【作者】 王晓屹；

【导师】 王跃涛；

【作者基本信息】 苏州大学 ， 影像医学与核医学， 2013， 硕士

【摘要】 目的：以超声声像图特征及患者性别、年龄作为甲状腺结节良、恶性鉴别诊断的自变量，建立Logistic回归模型，比较超声声像图特征及患者性别、年龄等因素在甲状腺结节良恶性鉴别诊断中的价值。材料与方法：收集经手术切除后病理证实的286例甲状腺结节，其中良性结节183例作为对照组，恶性结节103例作为病例组。收集两组患者的年龄、性别资料以及甲状腺结节的超声声像图特征，包括结节的数目、最大直径、形态、纵横比、边界、边缘特征、回声水平、内部质地、有无声晕、钙化情况、后方回声有无衰减和血流情况。两组之间年龄因素使用t检验，其余两组间参数采用两独立样本比较的χ2检验，比较两组的参数有无统计学差异。将有统计学差异的参数作为自变量，以甲状腺结节良恶性作为应变量，分别进行赋值，向前逐步法测定恶性结节的超声特征，最后建立Logistic回归模型，P＜0.05有统计学意义。结果：两组病例之间年龄经t检验，其余变量经χ2检验。两组病例之间结节的数目无统计学差异（P=0.718），其余两组之间参数均有统计学差异。经过最大似然估计前进法逐步回归，筛选出边缘特征、钙化情况、年龄、有无侵犯包膜、回声水平这五个最有价值的参数，其中边缘毛刺（OR＝20.830，r＝2.970）、微钙化（OR＝14.540，r＝1.796）、年龄（OR＝16.359，r＝-0.074）、有侵犯包膜（OR＝6.890，r＝1.351）、极低回声（OR＝9.747，r＝2.553）作为自变量进入Logistic回归模型。建立模型为：Logit（P）=0.719+2.970×边缘（2）+1.796×钙化（3）－0.074×年龄+1.351×包膜（1）+2.553×回声（4）。该模型经似然比检验有统计学意义（χ2=170.571，P=0.000），以该模型预报286例甲状腺结节，如果P>0.5时预报为恶性结节，P≤0.5时预报为良性结节，预报正确率达85.3%。结论：1.甲状腺结节的单、多发对良恶性的鉴别没有意义。2. Logistic回归模型筛选出5个对甲状腺结节恶性鉴别有意义的特征变量：患者的年龄、甲状腺结节的边缘特征、钙化情况、有无侵犯包膜、回声水平。3.患者的年龄与甲状腺结节的恶性呈负相关，年龄越小，其甲状腺结节的恶性可能性越大。4.边缘毛刺、微钙化、有侵犯包膜、极低回声与甲状腺结节的恶性呈正相关，这四个特征出现越多，其甲状腺结节的恶性可能性越大。5.进入方程的变量对方程重要程度：边缘毛刺＞年龄＞微钙化＞极低回声＞侵犯包膜。6.以超声声像图特征及患者的年龄等多个因素建立Logistic的回归模型有助于甲状腺结节的良、恶性鉴别。更多还原

【Abstract】 Purpose: Sonographic features, patient’s gender and age as the independent variablesto indentify the benign and malignant of thyroid nodule, establish a logistic regressionmodel, a benign thyroid nodule. Compare the values of sonographic characteristics, thepatient’s gender, age and other factors in the differential diagnosis of benign and malignantthyroid nodulesthe.Materials and methods: A retrospective analysis of two hundred and eighty six casesof thyroid nodules confirmed by pathology after operation excision,one hundred and eightythree nodules were benign as control group, one hundred and three nodules were malignantas case group. A retrospective collection of two groups of patients with age, the sexualdistinction and the ultrasonographic features of thyroid nodules, including nodule number,diameter, shape, boundary, internal echo level, edge feature, anteroposterior and transversediameter ratio, rear echo attenuation, halo, calcification and the blood flow. In addition,multiple logistic regression analysis with a forward stepwise method for selection ofsignificant variables was performed to determine independent ultrasound predictors formalignancy from the ultrasound characteristics that showed statistical significance(P<0.05). At last Logistic regression model was set up.Results: The t test was used to evaluate the age of patient between the benign andmalignant nodules, the χ2test was used to evaluate the other parameters of two groups. Nosignificant difference between the two groups in the number of nodules (P=0.718), theother parameters of two groups were significant differences between theparameters(P<0.05). According to the output result of SPSS19.0, there were fiveindependent variable selected on the final step of Logistic regression analysis: calcification of nodules,the blood flow, age, capsular invasion,the echo level. There were the five of themore valuable parameter:spiculated margin (OR=20.830,r=2.970), calcification(OR=14.540, r=1.796), age (OR=16.359,r=-0.074), capsular invasion (OR=6.890, r=1.351), marked hypoechogenicity(OR=9.747,r=2.553). The model was: logit(P)=0.719+2.970edge(2)+1.796calcification(3)-0.074age+1.351capsularinvasion(1)+2.553theecho level(4). The likelihood ratio test was used to evaluate the fitting situation of thewhole model.It was statistically significant (χ2=170.571,P=0.000), the modle was used topredict the two hundred and eighty six thyroid nodules. The modle could distinguish thetwo kinds of nodules. When the regression value P was more than0.5, the prediction wasmalignant nodules. When the regression value P was less the or equal to0.5, the predictionwas benign nodules. The correct rate of prediction was85.3%.Conclusion:①Thyroid nodules which were single or multiple can be helpless todifferential diagnosis of benign and malignant.②Five of fifteen independent variableswere selected by Logistic regression analysis: the age of patients, calcification, edgefeatures, capsular invasion, the echo level.③The age of patients was negatively correlatedwith malignant, the smaller the age, the more possibility of malignant thyroid nodules were.④Spiculated margin, microcalcification, capsular invasion and marked hypoechogenicityare positively correlated with malignant thyroid nodules, this four ultrasonic characteristicsmore appeared, the more possibility of malignant thyroid nodules were.⑤According theimportance of the model, arranged from high to low is: spiculated margin＞age＞microcalcification＞marked hypoechogenicity＞capsular invasion.⑥The binary Logisticregression model with the sonographic characteristics and the patients age can be helpful todifferentiation of benign and malignant thyroid nodules.更多还原