Logistic Regression Models Of Multi-parametric Ultrasound In The Assessment Of Benign And Malignant Cervical Focal Lesions

Objective: Two-dimensional gray-scale ultrasonography is a common method for the evaluation of cervical focal lesions,but its diagnostic performance is not satisfactory.Color Doppler flow imaging(CDFI),Pulsed-wave Doppler measurement,contrast-enhanced ultrasound(CEUS)and Ultrasonic elastography(UE)are complementary to two-dimensional gray-scale ultrasonography for the assessment of benign and malignant cervical focal lesions.However,what is the role of a diagnostic model based on mathematics of integrated characteristics of patients and focal lesions for the evaluation of the benign and malignant cervical focal lesions has not been fully understood.The objective of this study was to investigate that whether a multi-parameter ultrasound(MPUS)diagnostic model based on mathematics was helpful to the differential diagnosis of benign and malignant cervical lesions or not.Methods: A retrospective study had conducted for the assessment of benign and malignant cervical lesions,histopathological confirmed 232 cervical focal lesions with history of Duplex ultrasound(Duplex US)examination were included,in which there were 87 lesions in 86 patients received multi-parameter ultrasound(MPUS)evaluation(including ultrasound elastography,UE).Related characteristics and parameters of the patients and lesions were studied,two diagnostic models based on single-factor regression analysis and logistic regression model using the characteristics and parameters of the patients and lesions were established,comparisons of these characteristics and parameters in the differential diagnosis of benign and malignant cervical focal lesions were done,and the receiver-operating characteristic curve(ROC)and the area under the curve(AUC)for these characteristics and parameters were built,and the diagnostic performances were calculated,respectively.28 non-lymphatic cervical focal lesions of the 232 cervical focal lesions in 222 patients were selected and studied independently,with the same protocols as used in the previous entities.Results: Of the 232 cervical focal lesions in 222 patients,there were 110 benign lesions and 122 malignant lesions,a study by using single factor analysis showed that the size of the cervical focal lesions(short diameter,length diameter),shape,S/L(short diameter/length diameter),border,internal compositions,echogenic features,abnormal anechoic area,punctate hyperechogenic appearance,status of lymph nodal hilus,coalescence of lymph nodes,degree of vascularity,vascularity type and the funicular high echoic appearance of the Duplex US features between benign and malignant lesions were all significant differences(all P < 0.05);AUC of lymphatic hilum feature was > 0.70,and AUCs of other features were all < 0.70.Of the 87 lesions in 86 patients,there were 31 benign lesions and 56 malignant lesions.A study by using single factor analysis showed that the ages of the patients,the short diameter,shape,S/L,boundary,punctate hyperechogenic lymphatic hilus,coalescence of lymph nodes,degree of vascularity,vascularity type,and UE image score,elastography strain ratio(ESR)of the MPUS features between benign and malignant lesions all had significant differences(all P < 0.05),the AUC of degree of vascularity showed no statistical significance in the differentiation of benign and malignant lesions(P > 0.05).AUCs of lymphatic hilus feature,short diameter,UE image score,and ESR were all > 0.70,and AUCs of other features were all < 0.70.Single factor Logistic regression was used to select the optimal characteristic parameters of MPUS of the 87 cervical focal lesions,the results showed that age,short diameter,shape,S/L≥0.5,boundary,punctate hyperechogenic appearance,vascularity type,UE image score,ESR of lesions all had significant differences between benign and malignant cervical focal lesions(all P<0.05).The formula of Duplex US regression model were Logit(P)=-4.365+(1.460×age>40)+(1.581×S/L≥0.5)+(3.416×peripheral vascularity or mixed type of vascularity).The regression coefficients P values of the above characteristics were all<0.05.The sensitivity,specificity,positive predictive value and negative predictive value were94.4%,61.3%,86.3%and 80.9%,the predictive accuracy was 82.4%,and the AUC value of ROC curve was 0.861.The formula of MPUS regression model were Logit(P)=-5.962+(2.324×S/L≥0.5)+(3.911×peripheral vascularity or mixed type of vascularity)+(2.513×ESR>3.24).P values of the regression coefficients of the above characteristics were all<0.05,and the sensitivity,specificity,positive predictive value and negative predictive value were 98.1%,61.3%,81.5%and 95.0%,respectively,the predictive accuracy was 84.7%,and the AUC value of ROC curve was 0.894.The AUC of Duplex US,UE,Duplex US model and MPUS model were compared,and the difference of AUC values between Duplex US and UE,Duplex US model and UE,Duplex US model and MPUS model were not statistically significant(all P>0.05).The difference between the AUC values of Duplex US model Duplex US was statistically significant(P<0.05).The difference in AUC values between the MPUS model and Duplex US,and between the MPUS model and UE was statistically significant(P<0.05).Of the 232 cervical focal lesions in 222patients,there were 28 non-lymphatic cervical focal lesions in 27 patients,including24 benign lesions and 4 malignant lesions.A study using univariate analysis for the MPUS characteristics of these lesions showed that the difference of boundary characteristic between benign and malignant cervical focal lesions was significant(P<0.05),with AUC of 0.813,sensitivity of 87.5%and specificity of 75.0%.Conclusion: Duplex US model and MPUS model based on mathematics present significant higher diagnostic performances than those of Duplex US and UE for the differentiation between benign and malignant cervical focal lesions.