MRI计算机辅助诊断系统对不同经验医师鉴别乳腺小肿块良恶性的价值  

Value of MRI-CAD in the differentiation between small benign and malignant breast masses for readers with different level of experience

作　　者：宋颖[1] 欧阳汉[1] 叶枫[1] 李静[1] 周纯武[1] 

机构地区：[1]中国医学科学院肿瘤医院影像诊断科,北京100021

出　　处：《放射学实践》2016年第7期638-643,共6页Radiologic Practice

摘　　要：目的：探讨应用乳腺磁共振计算机辅助系统（MRI-CAD）对于不同经验的医师鉴别乳腺小肿块良恶性的价值。方法：回顾性分析行MRI检查的肿块型乳腺病灶235个（≤2.0cm）,均经病理证实或随诊2年以上。比较4名医师（2名经验缺乏者和2名经验丰富者）使用CAD系统前后的诊断敏感度、特异度。采用ROC曲线评价并比较4名医师使用CAD系统前后的诊断符合率。采用Kappa检验评价使用CAD系统前后医师间的一致性。结果：235个病灶中,155个良性病灶,80个恶性病灶。4位医师均检出了所有病灶。应用CAD系统前,4名医师诊断乳腺癌的特异度分别为55.5%、58.1%、72.3%、74.8%,使用CAD后分别为72.3%、76.1%、75.5%、70.3%;医师一、医师二（两名低年资医师）的诊断特异度前后差异具有统计学意义（P〈0.001）,医师三及医师四差异无统计学意义（P3=0.404,P4=0.265）。四位医师在使用CAD系统后诊断敏感度均略有提高,但差异均无统计学意义（P1=1.000,P2=1.000,P3=1.000,P4=0.480）。医师一、医师二及医师三在应用CAD系统后,总体诊断符合率均有明显提高,ROC曲线下面积分别由0.899（95%CI：0.853~0.934）、0.839（95%CI：0.785~0.883）、0.929（95%CI：0.888~0.958）提高至0.947（95%CI：0.910~0.972）、0.987（95%CI：0.962~0.997）、0.971（95%CI：0.940~0.988）,且差异具有统计学意义（P1=0.002,P2〈0.001,P3〈0.001）。医师四尽管在使用CAD系统后总体诊断符合率亦有提高,ROC曲线下面积由0.940（95%CI：0.901~0.966）提高至0.960（95%CI：0.927~0.981）,但差异无统计学意义（P=0.11）。四位医师在应用CAD系统后诊断一致性提高（P=0.004）。结论：MRI-CAD系统的应用有助于提高医师诊断乳腺小肿块的准确度及特异度,使得不同经验程度医师的诊断一致性提高,减少了不必要的活检,可作为提高缺乏经验医师诊断符合率的有力工具。Objective：To investigate the value of MRI in addition with computer-aided diagnostic（CAD）system in the evaluation of contrast-enhanced breast MRI for readers with different level of experience.Methods：A total of 235 small breast lesions（≤2.0cm）assessed on MRI were reviewed retrospectively.All had either pathology diagnosis or follow-up for2 years.The diagnosis sensitivity,specificity by 4readers（experienced and inexperienced）before and after using CAD system were compared.ROC was used to evaluate the diagnosis accuracy before and after using CAD of the 4readers.Interobserver variations were evaluated using kappa statistics.Results：Of the 235 breast lesions,there were 155 benign and 80 malignant masses.All lesions were detected by four readers.Before CAD was used,the specificity of these 4readers was55.5%、58.1%、72.3%、74.8%,respectively;which was 72.3%、76.1%、75.5%、70.3%respectively after using CAD.Statistic difference（P〈0.001）was existed between the 2inexperienced readers before and after using CAD;whereas no statistic difference（P=0.404 for reader 3,P=0.265 for reader 4）was existed in the 2experienced readers before and after using CAD.Diagnostic sensitivity in all four readers increased slightly by the use of CAD,yet there was no statistical significance（P1=1.000,P2=1.000,P3=1.000,P4=0.480）.The overall performance found in both inexperienced and one experienced readers（reader 3）improved significantly after using CAD（ROC,0.899,95%CI：0.853~0.934 vs 0.947,95% CI：0.910~0.972,P=0.002 in reader 1;ROC,0.839,95% CI：0.785~0.883 vs 0.987,95%CI：0.962~0.997,P〈0.001 in reader 2;ROC,0.929,95% CI：0.888~0.958 vs 0.971,95% CI：0.940~0.988,P〈0.001 in reader 3）,with significant statistic differences.Although there was improvement of overall accuracy after using CAD in reader 4,ROC 0.940,95%CI：0.901~0.966 vs 0.960,95% CI：0.927~0.981,no significant statistic difference was found（P=0.11）.The level of agreement between the readers improved significantly when u

关 键 词：乳腺肿瘤 图像处理 计算机辅助 磁共振成像 

分 类 号：R445.2[医药卫生—影像医学与核医学] R737.9[医药卫生—诊断学]