常规MRI纹理分析对腰椎多发性骨髓瘤与腰椎退行性病变的鉴别诊断价值.pdf

中国中西医结合影像学杂志2 0 2 3年9 月第2 1卷第5期骨骼肌肉影像学常规MRI纹理分析对腰椎多发性骨髓瘤与腰椎退行性病变的鉴别诊断价值朱心雨，郭阝立，黄鹏，马金宇，张雨柔昆明医科大学第二附属医院放射科，云南昆明6 50 10 1摘要】目的：探究常规MRI纹理分析对腰椎多发性骨髓瘤（MM)与腰椎退行性病变的鉴别诊断价值。方法：对6 4例腰椎MM患者和6 4例腰椎退行性病变患者进行分析。利用MaZda软件在T,WI、T,W I 和T2-STIR图像上手动勾画ROI并提取特征参数，用费希尔算法（Fisher）、分类错误率+平均相关系数算法（POE+ACC）、交互信息算法（MI)3种方法对提取的特征参数进行降维筛选，每种方法筛选出10 个最优纹理特征参数。对30 个最优纹理特征参数进行统计分析，利用ROC曲线评价各参数的诊断效能。结果：T,WI序列的9 个参数具有统计学意义,T,WI序列的8 个参数具有统计学意义,T,-STIR序列的14个参数具有统计学意义。排除允余参数后logistic回归分析显示，T,WI序列的Perc.10%，T z-ST IR序列的Perc.10%和135dr_GLevNonU是鉴别腰椎MM与退行性病变的独立预测因子。ROC曲线显示,3个纹理参数联合鉴别诊断腰椎MM与退行性病变的AUC为0.9 0 2,高于单一参数,差异均有统计学意义（均P0.05)。结论：常规MRI纹理分析有助于鉴别腰椎MM与腰椎退行性病变。【关键词腰椎多发性骨髓瘤；腰椎退行性病变；磁共振成像；纹理分析Value of routine MRI texture analysis for differential diagnosis of lumbar multiple myeloma and degeneration ZHUXinyu,GUO Li,HUANG Peng,MA Jinyu,ZHANG Yurou.Department of Radiology,Second Affiliated Hospital of Kunming MedicalUniversity,Kunming 650101,China.Abstract Objective:To investigate the value of conventional MRI texture analysis in dfferentiating lumbar multiplemyeloma(MM)and degeneration.Methods:Sixty-four patients with MM and 64 patients with lumbar degeneration werecollected and analyzed.ROIs were manually sketched and feature parameters were extracted from the TWI,T,WI and T,-STIRsequences using MaZda software.The extracted feature parameters were filtered by Fisher,POE+ACC and MI,and 10 optimalfeature parameters were filtered by each method.Statistical significance analysis was performed on the 30 optimal texturefeature parameters,and ROC curves were used to evaluate the diagnostic efficiency of each method.Results:Nine parameters onTWI,8 parameters on TWI and 14 parameters on T2-STIR were statistically significant.After excluding redundant parameters,logistic regression analysis showed that Perc.10%on T,WI,Perc.10%and 135dr_GLevNonU on T2-STIR were the independentpredictors for differentiating MM from degeneration.ROC curve showed that the AUC of the combined texture parameters for thedifferential diagnosis of MM and degeneration was 0.902,which was higher than that of the single-parameter,and the differenceswere statistically significant(all P0.8),然2后行logistic回归分析。采用ROC曲线分析差异有统计学意义的参数及其联合鉴别MM与腰椎退行性病变的诊断效能。利用DeLong 检验比较联合诊断与单一参数AUC的差异。以P0.05为差异有统计学意义。中国中西医结合影像学杂志2 0 2 3年9 月第2 1卷第5期2结果2.12组一般资料比较MM组6 4例,其中男34例,女30 例；年龄4579岁,平均(6 110)岁。对照组6 4例,其中男31例，女33例；年龄2 2 7 7 岁，平均(58 11)岁。2 组年龄表1T,WI纹理特征参数在2 组中的比较(x土s)纹理参数MM 组(n=64)Perc.10%99.76624.703 134.166 28.513Perc.01%97.020 26.987 137.00049.988Perc.50%138.26666.537 179.433 44.238Mean181.99366.388 221.59944.225S(2,0)SumAverg11.780 4.111S(1,0)SumAverg11.772 4.10614.5564.8880.020S(3,0)SumAverg11.786 4.11514.5794.9090.020S(0,1)SumAverg11.781 4.12114.5524.8730.021S(0,2)SumAverg11.793 4.12114.560 4.8750.021注：MM为多发性骨髓瘤。表3T,-STIR纹理特征参数在2 组中的比较(x土s）纹理参数MM 组(n=64)Perc.50%71.923 17.456Mean73.904 18.345Perc.10%53.01515.143Perc.01%39.476 13.952Perc.90%97.538 227.074Perc.99%125.430 36.032WavEnLL_s-1924.657685.380WavEnLL_s-2859.818 663.274WavEnLL_s-3835.453592.081135dr_GLevNonU581.842 304.796Horzl_GLevNonU483.674 251.311S(3,-3)Correlat0.3820.19445dr_GLevNonU578.831 306.606S(1,-1)SumAverg14.967 5.615注：MM为多发性骨髓瘤。2.3logistic回归分析结果排除允余参数后对T,WI序列的Perc.10%、S(2,0)SumAverg,T,WI序列的Perc.10%、W a v En LL_s-2 和T2-STIR序列的Perc.10%、Pe r c.9 9%、135d r-G Le v No n U、S(3,-3)Correlat 和 S(1,-1)SumAverg 共9 个纹理参数行logistic回归分析（表4),结果表明T,WI序列的Perc.10%,T2-STIR序列的 Perc.10%和 135dr_GLevNonU为鉴别MM与退行性病变的独立预测因子。2.4单参数与参数联合的诊断效能比较以T,WI序列的Perc.10%,T2-STIR序列的Perc.10%和135dr_GLevNonU作为独立检验变量及三者555.及性别比较,差异均无统计学意义(t=1.451,P=0.149;X2=0.281,P=0.596)。2.22组各序列上纹理特征参数比较经统计学分析后具有鉴别2 种疾病意义的各序列参数比较见表1 3。表2 T,WI纹理特征参数在2 组中的比较(xs)对照组(n=64)P值0.0010.0010.0070.00914.571 4.9020.020纹理参数Perc.10%110.033 39.176Perc.01%86.900 32.591S(3,0)Correlat0.519 0.171S(4,0)Correlat0.438 0.173WavEnLL_s-2621.106 75.452S(3,0)SumAverg9.889 3.261S(2,0)SumAverg9.891 3.260Mean151.197 53.432注：MM为多发性骨髓瘤。对照组(n=64)52.23811.79753.39412.95237.460 9.53526.5398.19470.968 20.88890.809 30.732519.358324.638504.043 312.164503.380300.605839.808422.689696.495 351.5770.2980.151832.153 422.44111.714 3.644联合检验变量行ROC曲线分析,3个参数鉴别诊断MM与腰椎退行性变的AUC分别为0.7 53、0.8 13和0.705，联合参数的AUC为0.9 0 2，优于单参数，差异有统计学意义（均P0.05；表5,图3)。3讨论纹理分析可从图像像素、灰阶、空间特征等多角度多方法提取大量人眼难以观察或易忽略的纹理参数，利用数学算法进行分析计算，经过降维筛选后的特征能最大程度展现影像图像中异质性微观差异,有助于肿瘤的诊断、鉴别、预测及预后评估等1415。MM 组(n=64)对照组(n=64)125.033 28.098106.000 24.7950.609 0.1010.531 0.110475.475 37.16011.697 3.60111.690 3.594175.016 32.764P值0.0010.0010.0010.0010.0010.0010.0010.0010.0020.0010.0010.0010.0010.001P值0.0060.0130.0160.0170.0200.0460.0470.042556.Perc.10%(T2-STIR)Perc.99%(T2-STIR)135dr_GLevNonU(T,-STIR)S(3,-3)Correlat(T,-STIR)S(1,-1)SumAverg(T,-STIR)Perc.10%(T,WI)S(2,0)SumAverg(T,WI)Perc.10%(T,WI)WavEnLL_s-2(T,WI)注：OR为比值比。表5单参数与参数联合的诊断效能比较(DeLong检验）比较方式P值参数联合vs.Perc.10%(T,WI)3.009参数联合vs.Perc.10%(T,-STIR)2.933参数联合vs.135dr_GLevNonU(T,-STIR)3.9960.00110080(%)酒薄6040200图 3Perc.10%(T,WI),Perc.10%(T-STIR),135dr_GLevNonU(Tz-STIR)和参数联合的ROC曲线目前,纹理分析广泛用于肿瘤诊疗中。Reinert 等16 分析了双能量CT图像在MM中的定性作用,并使用纹理分析方法评估椎体髓质受累程度，结果显示骨髓浸润程度与一阶特征呈正相关，与二阶共生矩阵特征呈负相关，骨髓图像纹理特征的变化与骨髓瘤特异性血液学参数有很好的相关性。Xiong等131利用T,WI和T,WI图像提取腰椎MM和不同腰椎转移肿瘤的纹理特征,结合不同的机器学习模型探究其鉴别诊断MM与腰椎转移肿瘤病变的潜力,结果显示，由TWI图像7 个特征构建的人工神经网络分类器在鉴别MM和不同亚型转移瘤诊断性能最高。中国中西医结合影像学杂志2 0 2 3年9 月第2 1卷第5期表4logistic回归分析结果变量1.136(1.051 0.998(0.965 0.997(0.96520.473(0.1253357.205)0.987(0.851 1.146)0.968(0.9400.997)1.057(0.873 1.280)1.000(0.9701.031)0.004(0.000103.556)本研究利用MaZda纹理分析软件提取MM患者Z值与腰椎退行性变患者MRI图像6 类近30 0 个纹理特0.002征参数进行对比。其中,Perc.和Mean来自于直方图0.003参数。Perc.10%是描述低于该百分位数所观察对象的百分比17-18 ,其可反映ROI内的微小变化。Mean代表ROI 的平均灰度值水平19-2 0 。本研究中MM患者的T2-STIR直方图参数值均高于退行性病变,可能与MM病灶散在多发、在T-STIR上呈较高信号有关。来自灰度游程矩阵的GLevNonU和灰度共生矩阵的SumAverge可反映ROI纹理明暗程度的异质性2 1。本研究中T2-STIR序列退行性病变的 GLevNonU值较MM组大,表明退行性病变椎体内部纹理的明亮度较低，内部纹理变化较大；这可能是由于骨髓脂肪变性、沉积引起骨质退变造成的。Correlat也称相关度，是用于评估图像的线性度,其来自于共生矩阵，Perc.10%(T,WI)Perc.10%(T2-STIR)135dr_GLevNonU(T,-STIR)参数联合2040100-特异度(%)OR(95%CI)1.228)1.033)0.999)可用于度量图像中行或列上灰度级的相似程度2-2 3。MM的T2-STIR上的Correlat值较高，原因可能为MM的病灶发生在骨髓组织（骨髓位于松质骨的骨小梁6080P值0.0010.9120.0020.2460.8660.0290.5680.9830.286100间）,病灶表面上虽呈散在、随机分布，但实则多沿椎体?的骨纹理排列2 4,这使MM累及的椎体图像在纵向或横向上具有相似度较高的灰度值，且其纵向或横向的纹理也较细致。WavEnLL_s-2来自于小波变换，小波变换通过高、低通滤波器将原始图像信号分成低频和高频分量，用于检测图像中的水平线、垂直线及交叉点和角点2 5,该高频分量反映图像的轮廓信息。本研究中T,WI和T,-STIR序列MM组的WavEnLL_s-2值高于腰椎退变组,可能是由于MM的骨质破坏多呈边缘清楚的穿凿状、鼠咬状,易与骨质疏松辨别。本研究运用Fisher、PO E+A C C、M I 方法，每种方法分别从各序列筛选出10 个最佳特征参数。其中Fisher为类间方差与类内方差之比,POE+ACC是基于分类错误概率和所选特征之间平均相关系数的最小化,MI是对纹理特征和类别变量之间依赖性的中国中西医结合影像学杂志2 0 2 3年9 月第2 1卷第5期衡量,这3种方法可降低纹理特征在所有类别中的分类错误率2 6 。在排除允余参数后,对9 个特征参数行logistic回归分析发现，T,WI序列的Perc.10%和T2-STIR序列的Perc.10%、135d r _G Le v No n U 是鉴别MM与退行性病变的独立预测因子。ROC曲线分析发现,3个参数联合的诊断效能较单一参数高(均P0.05）,表明选取的纹理特征参数均能很好地鉴别腰椎MM与腰椎退行性病变。本研究的不足之处：样本量较小，致使研究结果可能存在一定偏倚；未对腰椎退行性改变外的其他腰椎病变,以及累及腰椎的各型MM行进一步的分类研究,纹理分析对上述病变及病变亚型在鉴别上是否有帮助,有待进一步研究;仅对MRIT,WI、T,WI和T2-STIR进行分析,未对其他序列进一步研究，后续将逐步完善；采用人工方式勾画腰椎椎体，且未对椎体外的附件进行勾画，不可避免地存在人为因素的影响，未来将尝试通过半自动及自动的方式来对椎体和对应的附件进行勾画。综上所述，基于MRI图像的纹理特征参数在腰椎MM与腰椎退行性病变的鉴别中有一定临床价值,对指导临床制订有效的治疗方案具有重要意义。参考文献1 FIRTH J.Haematology:multiple myelomaJ.Clin Med(London,England),2019,19(1):58-60.2 BRIGLE K,ROGERS B.Pathobiology and diagnosis ofmultiple myelomaJ.Semin Oncol Nurs,2017,33(3):225-236.3 JOSHUA D E,BRYANT C,DIX C,et al.Biology and therapyof multiple myelomaJ.Med J Aust,2019,210(8):375-380.4 MESSIOU C,PORTA N,SHARMA B,et al.Prospectiveevaluation of whole-body MRI versus FDG PET/CT forlesion detection in participants with myelomaJ.RadiolImaging Cancer,2021,3(5):e210048.5】韩萍，于春水.医学影像诊断学M.北京：人民卫生出版社,2 0 16:57 1.6 MAYERHOEFER M E,MATERKA A,LANGS G,et al.Introduction to radiomicsJ.J Nucl Med,2020,61(4):488-495.7】瞿俊晨，贾传海，丁庆国，等.MRI图像纹理分析在腮腺良、恶性肿瘤鉴别诊断中的价值J中国口腔颌面外科杂志,2 0 2 2,2 0(2):17 3-17 6.8 POCIASK E,NURZYNSKA K,OBUCHOWICZ R,et al.Differential diagnosis of cysts and granulomas supportedby texture analysis of intraoral radiographsJ.Sensors(Basel),2021,21(22):7481.557.9 JIANG N,ZONG L,ZANG C,et al.Value of conventionalMRI texture analysis in the differential diagnosis ofphyllodes tumors and fibroadenomas of the breastJ.Breast Care(Basel),2021,16(3):283-290.10王敏红,周理想,冯湛常规MRI纹理分析鉴别脑胶质母细胞瘤和原发性中枢神经系统淋巴瘤的价值J中国癌症杂志,2 0 19,2 9(4)：2 8 4-2 8 8.11 MEYER H J,LEONHARDI J,HOHN A K,et al.CTTexture analysis of pulmonary neuroendocrine tumors-associations with tumor grading and proliferationJ.JClin Med,2021,10(23):5571.12 WU Z,BIAN T,DONG C,et al.Spinal MRI-basedradiomics analysis to predict treatment response in multi-ple myelomaJ.J Comput Assist Tomogr,2022,46(3):447-454.13 XIONG X,WANG J,HU S,et al.Differentiating betweenmultiple myeloma and metastasis subtypes of lumbarvertebra lesions using machine learning-based radiomicsJ.Front Oncol,2021,11:601699.14 LIU J,PEI Y,ZHANG Y,et al.Predicting the prognosisof hepatocellular carcinoma with the treatment of trans-catheter arterial chemoembolization combined with micro-wave ablation using pretreatment MR imaging texturefeatures J.Abdominal Radiology(New York),2021,46(8):3748-3757.15 HABASHI W,BDER-FARRAI A,SPACK N,et al.Three-dimensional surface texture analysis of fluorides effecton enamel erosionJ.J Clin Med,2021,10(19):4528.16 REINERT C P,KRIEG E,ESSER M,et al.Role of com-puted tomography texture analysis using dual-energy-basedbone marrow imaging for multiple myeloma characteri-zation:comparison with histology and established serologicparametersJJ.Eur Radiol,2021,31(4):2357-2367.17 FRUEHALD-PALLAMAR J,CZERNY C,HOLZER-FRU-EHWALD L,et al.Texture-based and diffusion-weighteddiscrimination of parotid gland lesions on MR imagesat 3.0 TeslaJ.NMR Biomed,2013,26(11):1372-1379.18 DENG X Y,CHEN H Y,YU J N,et al.Diagnostic valueof CT-and MRI-based texture analysis and imagingfindings for grading cartilaginous tumors in long bonesJ.Front Oncol,2021,11:700204.19宋新宇.磁共振T,WI图像纹理分析在前列腺增生与前列腺癌鉴别诊断中的应用J.实用医学影像杂志,2 0 2 1，22(6):629-631.20 HODGDON T,THORNHILL R E,JAMES N D,et al.CTtexture analysis of acetabular subchondral bone candiscriminate between normal and cam-positive hips J.Eur Radiol,2020,30(8):4695-4704.558.儿科影像学总肺体积、肺-肝信号强度比及ADC值评价胎儿肺成熟度的应用价值何永财，严杰文，曾亦如，崔运能广东省佛山市妇幼保健院放射科，广东佛山52 8 0 9 9【摘要】目的：探讨基于MRI测量总肺体积、肺-肝信号强度比及肺平均ADC值评价胎儿肺成熟度的应用价值。方法：收集42例2 6 39 周的胎儿，均行肺部MRI扫描，包括轴位及冠状位单激发自选回波序列（SSFSE）、轴位T,WI和DWI;分别测量并计算出胎儿总肺体积、肺-肝信号强度比及肺平均ADC值，分析其与胎龄的相关性；比较两肺体积及ADC值的差异。结果：42 例胎儿，总肺体积2 6.33110.7 4cm,平均（42.46 16.58)cm，与胎龄呈正相关（r=0.615,P0.001）。左、右肺平均体积分别为（2 6.40 8.94)cm、（34.3411.10)c m，差异有统计学意义（t=-11.26,P0.001）。肺-肝信号强度比为1.45 5.34，平均2.9 0 0.9 5，与胎龄呈正相关（r=0.671,P0.001）。肺ADC值为（1.54 3.2 3）10-mm%s，平均（2.9 10.37）10-mm/s，与胎龄无相关性（P=0.685）。左、右肺平均ADC值分别为（2.510.48）10-mm/s、（2.310.50）10-m m/s，差异无统计学意义（P=0.211）。结论：产前MRI可通过测量胎儿总肺体积及肺-肝信号强度比评价胎儿肺发育成熟度，而肺平均ADC值尚不能作为评价肺发育成熟度的一项参数。【关键词胎儿；肺发育；磁共振成像；表观扩散系数Application value of lung volume,signal intensity ratio of lung to liver and ADC value in evaluating fetal lung maturityHE Yongcai,YAN Jiewen,ZENG Yiru,CUI Yunneng.Department of Radiology,Foshan Women and Chilren Hospital,Foshan528099,China.Abstract Objective:To explore the application value of total lung volume,lung-to-liver signal intensity ratio and ADC valuebased on MR imaging based on MRI for evaluating the fetal lung maturity.Methods:Forty-two fetuses with 2639 weeksgestation were collected.All fetuses underwent lung MRI scanning,including axial and coronal SSFSE,axial T,WI and DWIsequence.The fetal total lung volume,lung-to-liver signal intensity ratio and ADC value were measured and calculated.Thecorrelation of the parameters and gestational age was analyzed,and the dfferences between the two lung volumes and ADCvalues were compared.Results:The total lung volume of 42 fetuses ranged from 26.33 to 10.74 cm,with an average of(42.4616.58)cm,which was positively correlated with gestational age(r=0.615,P0.001).The average volumes of the left and rightlungs were(26.408.94)cm and(34.3411.10)cm,respectively,and the difference was statistically significant(t=-11.26,P0.001).The lung-to-liver signal intensity ratio was 1.455.34,with an average of(2.90 0.95),which was positivelycorrelated with gestational age(r=0.671,P0.001).The lungD0I:10.3969/j.issn.1672-0512.2023.05.017ADC value ranged from 1.54 to 3.23 mm/s,with an average【基金项目】2 0 2 2 年佛山市自筹经费类科技计划项目（2 2 2 0 0 0 10of(2.91 0.37)10-3 mm/s,which was not correlated with03769)；2 0 19 年佛山市医学影像精准诊断工程技术研究中心资助项目(FS0AA_KJ819-4901-0049)。通信作者崔运能，Email:。中国中西医结合影像学杂志2 0 2 3年9 月第2 1卷第5期gestational age(P=0.685).The average ADC values of theleft and right lung were(2.510.48)10mm/s and(2.3121 SZCZYPINSKI P M,STRZELECKI M,MATERKA A,et al.MaZda-a software package for image texture analysisJJ.Comput Methods Programs Biomed,2009,94(1):66-76.22 LIU Y,FANG Q,JIANG A,et al.Texture analysis basedon U-Net neural network for intracranial hemorrhageidentification predicts early enlargement J.ComputMethods Programs Biomed,2021,206:106140.23 GAO X C,HUANG S Y.Identification of benign andmalignant breast masses based on texture analysis ofdifferent areas of mammographic images J.RadiologyPractice,2020,35(8):1037-1041.24范荣.基于IVIM-DWImapping图影像组学特征分析骨髓微环境的初步研究D.太原：山西医科大学,2 0 18.25 CHEN C,QIN Y,CHENG J,et al.Texture analysis offat-suppressed T,-weighted magnetic resonance imagingand use of machine learning to discriminate nasal andparanasal sinus small round malignant cell tumorsJ.Front Oncol,2021,11.26刘瑶,郑石磊,刘磐石磁共振T,WI的影像组学在前列腺癌预测中的临床研究J.航空航天医学杂志,2 0 2 0，31(3):259-262.（收稿日期2 0 2 3-0 1-17)