微钙化的计算机辅助分析与乳腺导管原位癌病理分级的对照研究

Computer-aided detection with microcalcifications for DCIS and the comparative analysis with pathological grading

目的对照研究计算机辅助检测(computer-aided detection,CAD)自动提取的微钙化多种特征与乳腺导管原位癌(ductal carcinoma in situ,DCIS)不同病理分级的相关性,探讨所有特征对于DCIS不同分级的意义,为开发辅助临床早期诊断和个体化治疗DCIS的计算机辅助诊断系统提供参考。方法经本院乳腺钼靶X线摄影检查发现微钙化并经最终病理学证实的203例DCIS患者影像资料,按高、中、低级别病理结果分组,CAD检测微钙化并自动提取反映微钙化形态学、统计学、纹理、分布特征的14个参数,进行对比分析。结果 1所有微钙化特征中,线样分支状钙化数、细颗粒状微钙化率、总数目、簇状分布率、种群密度在低、中、高级别3组间比较和低、高级别2组间比较时差异均具有统计学意义(P<0.01),较高的病理级别容易出现较高的线样分支状钙化数、总数目、簇状分布率、种群密度,较低的病理级别出现细颗粒状微钙化的概率较高。2除线样分支状钙化数外,其余所有特征参数在中级别组和低级别或高级别组间比较时,差异均无统计学意义(P>0.05)。3微钙化的四种分布类型仅在高级别组出现统计学差异,其中簇状分布的出现概率相对较高。结论通过CAD辅助检测DCIS微钙化获得的多种特征参数,与不同生物学行为的病理分级相关性不同,有利于深入研究计算机辅助诊断DCIS,为早期诊断DCIS和临床个性化治疗提供重要的参考依据。

Objective To evaluate the features of microcalcifications extracted by Computer-aided System （CAD） for Ductal Carcinoma in Situ （DCIS）and analyze their relations with nuclear grade,thus provide good refer-ence for the development of CAD to assist early diagnosis and tailored therapy. Methods Two hundreds and three female patients with microcalcifications on mammography were pathologically confirmed and retrospectively studied. From low nuclei grade to high nuclei grade ,all cases were divided into three groups. Automatic extraction and detec-tion of 14 characterized parameters to reflect 4 kinds of features, including morpahological ,statastical,texture and dis-tribution of microcalcifications.Then compared the difference of all features in different groups. Results ①Five fea-tures,including number of linear branching-like microcalcifications, rate of granular microcalcifications, total number, clustered distribution and population density,were not only statistically different among different groups but also be-tween low and high nuclei grade （P〈0.01）. Except rate of granular microcalcifications,the other parameters tended to increase with the grade. ②Except the number of linear branching-like microcalcifications,the rest features had no significant difference between medium nuclei grade and each of the other two grades. ③The difference of four kinds of distribution was statistically significant only in high nuclei grade ,in which clustered distribution was more usually seen. Conclusion All features of microcalcifications extracted by CAD mean different to different nuclei grades of DCIS.The conclusion may facilitate the management of DCIS.