基于Tirm序列联合直方图及纹理特征分析鉴别乳腺良恶性病变  

Differential Diagnosis of Benign and Malignant Breast Lesions by Texture Feature Combined with Histogram Parameter Based on Tirm Sequence

作　　者：李晓峰[1] 刘彩云 石岩[1] 罗小虎[1] 王刚[1] 姚标[1] LI Xiaofeng;LIU Caiyun;SHI Yan;LUO Xiaohu;WANG Gang;YAO Biao(Department of Radiology,Xuzhou Cancer Hospital,Xuzhou 221004,China)

机构地区：[1]徐州市肿瘤医院放射科,徐州221004

出　　处：《肿瘤防治研究》2018年第12期1009-1013,共5页Cancer Research on Prevention and Treatment

基　　金：徐州市科技局社会发展项目(KC16SL105)

摘　　要：目的分析乳腺良恶性病变在T2反转恢复压脂序列上的直方图及纹理特征,并评价诊断乳腺良恶性病变的效能。方法回顾性分析100例行MRI检查并经病理证实的乳腺肿瘤资料。纹理特征提取采用灰度共生矩阵法得到病灶能量值、熵值、惯量值、自相关性、逆差矩及群显著性;直方图分析得到平均值、偏度、峰度。分别评价不同参数值在乳腺良恶性病变中的诊断效能。构建受试者工作特征曲线(ROC)结合临床实际评价乳腺良恶性组间的差异,确定诊断临界值。结果各参数在良恶性组间比较,自相关性、群显著性、平均值及峰度值差异均有统计学意义(均P<0.001);能量值、熵值、惯量值、逆差距及偏度值差异无统计学意义(P>0.05)。构建ROC将自相关性、群显著性、平均值及峰度值分别用于诊断乳腺良恶性病变,均有统计学意义(P<0.001),联合四组参数诊断良恶性病变,曲线下面积为0.868,敏感度为90.57%,特异性为72.34%。结论基于Tirm序列的直方图及纹理分析可用于乳腺良恶性病变的鉴别诊断,自相关性、群显著性、平均值及峰度值在Tirm序列上对诊断乳腺良恶性病变有一定意义,联合诊断可提高诊断效能。Objective To explore histogram parameters and texture features of benign and malignant breast lesion on turbo inversion recovery magnitude(Tirm)sequence,and evaluate which parameter could help best differentiate benign from malignant breast lesion.Methods This retrospective study included 100 breast cancer patients who underwent conventional MRI and confirmed pathologically.Texture features were derived from the gray level co-occurrence matrix(GLCM),and entropy,energy,correlation,inertia,inverse difference moment,cluster prominence and mean value,skewness,kurtosis of histogram parameters were calculated.Then we assessed the diagnosis efficacy with these parameters among the variety kinds of benign and malignant breast lesions respectively,and establish the receiver-operating characteristic curve(ROC).We assessed the differences between benign and malignant groups by the Yoden index combined with clinic for the cut-off values.Results The differences of correlation,cluster prominence,mean value and kurtosis were statistically significant between the benign and malignant groups(all P<0.001);the differences of energy,entropy,inertia,inverse difference moment and skewness were not statistically significant(all P>0.05).The results of ROC with correlation,cluster prominence,mean value and kurtosis on the diagnosis of benign and malignant breast lesions were statistically significant,respectively(P<0.001).Combined the four parameters on diagnosis benign and malignant lesions,the AUC was 0.868,the sensitivity was 90.57%and the specificity was 72.34%.Conclusion The histogram analysis and texture analysis based on Tirm sequence could be used for the differential diagnosis of benign and malignant breast lesions.Correlation,cluster prominence,mean value and kurtosis have certain significance in the diagnosis.The combined diagnosis could improve the differential ability of benign and malignant breast lesions.

关 键 词：磁共振成像 乳腺肿瘤 直方图分析 纹理分析 鉴别诊断 

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]