基于TIRM序列不同区域的GLCM纹理特征对乳腺肿块的诊断价值  

The Diagnostic Value of Gray Level Co-occurrence Matrix Texture Features Based on Different ROIs of TIRM Sequence in Breast Masses

作　　者：钱丽[1] 李通 唐彩银 QIAN Li;LI Tong;TANG Cai-yin(Department of Radiology,The Affiliated Taizhou People’s Hospital of Nanjing Medical University,Taizhou 225300,Jiangsu Province,China)

机构地区：[1]南京医科大学附属泰州人民医院影像科,江苏泰州225300

出　　处：《中国CT和MRI杂志》2023年第9期101-103,共3页Chinese Journal of CT and MRI

基　　金：泰州市社会发展项目(SSF20210111)。

摘　　要：目的探究基于磁共振T2WI反转恢复压脂(TIRM)序列乳腺肿块中央区、边缘区和整体区三个不同区域的灰度共生矩阵(GLCM)纹理特征鉴别乳腺良恶性肿块的诊断价值。方法回顾性收集本院2016年5月至2021年4月经手术或穿刺病理证实并行术前乳腺MRI检查的88例乳腺肿瘤患者资料,包括良性病灶48个、恶性病灶50个。将病灶于TIRM序列中的最大层面图像导入MaZda软件,分别在病灶整体、中央和边缘勾画感兴趣区(ROI)并提取GLCM纹理参数,比较两组间各纹理参数的差异并评估其诊断效能。进一步分别联合三个不同ROI有统计学差异的纹理参数并建立二元Logistic回归诊断模型,评价其鉴别乳腺良恶性肿块的诊断效能,寻找最佳感兴趣区。结果两组间差异有统计学意义(P<0.05)的纹理参数共计20个,其中整休区6个,边缘区7个,中央区7个。良性组中三个不同ROI的和方差、和熵、熵、差熵均低于恶性组,而角二阶矩均高于恶性组。其中以边缘区的差熵诊断效能最高,AUC为0.783,敏感度和特异度分别为76.0%、79.2%。两组边缘区的联合诊断模型效能最优,AUC为0.840,敏感度和特异度分别为82.0%和81.3%。结论基于TIRM序列乳腺肿块边缘区的GLCM纹理特征对鉴别乳腺良恶性病变诊断价值较高,为临床诊断提供了新的思路。Objective To investigate the value of gray level co-occurrence matrix(GLCM)texture features based on different regions of interest on T2WI turbo inversion recovery magnitude(TIRM)sequence in differentiating benign and malignant breast masses.Methods A total of 88 patients(98 lesions)with breast lesions who confirmed by surgical or biopsy pathology and underwent preoperative MRI were retrospectively collected from our Hospital from May 2016 to April 2021,including 48 benign lesions and 50 malignant ones.The maximum cross-sectional image of lesion in the TIRM sequence was imported into the MaZda software,and each lesion was divided into three different regions of interest(ROI):the whole,core and edge part.Then we extracted the GLCM parameters of the three ROIs,compared statistical significance of the parameters in different groups and evaluated the diagnostic efficacy of each parameter in differentiating benign and malignant lesions.Furthermore,according to the statistically significant texture parameters of each region,three multi-parameters Logistic regression diagnosis models based on different regions of interest were established,and the diagnostic efficacy of the best diagnosis model was evaluated.Results There were 20 texture features with statistically significant difference between the two groups(P<0.05).Among them,there were 6 features from the whole part,7 features from the edge part and 7 features from the core part.The sum of variance(SumVarnc),sum entropy(SumEntrp),entropy and difference entropy(DifEntrp)of the benign group were lower than those of the malignant group,while angular second moment(AngScMom)was higher than that of the malignant group.Among the texture parameters,the AUC of difference entropy in the edge part was the largest(0.783)with sensitivity 76.0%and specificity 79.2%.Among the three different diagnosis models,the model based on the edge area had the best diagnostic performance,whose AUC was 0.840,sensitivity was 82.0%and specificity was 81.3%.Conclusion The GLCM texture features of

关 键 词：乳腺肿瘤 磁共振成像 纹理分析 感兴趣区 

分 类 号：R737.9[医药卫生—肿瘤] R445.2[医药卫生—临床医学]