微钙化的计算机辅助分析对乳腺导管原位癌及微浸润的诊断价值

Performance of computer-aided detection of microcalcifications for ductal carcinoma in situ with or without microinvasion

目的:探讨在乳腺钼靶X线摄影检查中采用计算机辅助检测系统(CAD)对伴有微钙化的乳腺导管原位癌(DCIS)及其微浸润的诊断效能。方法:回顾性分析经本院乳腺钼靶X线摄影检查发现微钙化并经病理学证实的654例乳腺病变患者的病例资料,其中良性病变451例,DCIS(有/无微浸润)203例,使用CAD系统进行微钙化特征的提取和分类,比较14个特征参数在两组病变的差异,分析CAD的诊断效能。结果:所有特征参数中线样分枝状钙化数、细颗粒状微钙化率、段样分布、种群密度在两组病变间的差异有统计学意义(P<0.05),原位癌及微浸润组的特征参数值较高;四个特征参数对判别两组病变的ROC曲线下面积(AUC)分别为0.752、0.734、0.729和0.714,低于CAD系统综合检测法的判别效能(AUC=0.873),其诊断敏感度、特异度和符合率分别为97.3%、75.8%和85.1%。结论:乳腺导管原位癌及微浸润的微钙化特征具有相对特异性,基于多特征分析的CAD系统对该类病变可达到较高的诊断效能,可为乳腺癌早期病变的诊断和临床个性化治疗提供重要的参考依据。

Objective：To explore the diagnostic performance of computer-aided detection （CAD） for breast ductal carcinoma in situ （DCIS） and DCIS with mieroinvasion by detection and classification of microcalcifications on mammography. Methods：Data of 654 female patients with breast lesoins accompanied with microealcification on mammography confirmed by pathology were retrospectively studied,including benign lesions in 451 cases and DCIS or DCIS with mieroinvasion in 203 cases. After an automatic detection of microcalcifieation features by CAD,the differences of 14 features were compared between the two groups, and the diagnostic performance of CAD was analyzed. Results：In all 14 microcalcification features,only 4 （number of linear branching-like microealeifications,ratio of granular microcalcifications, sample distribution and population density） had significant differences between the two groups （P〈0.05）, the value of these features tended to be higher in DCIS and DCIS with microinvasion group. Area under the ROC curve （AUC） of the four features were 0. 752,0. 734, 0. 729 and 0. 714, CAD comprehensive method using all features for analysis had the best diagnostic efficacy （AUC= 0. 873） with sensitivity of 97.3 %, specificity of 75.8%, and accuracy of 85.1%. Conclusion： Microcalcifications of DCIS and DCIS with microinvasion have relatively characteristic features. Based on comprehensive method using multiple microinvasion features for analysis, the diagnostic performance of CAD can be highly improved, which may facilitate the early detection and personal management of DCIS and DCIS with microinvasion.