| Objective:To explore the texture features model of CT arterial phase and venous phase based on wavelet transform and the subjective diagnosis model,combined prediction model in primary thyroid papillary carcinoma.The four models predict the value of central group lymph node metastasis.Methods:1.548 patients(612 nodules)with clinicopathologically confirmed Papillary Thyroid Carcinoma(PTC)was collected from December 2013 to December 2017 in the first affiliated hospital of Kunming Medical University.Collecting clinical data of age,sex,and whether the PTC patients were complicated with Hashimoto's thyroiditis(HT)or adenoma/nodular goiter.Divided into metastatic group and non-metastatic group according to pathology.2.8 CT subjective signs of PTC were independently evaluated by two physicians with extensive experience in head and neck imaging diagnosis.The evaluate content consist of:nodule position,number,size,morphology,calcification,enhanced posterior border,thyroid capsule invasion,surrounding tissue invasion.Kappa test was used to evaluate the consistency of the two subjects' evaluation of the subjective signs of 30 randomly selected patients.Chi-square test was used to compare the differences in age,gender,concurrent HT or adenoma/nodular goiter and CT subjective signs between the two groups.Subjective signs of CT with statistically significant single factors were included in the multivariate logistic regression analysis to construct a physician's subjective diagnosis model.3.Texture features after wavelet transform(including 576 first-order,second-order and high-order texture features)were extracted from portal venous-phase computed tomography(CT)of PTC.The difference of texture feature parameters between the two groups was compared using independent sample t-test(normal distribution)or Mann-Whitney U test(non-normal)Distribution).The receiver operating characteristic(ROC)curve was used to analyze the statistically significant characteristic parameters of CT images in the arteriovenous and venous phases to predict the efficacy of central lymph node metastasis(CLNM).Manual screening of arteriovenous diagnosis CLNM Area Under Curve(AUC)values ranked the top 10 texture features as the best feature parameters,Multivariate logistic regression analysis was performed after multicollinearity test to construct the arterial texture feature model and venous texture features model.4.Integrating multivariate logistic regression into the multivariate logistic regression of clinical,physician subjective diagnosis features and texture features,Constructing a combined prediction model.5.The ROC curve was used to examine the physician's subjective diagnosis model,the arterial and venous texture features models,and the combined prediction model to diagnose the efficacy of PTC central lymph node metastasis.The parameters included:AUC,accuracy,sensitivity and specificity,Using Delong test to compare the AUC of the four models.Results:1.General conditions:In 160 cases(189 pieces)of PTC nodules,postoperative pathology confirmed 110 cases of CLN metastasis and 79 cases of non-metastasis group.There were 43 males(43/160,26.88%)and 117 females(117/160,73.13%),aged 15 to 74 years,mean age(43.43±0.93)years old.The maximum diameter of PTC primary lession was 5.1?42.6mm,and the average was about 13.57±7.30mm.The mean internal time between CT and surgery was 6(0?14)d.PTC with HT in 19 cases(19/160,11.88%),without HT(141/160,88.13%);2 cases with adenoma(2/160,1.25%),41 cases with nodular goiter(41/160,25.63%),no adenoma/nodular goiter(117/160,73.13%).There was a statistically significant difference in age between the two groups(P<0.05),and CLN metastasis was more likely to occur at age<45 years.The other three clinical factors were not statistically different between the two groups(P>0.05).2.Comparison of CT subjective signs:There were significant differences in thyroid capsule invasion and surrounding tissue invasion between the two groups(P<0.05).There was no significant difference in the residual subjective signs between the two groups(P>0.05).The two observers had a good agreement on the 7 subjective signs of the 30 patients(Kappa values were>0.61),while the assessment of "surrounding tissue invasion"was moderate,with a Kappa value of 0.58.3.Texture analysis of CT images in different phases showed that:There were statistical differences between the two groups in the arterial phase and the AUC values of the optimal texture parameters for the diagnosis of PTC CLNM were:0.608,0.603,0.611,0.609,0.605,0.608,0.61,0.605 0.607,0.604(P<0.05);and The optimal texture parameters in the venous phase were:0.630,0.616,0.610,0.607,0.605,0.604,0.602,0.599,0.595,0.595(P<0.05).4.The statistically significant variables,including age,thyroid capsule invasion,surrounding tissue invasion,and optimal texture parameters of arteriovenous phase were included in the multivariate logistic regression to construct a combined prediction model.Regression analysis showed that age,thyroid capsule invasion and texture features were independent risk factors for the diagnosis of CLNM in combined prediction model(P<0.05).5.The physician's subjective diagnosis model,arterial phase texture feature model venous phase texture feature model and combined prediction model diagnosis CLNM AUC values and 95%CI(Confidence Interval)were 0.655(0.58-0.72),0.668(0.60-0.74),0.693(0.62-0.76),0.803(0.739-0.867);Sensitivity values were 62.21%,66.4%,62.84%,76.07%;Specificity was 82.35%,60.34%,60.47%,70.83%;Accuracy values were 64%64.6%,63%,74.1%.The physician's subjective diagnosis model pre diets the sensitivity of the CLNM was lower,but the specificity was higher than the arteriovenous phase texture feature model.The arteriovenous phase texture feature models AUC and sensitivity were higher than the subjective diagnosis model.and the combined prediction model's AUC,sensitivity and accuracy of the combined prediction model were the highest among the four models.6.There were no significant differences in the AUC values between the subjective diagnosis model,the arterial phase texture feature model,and the venous phase texture feature model(P>0.05).But the AUC values of the combined prediction model were higher than those of the physician's subjective diagnosis model and the arteriovenous texture feature model(P=0.046,0.001,0.003).Conclusions:1.Based on the CT image of the thyroid papillary carcinoma,the CT image analysis model has a certain value for predicting the central lymph node metastasis.The arterial and venous texture features predict the AUC value of the central lymph node metastasis of thyroid papillary carcinoma is 0.668,0.693.2.The arterial phase and venous phase wavelet texture model Based on the CT image of thyroid papillary carcinoma primary lession has no difference in predicting the efficacy of central lymph node metastasis,There is no difference between CT morphological subjective model and arteriovenous phase texture feature model in predicting the efficacy of central lymph node metastasis.3.Combined with wavelet texture features,CT subjective signs and clinical risk factors constructed combined prediction model based on the CT image of thyroid papillary carcinoma primary lession,Which has a high AUC value,sensitivity and accuracy for predicting central lymph node metastasis(0.803?76.07%?74.1%). |
PDF Full Text Request