; sensitivity ranged from 0.89 to 1.00, and specificity from 0.38 to 0.85.
CAD system can help improve the diagnostic accuracy of PCa especially using the support vector machines classifier. Whether the performance of the CAD system depends on the specific locations of the prostate needs further investigation.
已开发出用于准确、自动诊断前列腺癌（PCa）的计算机辅助检测（CAD）系统，然而，不同CAD系统的诊断测试准确性仍存在争议。本系统评价旨在评估基于磁共振成像的CAD系统对PCa的诊断准确性。
系统检索Cochrane图书馆、PubMed、EMBASE和中国生物医学光盘数据库，直至2019年3月，查找原始诊断研究。两名独立的评价者选择基于磁共振成像诊断PCa的CAD研究，并提取所需数据。计算合并敏感性、特异性和汇总接受者操作特征曲线下面积，以评估CAD系统的诊断准确性。
我们的分析纳入了15项研究，涉及1945例患者。诊断性Meta分析显示，CAD系统的总体敏感性为0.47至1.00，特异性为0.47至0.89。CAD系统的合并敏感性为0.87（95%CI：0.76 - 0.94），合并特异性为0.76（95%CI：0.62 - 0.85），曲线下面积（AUC）为0.89（95%CI：0.86 - 0.91）。亚组分析显示，在CAD分类器中，支持向量机产生的AUC最佳，敏感性为0.87至0.92，特异性为0.47至0.95。在前列腺的不同区域中，CAD系统在移行区产生的AUC优于外周区和中央腺；敏感性为0.89至1.00，特异性为0.38至0.85。
CAD系统有助于提高PCa的诊断准确性，尤其是使用支持向量机分类器时。CAD系统的性能是否取决于前列腺的特定位置需要进一步研究。
