分享:
分享到微信朋友圈
X
临床研究
基于多模态磁共振影像组学鉴别唾液腺多形性腺瘤和基底细胞腺瘤
闫小凡 邵硕 郑宁 崔景景 苑子茵 李森

Cite this article as: Yan XF, Shao S, Zheng N, et al. Differentiating salivary gland pleomorphic adenoma from basal cell adenoma based on multimodal magnetic resonance imaging radiomics[J]. Chin J Magn Reson Imaging, 2022, 13(7): 22-28.本文引用格式:闫小凡, 邵硕, 郑宁, 等. 基于多模态磁共振影像组学鉴别唾液腺多形性腺瘤和基底细胞腺瘤[J]. 磁共振成像, 2022, 13(7): 22-28. DOI:10.12015/issn.1674-8034.2022.07.005.


[摘要] 目的 探讨基于表观弥散系数(apparent diffusion coefficient, ADC)图、T1WI及T2WI序列构建的影像组学模型鉴别唾液腺多形性腺瘤(pleomorphic adenoma, PA)和基底细胞腺瘤(basal cell adenoma, BCA)的价值。材料与方法 回顾性分析2015年1月至2021年10月来自济宁市第一人民医院的唾液腺129例PA和48例BCA患者的MR图像,并将其以8∶2的比例随机划分为训练集(n=141)与测试集(n=36)。在横断位ADC、T1WI及T2WI图像上手动勾画肿瘤的三维容积感兴趣区域,提取影像组学特征;采用方差阈值法、方差分析(analysis of variance, ANOVA)及基于5折交叉验证的最小绝对收缩与选择算法(least absolute shrinkage and selection operator, LASSO)筛选最有价值的特征,将筛选出的特征结合逻辑回归(logistic regression, LR)与支持向量机(support vector machine, SVM)两种分类器后进行模型训练,并在测试集中验证。绘制ROC曲线来评估LR模型与SVM模型鉴别PA和BCA的效能。此外,使用Delong Test对模型进行比较,使用决策曲线及校准曲线对模型进行评价。结果 分别从ADC、T1WI、T2WI及联合序列(ADC+T1WI+T2WI)图像中得到15、3、15及23个最优特征。在训练集中,基于ADC图、T1WI图、T2WI图、联合模型构建的LR与SVM模型的曲线下面积(area under the curve, AUC)分别为0.955、0.961、0.812、0.813、0.939、0.949、0.994、0.995;基于ADC、T1WI、T2WI及联合序列图像构建的LR模型鉴别诊断PA和BCA的AUC值分别为0.906、0.780、0.868及0.972,SVM模型的AUC值分别为0.924、0.783、0.847及0.959;在训练集中,基于联合序列模型优于基于T1WI或T2WI影像组学模型(P<0.05),与基于ADC影像组学模型差异无统计学意义(P>0.05),联合序列模型的准确率、敏感度及特异度分别为98.6%~98.7%、96.4%~98.4%、98.8%~99.4%,ADC影像组学模型的准确率、敏感度及特异度分别为91.4%~91.8%、75.0%~79.7%、95.7%~98.1%;在测试集中,各模型间的AUC值均无显著性差异(P>0.05)。结论 多序列联合模型及ADC影像组学模型鉴别多形性腺瘤和基底细胞腺瘤优于T1WI及T2WI序列,且与ADC影像组学模型比较,联合序列模型具有较高的准确率、敏感度及特异度。
[Abstract] Objective To explore the value of radiomics models based on ADC, T1WI and T2WI in differentiating salivary gland pleomorphic adenoma (PA) from basal cell adenoma (BCA).Materials and Methods The MR images of 129 cases with PA and 48 cases with BCA from Jining First People's Hospital from January 2015 to October 2021 were retrospectively analyzed, and then these data were randomly divided into training sets (n=141) and test sets (n=36) at a ratio of 8∶2. The three-dimensional volume region of interest of the tumor was manually delineated on the axial ADC, T1WI and T2WI images, and radiomics features were extracted; the variance threshold method, analysis of variance (ANOVA) and least absolute shrinkage and selection operator (LASSO) based on 5-fold cross validation were used to single out the most valuable radiomic features, and these selected features were combined with two classifiers, logistic regression (LR) and support vector machine (SVM), for training the models, and then the models were verified in the test sets. ROC curve was drawn to evaluate the efficacy of LR and SVM models in differentiating PA from BCA. In addition, the Delong Test was used to compare the models, and the decision curve and calibration curve were used to evaluate the models.Results A total of 15, 3, 15 and 23 optimal features were obtained from ADC, T1WI, T2WI and combined sequence (ADC+T1WI+T2WI) image respectively. In the training set, the area under the curve (AUC) of the LR and SVM models constructed based on the ADC map, T1WI map, T2WI map, and joint model were 0.955, 0.961, 0.812, 0.813, 0.939, 0.949, 0.994, 0.995, respectively. The AUC values of the LR model constructed based on ADC, T1WI, T2WI and combined sequence image for differential diagnosis of PA and BCA were 0.906, 0.780, 0.868 and 0.972, respectively, and the AUC values of the SVM model were 0.924, 0.783, 0.847 and 0.959, respectively. In the training sets, the combined sequence models were better than the T1WI or T2WI-based radiomics models (P<0.05), and there was no significant difference between the combined sequence models and the ADC-based radiomics models (P>0.05), the accuracy, sensitivity and specificity of the combined sequence models were 98.6%-98.7%, 96.4%-98.4%, 98.8%-99.4% respectively, the accuracy, sensitivity and specificity of the ADC radiomics models were 91.4%-91.8%, 75.0%-79.7%, 95.7%-98.1% respectively. In the test sets, there was no significant difference in AUC between the models (P>0.05).Conclusions The combined sequence models and ADC-based radiomics models were better than the T1WI and T2WI-based radiomics models in differentiating pleomorphic adenoma and basal cell adenoma. Compared with ADC-based radiomics models, the combined sequence models had higher accuracy, sensitivity and specificity.
[关键词] 唾液腺肿瘤;多形性腺瘤;基底细胞腺瘤;影像组学;磁共振成像
[Keywords] salivary gland tumors;pleomorphic adenoma;basal cell adenoma;radiomics;magnetic resonance imagining

闫小凡 1   邵硕 2   郑宁 2*   崔景景 3   苑子茵 2   李森 1  

1 山东第一医科大学(山东省医学科学院),济南 250000

2 济宁市第一人民医院磁共振室,济宁 272000

3 上海联影智能医疗科技有限公司,上海 200000

郑宁,E-mail:zhengning_369@163.com

作者利益冲突声明:全体作者均声明无利益冲突。


收稿日期:2022-02-05
接受日期:2022-06-22
中图分类号:R445.2  R739.8  R730.261 
文献标识码:A
DOI: 10.12015/issn.1674-8034.2022.07.005
本文引用格式:闫小凡, 邵硕, 郑宁, 等. 基于多模态磁共振影像组学鉴别唾液腺多形性腺瘤和基底细胞腺瘤[J]. 磁共振成像, 2022, 13(7): 22-28. DOI:10.12015/issn.1674-8034.2022.07.005.

       多形性腺瘤(pleomorphic adenoma, PA)是唾液腺最常见的良性肿瘤,常见于中青年人,女性好发[1];在组织学上其由上皮组织、黏液和软骨样组织构成,亦称为“混合瘤”[2]。基底细胞腺瘤(basal cell adenoma, BCA)在唾液腺良性肿瘤中位居第三位,约占唾液腺全部上皮源性肿瘤的1%~2%[3];组织学上,BCA由基底细胞样细胞排列的基底细胞层和基底膜样结构构成,缺乏黏液软骨样基质,根据细胞的生长模式可将其分为4种病理亚型:梁状型(60%)、管状型(30%)、实性型及膜性型[4];临床及影像上对该病认识不足,则易与多形性腺瘤混淆。手术切除是治疗唾液腺肿瘤最有效的方法,选择术式主要依赖于肿瘤类型及其生物学特性。多形性腺瘤易复发、会恶变,据报道其术式选择单纯瘤体切除术,复发率高达70%,故多采用根治性外科手术[5, 6, 7];而基底细胞瘤的术式及预后与PA不同[8],只采用单纯瘤体切除术或部分腺叶切除术,可达到低复发、预后好的效果。因此,术前准确定性PA和BCA对临床医生制订手术方案具有指导意义。在临床工作中,一些可观察到的常规影像征象,如密度/信号、边缘和病变形态,可能有助于唾液腺肿瘤的诊断,但有时PA和BCA的影像学表现存在部分重叠,且人工阅片具有主观差异,部分病例鉴别困难[9, 10, 11]。影像组学[12]是指从CT、MRI或正电子发射计算机断层显像(positron emission tomography, PET)等图像中高通量地挖掘人眼无法观测到的潜在影像特征,并转换为可视化数据进行量化分析,具有无创、可重复性操作等优势,近年来已用于头及颈部疾病的研究中[13],并且已有多位学者[14, 15, 16]对唾液腺良恶性肿瘤或最常见的两种良性肿瘤(多形性腺瘤和腺淋巴瘤)开展了相关纹理分析或影像组学研究,但目前尚未有在唾液腺良性肿瘤中的第三大常见类型(基底细胞腺瘤)方面的磁共振影像组学研究。因此本研究基于表观弥散系数(apparent diffusion coefficient, ADC)图、T1WI及T2WI序列构建影像组学模型,以探讨其鉴别唾液腺多形性腺瘤和基底细胞腺瘤的诊断价值。

1 材料与方法

1.1 一般资料

       本研究回顾性分析济宁市第一人民医院2015年1月至2021年10月177例经术后病理证实的唾液腺上皮源性良性肿瘤患者资料,且通过了济宁市第一人民医院伦理委员会批准,免除受试者知情同意,批准文号:2022伦审研第(005)号。

       纳入标准:(1)所有患者的ADC、T1WI及T2WI图像均在同一台MRI扫描仪上获得,且采用相同的成像参数;(2)临床信息及影像学资料完整;(3)MRI扫描前未接受任何有创性检查(穿刺、放化疗等);(4)经病理学检查证实为原发性PA或BCA。排除标准:(1)图像质量差(图像存在明显的运动伪影或图像变形等),影响数据分析;(2)病灶最大直径小于1.0 cm,难以精准勾画病灶感兴趣区。

1.2 仪器与方法

       所有ADC、T1WI及T2WI原始图像均来自于济宁市第一人民医院影像归档和通信系统(picture archiving and communication system, PACS)。采用Siemens Trio 3.0 T超高场MRI扫描仪,线圈为头部专用12通道线圈与颈部专用4通道线圈。常规MRI中T1WI扫描参数:TE 23 ms,TR 689 ms;T2WI扫描参数:TE 75 ms,TR 5280 ms。DWI扫描参数:b值选择为0 s/mm²和1000 s/mm²,TE 94 ms,TR 4200 ms,FOV 24 cm×24 cm,NEX 3,矩阵 192×173,层厚3 mm,层间距0.3 mm,总采集时间102 s。将所有病例图像以DICOM格式存入硬盘。

1.3 图像分割、特征提取与降维、模型建立

       将符合标准的图像导入ITK-SNAP开源软件(美国宾夕法尼亚大学宾州图像计算与科学实验室;http://www.ITK-SNAP.org),由2名分别具有2年、15年颌颈部影像诊断工作经验的住院医师、副主任医师在横断位ADC、T1WI及T2WI图像上沿病灶边缘逐层(应至少包含3个连续层面)手动勾画感兴趣区域(region of interest,ROI),并生成肿瘤的三维容积感兴趣区域(volume of interest,VOI)。医师A随机选择的一部分图像进行ROI勾画,2周后重复勾画该部分图像以评估观察者内部的可重复性;医师B勾画与医师A相同的图像,来评估观察者间的可重复性。组内相关系数(intraclass correlation coefficient, ICC)大于0.75则认为可重复性较好。由医师A继续勾画剩余图像。勾画ROI时应注意:(1)病灶的大小、边缘及形态等,只勾画肿瘤侵犯的区域,避开正常组织;(2)不需避开病灶内的出血和囊变坏死区。将所有VOI图像整理后导入uAI Research Portal平台(中国上海联影智能医疗科技有限公司;http://urp.united-imaging.com:8080/#/),去掉无效数据(如文本信息、空数据和无效数字等)后分别得到了2600个ADC特征、2600个T1WI特征及2600个T2WI特征信息,联合序列(ADC+T1WI+T2WI)一共提取了7800个特征。通过uAI Research Portal平台对全部特征进行选择,使用mRMR法剔除掉冗余或不相关特征。首先,采用Z-Score方法对特征进行标准化;随后,利用方差阈值法,选择0.8为阈值,筛选出方差大于0.8的特征参数;其次,以用于单变量特征选择的SelectKBest方法,用方差分析(analysis of variance,ANOVA)选出P<0.05的最佳特征;最后,在最小绝对收缩与选择算法(least absolute shrinkage and selection operator, LASSO)过程中使用5折交叉验证选出在最佳Alpha时的特征;整理筛选出的特征建立逻辑回归(logistic regression, LR)与支持向量机(support vector machine,SVM)2种机器学习模型。将病例数据以8∶2的比例随机划分为训练集(n=141)与测试集(n=36),使用训练集进行特征的选择与机器学习模型的构建,使用测试集对模型进行验证。绘制受试者工作特征曲线(receiver operating characteristic curve, ROC),并计算曲线下面积(area under the curve, AUC)、准确率、敏感度及特异度,以评估不同影像模态中LR模型与SVM模型鉴别诊断PA和BCA的效能。

1.4 统计学分析

       采用SPSS 23.0软件进行全部统计学数据分析。计量资料中符合正态分布的组间差异分析采用独立样本t检验,以(x¯±s)表示,不符合正态分布的组间差异分析采用非参数检验(Mann-Whitney U检验),以MQn)表示,计数资料的组间差异分析采用χ²检验,以频数表示;P<0.05均表明差异具有统计学意义。运用ROC曲线分析PA和BCA的ADC平均值,并得出二者的ADC平均值诊断阈值及该阈值下的AUC值、敏感度及特异度。ICC用于评价同一位医师先后两次及两位医师之间勾画病灶ROI的可重复性。绘制ROC曲线来评估模型鉴别PA和BCA的效能。此外,使用Delong Test对各模型间ROC曲线进行比较,使用决策曲线及校准曲线对模型进行评价。

2 结果

2.1 一般资料比较

       本研究包括129例多形性腺瘤[男58例,女71例,年龄10~76(45.67±15.75)岁](图1)和48例基底细胞腺瘤[男21例,女27例,年龄27~76(55.15±10.02)岁](图2),其中位于腮腺155例,下颌下腺21例,小涎腺1例。PA组和BCA组的年龄、形状及ADC平均值差异具有统计学意义(P<0.01),比较发现PA的好发年龄稍低于BCA,PA形态为浅分叶的概率高于BCA,PA的ADC平均值高于BCA;但二者间性别比例和长径的差异均无统计学意义(表1)。绘制ROC曲线,当ADC平均值诊断阈值为1.31×10-3 mm2/s时,鉴别PA和BCA的效能最优,相应AUC值为0.843,敏感度为78.3%,特异度为75%(图3)。

图1  A 女,43岁,左侧腮腺多形性腺瘤。1A:横断位T1WI示左侧腮腺椭圆形等信号影;1B:T2WI呈不均匀高信号,边界清晰,周边可见线样T2WI低信号包膜影;1C:表观弥散系数约为1.50×10-3 mm2/s;1D:镜下见肿瘤由上皮成分及较多黏液样基质构成,上皮成分由腺上皮细胞和肌上皮细胞构成,肌上皮细胞增生,细胞多呈梭形、卵圆形、星状,呈片状或弥漫分布,上皮成分无异型,核分裂少见。
Fig. 1  Female, 43 years old, left parotid gland pleomorphic adenoma. 1A: Axial T1WI shows oval equal signal shadow of left parotid gland; 1B: T2WI shows uneven high signal, clear boundary, and linear T2WI low signal envelope shadow can be seen around; 1C: The apparent diffusion coefficient (ADC) is about 1.50×10-3 mm2/s; 1D: Microscopically, the tumor is composed of epithelial components and more mucoid matrix. The epithelial components are composed of glandular epithelial cells and myoepithelial cells. Myoepithelial cells proliferate. Most of the cells are spindle, oval, star, flake or diffuse distribution. The epithelial components have no atypia and nuclear division is rare.
图2  女,49岁,左侧腮腺基底细胞腺瘤。2A:横断位T1WI示左侧腮腺椭圆形低信号影;2B:T2WI呈稍低信号,其内见点片状长T2信号,边界清晰,周围见短T2线样包膜影;2C:表观弥散系数约为1.13×10-3 mm2/s;2D:镜下见肿瘤由较单一的基底样细胞构成,大小较一致,细胞呈圆形、梭形,肿瘤团外围的细胞核深染,呈栅栏状排列,局部细胞丰富,呈片状实性生长,可见核分裂像;间质为纤维组织。
Fig. 2  Female, 49 years old, left parotid gland basal cell adenoma. 2A: Axial T1WI shows oval low signal shadow of left parotid gland; 2B: T2WI shows slightly low signal, with patchy long T2 signal, clear boundary, and short T2 linear envelope shadow can be seen around; 2C: The apparent diffusion coefficient (ADC) value is about 1.13×10-3 mm2/s; 2D: Microscopically, the tumor is composed of single basal like cells with uniform size. The cells are round and spindle shaped. The nuclei at the periphery of the tumor mass are deeply stained and arranged in a fence shape. Local cells are rich and grow in flake solid growth. Mitotic images can be seen. The stroma is fibrous tissue.
图3  唾液腺多形性腺瘤和基底细胞腺瘤的表观弥散系数(ADC)诊断阈值的受试者工作特征(ROC)曲线。当阈值为1.31×10-3 mm2/s时,诊断多形性腺瘤时的曲线下面积为0.843,敏感度为78.3%,特异度为75%。
Fig. 3  The receiver operating characteristic curve of apparent diffusion coefficient (ADC) value diagnostic threshold of salivary pleomorphic adenoma and basal cell adenoma. When the threshold is 1.31×10-3 mm2/s, the area under the curve of pleomorphic adenoma (PA) is 0.843, the sensitivity is 78.3%, and the specificity is 75%.
表1  唾液腺多形性腺瘤和基底细胞腺瘤患者的临床资料比较
Tab. 1  Comparison of clinical data between pleomorphic adenoma and basal cell adenoma of salivary gland

2.2 测量一致性

       使用一致性检验,医师A先后两次、医师A与医师B之间对病灶进行手动分割的ICC范围分别为0.771~0.933、0.759~0.906,可重复性均较好。

2.3 影像组学特征

       于uAI Research Portal平台(上海联影智能医疗科技有限公司),对从ADC、T1WI、T2WI及联合序列图像中提取的特征进行降维与筛选,最终分别得到15、3、15及23个有效特征(图4)。

图4  图像特征及相关系数。4A:表观弥散系数(ADC);4B:T1WI;4C:T2WI;4D:联合序列(ADC+T1WI+T2WI)。
Fig. 4  Image features and correlation coefficient. 4A: Apparent diffusion coefficient (ADC); 4B: T1WI; 4C: T2WI; 4D: Combined sequence (ADC+T1WI+T2WI).

2.4 影像组学预测模型

       结果显示,在训练集中,基于ADC图构建的LR与SVM模型的AUC值分别为0.955、0.961;基于T1WI图构建的LR与SVM模型的AUC值分别为0.812、0.813;基于T2WI图构建的LR与SVM模型的AUC值分别为0.939、0.949;基于联合序列构建的LR与SVM模型的AUC值分别为0.994、0.995,其中LR模型的准确率、敏感度及特异度分别为98.7%、98.4%及98.8%,SVM模型的准确率、敏感度及特异度分别为98.6%、96.4%及99.4%。在测试集中,基于ADC、T1WI、T2WI及联合序列图像构建的LR模型鉴别诊断PA和BCA的AUC值分别为0.906、0.780、0.868及0.972,SVM模型的AUC值分别为0.924、0.783、0.847及0.959;其中联合序列构建的LR模型的准确率、敏感度及特异度分别为91.6%、89.8%及92.3%,SVM模型的准确率、敏感度及特异度分别为93.3%、85.6%及96.1%(表23图5, 6, 7)。Delong Test显示:基于联合序列影像组学模型优于基于T1WI或T2WI影像组学模型(P<0.05),与基于ADC影像组学模型差异无统计学意义(P>0.05);在测试集中,各模型间的AUC值均无显著性差异(P>0.05)。

图5  分别基于表观弥散系数(ADC)、T1WI、T2WI及联合序列构建的逻辑回归与支持向量机模型的受试者工作特征(ROC)曲线。5A:ADC模型训练集;5B:ADC模型测试集;5C:T1WI模型训练集;5D:T1WI模型测试集;5E:T2WI模型训练集;5F:T2WI模型测试集;5G:联合序列模型训练集;5H:联合序列模型测试集。
Fig. 5  The receiver operating characteristic (ROC) curves of logistic regression and support vector machine models based on apparent diffusion coefficient (ADC), T1WI, T2WI and combined sequence respectively. 5A: The training sets of the ADC models; 5B: The test sets of the ADC models; 5C: The training sets of the T1WI models; 5D: The test sets of the T1WI models; 5E: The training sets of the T2WI models; 5F: The test sets of the T2WI models; 5G: The training sets of the combined sequence models; 5H: The test sets of the combined sequence models.
图6  分别基于表观弥散系数(ADC)、T1WI、T2WI及联合序列构建的逻辑回归与支持向量机模型的决策曲线。6A:ADC模型训练集;6B:ADC模型测试集;6C:T1WI模型训练集;6D:T1WI模型测试集;6E:T2WI模型训练集;6F:T2WI模型测试集;6G:联合序列模型训练集;6H:联合序列模型测试集。
Fig. 6  The decision curves of logistic regression and support vector machine models based on apparent diffusion coefficient (ADC), T1WI, T2WI and combined sequence respectively. 6A: The training sets of the ADC models; 6B: The test sets of the ADC models; 6C: The training sets of the T1WI models; 6D: The test sets of the T1WI models; 6E: The training sets of the T2WI models; 6F: The test sets of the T2WI models; 6G: The training sets of the combined sequence models; 6H: The test sets of the combined sequence models.
图7  分别基于表观弥散系数(ADC)、T1WI、T2WI及联合序列构建的逻辑回归与支持向量机模型的校准曲线。7A:ADC模型训练集;7B:ADC模型测试集;7C:T1WI模型训练集;7D:T1WI模型测试集;7E:T2WI模型训练集;7F:T2WI模型测试集;7G:联合序列模型训练集;7H:联合序列模型测试集。
Fig. 7  The calibration curves of logistic regression and support vector machine models based on apparent diffusion coefficient (ADC), T1WI, T2WI and combined sequence respectively. 7A: The training sets of the ADC models; 7B: The test sets of the ADC models; 7C: The training sets of the T1WI models; 7D: The test sets of the T1WI models; 7E: The training sets of the T2WI models; 7F: The test sets of the T2WI models; 7G: The training sets of the combined sequence models; 7H: The test sets of the combined sequence models.
表2  逻辑回归模型在训练集与测试集中的诊断效能
Tab. 2  Diagnostic efficiency of logistic regression model in the training and test sets
表3  支持向量机模型在训练集与测试集中的诊断效能
Tab. 3  Diagnostic efficiency of support vector machine model in the training and test sets

3 讨论

       本研究旨在探讨多模态磁共振影像组学模型在唾液腺基底细胞腺瘤中的应用价值。结果显示,在训练集中,联合序列模型的诊断效能优于T1WI或T2WI模型;而联合序列影像组学模型与ADC影像组学模型的AUC间的差异无统计学意义,但联合序列模型的准确率、敏感度及特异度均较高于ADC影像组学模型,提示多序列图像联合能提供更多肿瘤内部特征。在测试集中,由于样本量较少,联合序列影像组学模型与基于ADC、T1WI或T2WI影像组学模型的AUC之间的差异无统计学意义,但各影像组学模型在鉴别PA与BCA时均具有较好的诊断价值。

3.1 影像组学模型的诊断效能

       在最近的几项研究中,郑韵琳等[17]基于多期CT增强构建影像组学模型鉴别腮腺PA和BCA,得出动脉期的影像组学模型及与临床资料建立的联合模型的AUC值为0.968~0.973,与本研究联合序列模型及ADC影像组学模型诊断效能相似。与郑韵琳等增强CT研究相比,本研究使用无电离辐射、多参数和软组织分辨率更高的磁共振检查,多参数成像能提供更多的肿瘤内部特征,且不需要使用增强对比剂,避免对比剂过敏风险。彭媛媛等[18]对腮腺多形性腺瘤和腺淋巴瘤患者的增强T1WI图像进行定量纹理分析,使用费希尔参数法+最小分类误差与最小相关系数法+协同信息法联合法,筛选出30个对鉴别诊断有显著效能的纹理参数,其中P50、P90、WavEnLL_s-2及WavEnLL_s-3的鉴别诊断能力最佳,相应的AUC值分别为0.858、0.864、0.901及0.905;吴艳等[19]基于99例腮腺肿瘤患者的T2WI图像上的最大层面去勾画二维ROI提取特征,采用两种降维方法(最小冗余最大相关算法和LASSO回归分析)先后对特征进行处理,最终筛选出8个最佳特征并建立影像组学标签,将临床资料与影像组学标签联合构建LR模型以鉴别腮腺多形性腺瘤和腺淋巴瘤,最终联合模型在训练集和验证集的AUC值分别为0.90、0.96。与以上研究相比,本研究是在多序列图像上勾画肿瘤三维ROI来构建影像组学模型且将组学应用于鉴别诊断PA和BCA,比纹理分析或单序列模型所提取的特征数量及类型更多,可以对图像信息相互补充,从而更全面地反映出肿瘤内部的异质性[20]。以上研究均证实影像组学模型鉴别诊断唾液腺良性肿瘤的可靠性及价值。

3.2 影像组学特征的客观性

       本研究收集的图像均来自于同一台MRI扫描仪,先进行图像预处理以降低采集误差与个体差异的影响,手动勾画肿瘤区域的三维ROI并进行特征的提取,最终筛选出4类特征参数,即:一阶统计学特征,形态学特征,纹理特征中的灰度共生矩阵(gray level cooccurence matrix, GLCM)、灰度游程矩阵(gray level run length matrix, GLRLM)、灰度区域大小矩阵(gray level size zone matrix, GLSZM)、灰度相关矩阵(gray level dependence matrix, GLDM)、邻域灰度差分矩阵(neighbouring gray tone difference matrix, NGTDM),以及小波特征。既往已有研究[21, 22, 23]表明图像灰度数据信息能定量分析肿瘤内部的异质性,由此表明影像组学模型在肿瘤预测方面的客观性。

3.3 机器学习算法的选择

       近年来,机器学习分类器模型被大量引入,合适的机器学习算法能够提升模型的预测效能及稳定性。有学者推断[24]逻辑回归与支持向量机分类器适用于小数据集分析,构建的模型均较为稳定,本研究构建的正是小样本模型,结果显示测试集与训练集的拟合度较佳,且LR模型与SVM模型鉴别诊断唾液腺PA和BCA时均具有良好的性能,诊断价值无显著差异(P>0.05)。既往研究报道了两分类器在临床其他方向的应用,有学者推测LR分类器的效能低于SVM,也有学者推测两者效能具有可比性[25, 26]

       此外,ADC值稍低提示为基底细胞腺瘤。本研究得出的鉴别PA和BCA的ADC值诊断阈值(1.31×10-3 mm2/s),AUC值为0.843,与Mukai等[27]报道的结果(1.31×10-3 mm2/s)一致;且基于ADC图构建的影像组学模型与ADC平均值鉴别两者的诊断效能均较好(P>0.05)。

3.4 局限性

       本研究仍存在的局限性:(1)本研究为单中心性研究,模型缺乏外部验证,存在过拟合及稳定性等问题,且对于影像组学相关研究,病例数仍相对较少且分布欠均衡,可进一步扩充样本量,并进行多中心性研究;(2)本研究对影像组学特征的提取仅基于ADC、T1WI及T2WI序列,后续将进一步结合临床资料构建模型,观察其诊断价值。

       综上所述,基于ADC、T1WI及T2WI序列构建影像组学模型对唾液腺PA和BCA的鉴别诊断提供了一种潜在方法,有助于术前准确定性。

[1]
Almeslet AS. Pleomorphic adenoma: a systematic review[J]. Int J Clin Pediatr Dent, 2020, 13(3): 284-287. DOI: 10.5005/jp-journals-10005-1776.
[2]
王海滨, 郇通, 韩志江, 等. 腮腺多形性腺瘤常规MR征象与病理相关性分析[J]. 中国临床医学影像杂志, 2020, 31(11): 783-787. DOI: 10.12117/jccmi.2020.11.006.
Wang HB, Xun T, Han ZJ, et al. Analysis of correlation between conventional MR signs and pathology of parotid pleomorphic adenoma[J]. J China Clin Med Imaging, 2020, 31(11): 783-787. DOI: 10.12117/jccmi.2020.11.006.
[3]
Nagao K, Matsuzaki O, Saiga H, et al. Histopathologic studies of basal cell adenoma of the parotid gland[J/OL]. Cancer, 1982, 50(4): 736-745 [2022-02-05]. https://doi.org/10.1002/1097-0142(19820815)50:4<736::AID-CNCR2820500419>3.0.CO;2-M.
[4]
Rajasenthil V, Ravi A. Basal cell adenoma of the parotid gland-A rare entity[J]. Indian J Surg, 2014, 76(5): 408-410. DOI: 10.1007/s12262-013-1010-5.
[5]
Witt RL, Nicolai P. Recurrent benign salivary gland neoplasms[J]. Adv Otorhinolaryngol, 2016, 78: 63-70. DOI: 10.1159/000442126.
[6]
Valstar MH, Mast H, Ten Hove I, et al. Malignant transformation of salivary gland pleomorphic adenoma: proof of principle[J]. J Pathol Clin Res, 2021, 7(5): 432-437. DOI: 10.1002/cjp2.216.
[7]
Espinosa CA, Fernández-Valle Á, Lequerica-Fernández P, et al. Clinicopathologic and surgical study of pleomorphic adenoma of the parotid gland: analysis of risk factors for recurrence and facial nerve dysfunction[J]. J Oral Maxillofac Surg, 2018, 76(2): 347-354. DOI: 10.1016/j.joms.2017.07.171.
[8]
Psychogios G, Bohr C, Constantinidis J, et al. Review of surgical techniques and guide for decision making in the treatment of benign parotid tumors[J]. Eur Arch Otorhinolaryngol, 2021, 278(1): 15-29. DOI: 10.1007/s00405-020-06250-x.
[9]
程景风, 郑宁, 邵硕, 等. 3.0T多参数MRI在鉴别腮腺基底细胞腺瘤和多形性腺瘤中的诊断价值[J]. 实用放射学杂志, 2020, 36(4): 550-553. DOI: 10.3969/j.issn.1002-1671.2020.04.011.
Cheng JF, Zheng N, Shao S, et al. Diagnostic value of 3.0T multiparametric MRI for discriminating parotid basal cell adenoma from pleomorphic adenoma[J]. J Pract Radiol, 2020, 36(4): 550-553. DOI: 10.3969/j.issn.1002-1671.2020.04.011.
[10]
Chiu NC, Wu HM, Chou YH, et al. Basal cell adenoma versus pleomorphic adenoma of the parotid gland: CT findings[J]. AJR Am J Roentgenol, 2007, 189(5): W254-W261. DOI: 10.2214/AJR.07.2292.
[11]
廖建勇, 杜静波, 刘迎新, 等. 腮腺基底细胞腺瘤和多形性腺瘤的双期增强CT的诊断价值[J]. 武警医学, 2020, 31(5): 382-385. DOI: 10.14010/j.cnki.wjyx.2020.05.005.
Liao JY, Du JB, Liu YX, et al. The diagnostic value of dual-phase CT enhancement appearances of basal cell adenoma and pleomorphic adenoma in the parotid gland[J]. Med J Chin People's Armed Police Force, 2020, 31(5): 382-385. DOI: 10.14010/j.cnki.wjyx.2020.05.005.
[12]
Lambin P, Leijenaar RTH, Deist TM, et al. Radiomics: the bridge between medical imaging and personalized medicine[J]. Nat Rev Clin Oncol, 2017, 14(12): 749-762. DOI: 10.1038/nrclinonc.2017.141.
[13]
王安然, 李泉江, 黄钟馨, 等. 影像组学在头及颈部疾病诊疗中的应用进展[J]. 磁共振成像, 2021, 12(1): 100-102. DOI: 10.12015/issn.1674-8034.2021.01.023.
Wang AR, Li QJ, Huang ZX, et al. Progress in the application of radiomics in head and neck diseases[J]. Chin J Magn Reson Imaging, 2021, 12(1): 100-102. DOI: 10.12015/issn.1674-8034.2021.01.023.
[14]
Gabelloni M, Faggioni L, Attanasio S, et al. Can magnetic resonance radiomics analysis discriminate parotid gland tumors? A pilot study[J/OL]. Diagnostics (Basel), 2020, 10(11): E900 [2022-02-05]. https://www.mdpi.com/2075-4418/10/11/900. DOI: 10.3390/diagnostics10110900.
[15]
余先超, 孙宇凤, 李鹏, 等. 影像组学在腮腺多形性腺瘤与腺淋巴瘤鉴别诊断中的应用[J]. 现代肿瘤医学, 2021, 29(5): 837-840. DOI: 10.3969/j.issn.1672-4992.2021.05.025.
Yu XC, Sun YF, Li P, et al. Application of radiomics in the differential diagnosis of parotid pleomorphic adenoma and adenolymphoma of the parotid gland[J]. J Mod Oncol, 2021, 29(5): 837-840. DOI: 10.3969/j.issn.1672-4992.2021.05.025.
[16]
周宇堃, 甄俊平, 靳波, 等. 基于T1WI及IDEAL-T2WI影像组学模型鉴别腮腺多形性腺瘤和腺淋巴瘤[J]. 中国医学影像技术, 2020, 36(5): 675-679. DOI: 10.13929/j.issn.1003-3289.2020.05.008.
Zhou YK, Zhen JP, Jin B, et al. Differentiating parotid gland pleomorphic adenoma from adenolymphoma based on T1WI and IDEAL-T2WI radiomics models[J]. Chin J Med Imaging Technol, 2020, 36(5): 675-679. DOI: 10.13929/j.issn.1003-3289.2020.05.008.
[17]
郑韵琳, 刘欢, 文明. 基于多期CT影像组学模型及联合模型鉴别腮腺多形性腺瘤与基底细胞瘤[J]. 中国医学影像学杂志, 2022, 30(3): 220-227. DOI: 10.3969/j.issn.1005-5185.2022.03.006.
Zheng YL, Liu H, Wen M. Radiomics analysis based on multi-phase CT and combined model differentiated pleomorphic adenoma from basal cell adenoma[J]. Chin J Med Imaging, 2022, 30(3): 220-227. DOI: 10.3969/j.issn.1005-5185.2022.03.006.
[18]
彭媛媛, 任翠萍, 程敬亮, 等. 增强T1WI纹理分析对腮腺多形性腺瘤与腺淋巴瘤的鉴别诊断[J]. 中国医学影像学杂志, 2021, 29(3): 207-211. DOI: 10.3969/j.issn.1005-5185.2021.03.004.
Peng YY, Ren CP, Cheng JL, et al. Enhanced-T1WI texture analysis for differential diagnosis of parotid pleomorphic adenoma and adenoma lymphoma[J]. Chin J Med Imaging, 2021, 29(3): 207-211. DOI: 10.3969/j.issn.1005-5185.2021.03.004.
[19]
吴艳, 谢元亮, 张树桐, 等. 基于T2WI影像组学及联合诊断模型鉴别腮腺多形性腺瘤与腺淋巴瘤[J]. 放射学实践, 2020, 35(12): 1525-1531. DOI: 10.13609/j.cnki.1000-0313.2020.12.006.
Wu Y, Xie YL, Zhang ST, et al. Radiomic analysis based on MR-T2WI and combined diagnostic model for identification of pleomorphic adenoma from adenolymphoma of parotid gland[J]. Radiol Pract, 2020, 35(12): 1525-1531. DOI: 10.13609/j.cnki.1000-0313.2020.12.006.
[20]
董诗洁, 胡晓欣, 王葳, 等. 基于多序列MRI与多体系影像组学模型预测子宫颈癌淋巴结转移的研究[J]. 中国癌症杂志, 2021, 31(6): 460-467. DOI: 10.19401/j.cnki.1007-3639.2021.06.004.
Dong SJ, Hu XX, Wang W, et al. Prediction of lymph node metastasis of cervical cancer based on multi-sequence MRI and multi-system imaging omics model[J]. China Oncol, 2021, 31(6): 460-467. DOI: 10.19401/j.cnki.1007-3639.2021.06.004.
[21]
Rizzo S, Botta F, Raimondi S, et al. Radiomics: the facts and the challenges of image analysis[J]. Eur Radiol Exp, 2018, 2(1): 36. DOI: 10.1186/s41747-018-0068-z.
[22]
石博文, 叶靖, 段绍峰, 等. T2WI序列纹理分析联合机器学习的影像组学术前预测肾透明细胞癌的病理分级[J]. 中国医学影像学杂志, 2020, 28(8): 629-632. DOI: 10.3969/j.issn.1005-5185.2020.08.018.
Shi BW, Ye J, Duan SF, et al. Radiomics based on T2WI sequence texture analysis combined with machine learning imaging in predicting the pathological grade of clear cell renal cell carcinoma before surgery[J]. Chin J Med Imaging, 2020, 28(8): 629-632. DOI: 10.3969/j.issn.1005-5185.2020.08.018.
[23]
胡云涛, 黄小华, 刘念, 等. 基于磁共振T2WI序列影像组学预测急性胰腺炎复发的价值[J]. 磁共振成像, 2021, 12(10): 12-15, 21. DOI: 10.12015/issn.1674-8034.2021.10.003.
Hu YT, Huang XH, Liu N, et al. The value of T2WI sequence-based radiomics in predicting recurrence of acute pancreatitis[J]. Chin J Magn Reson Imaging, 2021, 12(10): 12-15, 21. DOI: 10.12015/issn.1674-8034.2021.10.003.
[24]
Liang M, Cai ZT, Zhang HM, et al. Machine learning-based analysis of rectal cancer MRI radiomics for prediction of metachronous liver metastasis[J]. Acad Radiol, 2019, 26(11): 1495-1504. DOI: 10.1016/j.acra.2018.12.019.
[25]
Zhou ZG, Folkert M, Cannon N, et al. Predicting distant failure in early stage NSCLC treated with SBRT using clinical parameters[J]. Radiother Oncol, 2016, 119(3): 501-504. DOI: 10.1016/j.radonc.2016.04.029.
[26]
Wu HY, Gong CA, Lin SP, et al. Predicting postoperative vomiting among orthopedic patients receiving patient-controlled epidural analgesia using SVM and LR[J/OL]. Sci Rep, 2016 [2022-02-05]. https://wwwnature.53yu.com/articles/srep27041. DOI: 10.1038/srep27041.
[27]
Mukai H, Motoori K, Horikoshi T, et al. Basal cell adenoma of the parotid gland; MR features and differentiation from pleomorphic adenoma[J/OL]. Dentomaxillofac Radiol, 2016, 45(4) [2022-02-05]. https://www.birpublications.org/doi/full/10.1259/dmfr.20150322. DOI: 10.1259/dmfr.20150322.

上一篇 MRI直方图在胚胎发育不良性神经上皮瘤和弥漫性星形细胞瘤鉴别诊断中的价值
下一篇 合成MRI定量参数图与ADC图直方图特征对乳腺癌Ki-67表达状态的预测研究
  
诚聘英才 | 广告合作 | 免责声明 | 版权声明
联系电话:010-67113815
京ICP备19028836号-2