中国癌症杂志 ›› 2018, Vol. 28 ›› Issue (5): 361-368.doi: 10.19401/j.cnki.1007-3639.2018.05.007

• 论著 • 上一篇    下一篇

三种不同磁共振扩散加权成像模型在鉴别乳腺良恶性病灶中的价值研究

夏冰清,黎鑫乐,孙 琨,柴维敏   

  1. 上海交通大学医学院附属瑞金医院放射科,上海 200025
  • 出版日期:2018-05-30 发布日期:2018-06-12
  • 通信作者: 柴维敏 E-mail:Chai_weimin@126.com

Different models of diffusion-weighted magnetic resonance imaging in differential diagnosis of benign and malignant breast lesions

XIA Bingqing, LI Xinle, SUN Kun, CHAI Weimin   

  1. Department of Radiology, Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai 200025, China
  • Published:2018-05-30 Online:2018-06-12
  • Contact: CHAI Weimin E-mail: Chai_weimin@126.com

摘要: 背景与目的:近年来扩散加权成像(diffusion-weighted imaging,DWI)技术的开展及应用,大大提高了磁共振(magnetic resonance,MR)的特异度,其中体素内不相干运动(intravoxel incoherent motion,IVIM)模型和扩散峰度成像(diffusion kurtosis imaging,DKI)模型作为新兴技术,已在临床研究中取得一定的进展。该研究探讨DWI的单指数模型、IVIM模型和DKI模型在乳腺良恶性病灶中的鉴别诊断价值。方法:该研究为前瞻性研究,纳入标准:超声或X线BI-RADS 4类及以上患者。排除标准:① 乳腺MR检查前已进行穿刺检查、新辅助化疗或手术的患者;② 图像运动伪影较重。所有患者术前均行双侧乳腺MR检查,扫描序列包括快速反转恢复(turbo inversion recovery magnitude,TIRM)、多b值DWI(RS-EPI)和T1W动态增强扫描。选取病灶实性成分最大层面且避开明显坏死、囊变液化区绘制感兴趣区(region of interest,ROI),分别测量单指数模型参数表观弥散系数(apparent diffusion coefficient,DADC)值、IVIM模型参数[真实扩散系数(tissue diffusivity coefficient,DDT)、灌注相关扩散系数(perfusion-related diffusivity coefficient,D*)、灌注分数(perfusion fraction,f)]和DKI模型参数[峰度系数(kurtosis coefficient,K)、扩散系数(diffusivity coefficient,DDK)]。采用独立样本t检验分别比较乳腺良恶性病灶组织上述参数的差异。采用受试者工作特征(receiver operating characteristic,ROC)曲线评价3种模型参数的诊断效能。采用Z检验比较各参数曲线下面积(area under curve,AUC)的差异。结果:依据上述标准共纳入80例患者(83个病灶),其中良性病灶38个,恶性病灶45个。3种不同扩散模型中DADC值、DDT值、K值及DDK值在鉴别乳腺良恶性病灶中差异均有统计学意义(P均<0.05),其最佳阈值分别为DADC值1.08×10-3 mm2/s、DDT值1.06×10-3 mm2/s、K值0.756及DDK值1.36×10-3 mm2/s。而D*值和f值在良恶性病灶之间存在较大重叠,差异无统计学意义(P>0.05)。ROC曲线显示,K值和DDT值在鉴别乳腺良恶性病灶的AUC值最高,分别为0.956和0.947,K值的灵敏度和特异度为91.1%和89.5%,DDT值的灵敏度和特异度为93.3%和84.2%;DADC值和DDK值其次,AUC分别为0.933和0.923,DADC值的灵敏度和特异度为88.9%和84.2%,DDK值的灵敏度和特异度为91.1%和84.2%。最后,DADC值、DDT值、K值及DDK值在鉴别乳腺良恶性病灶中的ROC曲线的AUC差异均无统计学意义(P均>0.05)。结论:三种不同扩散加权成像模型在鉴别乳腺良恶性病灶中均有较好的诊断价值,其中IVIM和DKI的诊断效能较单指数模型略高,但
三者间差异无统计学意义。单指数模型扫描时间短,后处理简单,在临床应用价值很高。

关键词: 乳腺癌, 扩散加权成像, 磁共振, 体素内不相干运动, 扩散峰度成像

Abstract: Background and purpose: In recent years, the development and application of diffusion-weighted imaging (DWI) have greatly improved the specificity of magnetic resonance (MR). The intravoxel incoherent motion (IVIM) model and the diffusion kurtosis imaging (DKI) model, as new technologies, have made some progress in clinical research. This study aimed to evaluate the diagnostic value of mono-exponential, IVIM and DKI models of DWI in characterizing benign and malignant breast lesions. Methods: Patients diagnosed as Breast Imaging Reporting and Data System (BI-RADS) category 4 or higher using mammography or ultrasonography were seleted for breast MR imaging. The exclusion criteria included: ① Patients who underwent core-needle biopsy or preoperative chemotherapy or surgery; ② Patients whose MR images had substantial motion artifacts. Turbo inversion recovery magnitude (TIRM), multi-b DWI (readout-segmented echo-planar imaging) and dynamic contrast-enhanced T1WI were performed in all patients. Region of interests (ROIs) were drawn on apparent diffusion coefficient (DADC) maps on the slice with the largest tumor area using b=50 and 1 000 s/mm2, avoiding necrotic or cystic parts. The parameters of mono-exponential (DADC), IVIM [molecular diffusion coefficient (DDT), perfusion-related diffusion coefficient (D*) and perfusion fraction (f)] and diffusion kurtosis model [diffusivity coefficient (DDK), kurtosis coefficient (K)] were measured by two radiologists. The difference in the parameters between malignant tumors and benign lesions was analyzed by independent sample t test. Receiver operating characteristic (ROC) curve was performed to compare the diagnostic value of different parameters based on the area under curve (AUC). Z test was performed to compare the difference of each AUC. Results: Eighty patients (83 lesions) were included in our study, and there were 38 breast benign lesions and 45 malignant lesions. DADC, DDT, K and DDK values were all statistically significant for the differential diagnosis of malignant and benign breast lesions (P<0.05). The optimal threshold values were DADC 1.08 ×10-3 mm2/s, DDT value 1.06×10-3 mm2/s, K value 0.756 and DDK value 1.36×10-3 mm2/s. The D* and f values between benign and malignant lesions had a large degree of overlap, and the difference was not statistically significant (P>0.05). The ROC curve area showed that the AUC of K and DDT values in differential diagnosis of benign and malignant breast lesions were the highest, 0.956 and 0.947, respectively. The sensitivity and specificity of K value were 91.1% and 89.5%. The sensitivity and specificity of DDT value were 93.3% and 84.2%. The AUC of DADC and DDK value were 0.933 and 0.923, respectively. The sensitivity and specificity of DADC value were 88.9% and 84.2%. The sensitivity and specificity of DDK value were 91.1% and 84.2%. Finally, the AUC of DADC, DDT, K and DDK values in the differential diagnosis of benign and malignant breast lesions were not statistically significant (P>0.05). Conclusion: The three models all had good performance in differential diagnosis of benign and malignant breast lesions. IVIM and DKI showed higher AUC, but the AUC had no statistically significant difference among all models. The mono-exponential model had good clinical value with the advantages of short detection time and easy postprocessing.

Key words: Breast cancer, Diffusion-weighted imaging, Magnetic resonance, Intravoxel incoherent motion, Diffusion kurtosis imaging