Logistic Regression Analysis Of Ultrasound Multimodality Features Of Breast Cancer

Objective:Logistic regression analysis was used to analyze the ultrasound multimodality features of breast mass with BI-RADS 3~5,and to screen closely related risk factors of breast cancer and establish a multimodal diagnosis model.Methods:109 patients were treated in the Hospital of Shanxi Medical University from September 2015 to December 2017.A total of 125 breast masses were diagnosed by pathological findings.Two-dimensional gray-scale ultrasound,CDFI,UE,and CEUS were performed on all selected breast masses.The characteristics of each modality were recorded and analyzed.Binary Logistic regression analysis was used to screen closely related risk factors of breast cancer and establish a multimodal diagnosis model.Comparison of diagnostic efficacies of single modality(2-D gray-scale ultrasound,CDFI,UE,CEUS)and multimodal diagnostic models.Results:(1)This study included 109 patients with a total of 125 tumors,including 67 benign tumors and 58 malignant tumors.The age of patients of malignant tumors was greater than that of benign tumors(P<0.05);There was no significant difference in mean maximum diameter between the two groups(P>0.05).(2)Univariate analysis: patient age,two-dimensional gray-scale ultrasound features(morphology,orientation,margin,posterior echo,microcalcification,lymph node),CDFIgrade,UE score,and CEUS features(enhanced level,post-contrast bolus boundary,morphology,and range changes,penetrating blood vessels or radial enhancement),a total of 14 factors were statistically significant in the benign and malignant groups(P<0.05).(3)Logistic regression equation: Logit(P)=-18.544 + 1.321 * Non-parallel position +2.691 * Non-circumscribed edge + 2.185 * Microcalcification + 1.698 * CDFI classification(Grade II/III)+ 2.118 * UE score(4~5 points)+ 1.868 * CEUS range expanded + 1.152 * Penetrating blood vessels or radial enhancement.According to the OR value,7 closely related risk factors were non-circumscribed edge,intratumor microcalcification,and CEUS range expanded,penetrating blood vessels or radial enhancement,UE score(4~5 points),non-parallel position,CDFI classification(Grade II/III)in descending order.(4)The area under the ROC curves of the five methods of two-dimensional gray-scale ultrasound,CDFI,UE,CEUS and ultrasonic multimodal diagnosis model were 0.775,0.627,0.592,0.755 and 0.863,respectively.Conclusion:(1)The risk factors for breast cancer are closely related to the non-circumscribed edge,intratumor microcalcification,and CEUS range expanded,penetrating blood vessels or radial enhancement,UE score(4~5 points),non-parallel position,CDFI classification(Grade II/III)in descending order.(2)The ultrasound multimodal diagnosis model has better diagnostic performance than single modality,and can improve the accuracy of the diagnosis and discrimination of benign and malignant diagnosis of breast mass with BI-RADS type 3~5.