Application Of Prediction Model Based On Elastography And Doppler Ultrasonography In Breast Lesions

Application of ultrasonic elastography prediction model in breast lesionsObjectiveTo evaluate the application value of prediction model based on ultrasonic elastography combined with classification algorithm in differentiating benign and malignant breast lesions.MethodsA total of 201 cases of breast lesions diagnosed as BI-RADS 4 by ultrasound in the Provincial Hospital of Anhui Medical University from March 2015 to May 2018 were selected.When several lesions were found in the same patient,only the most suspicious or the largest lesions were included in the study,a total of 201 breast lesions.Each lesion was examined by Virtual touch Tissue Quantification(VTQ)and Virtual touch Tissue Imagine(VTI),and the shear wave velocity of the lesion and the gland,VTI image area of the lesion and two-dimensional image area of lesion were measured.Using the pathological results(including 91 benign cases and 110 malignant cases)as the gold standard,the diagnostic performance of various ultrasonic elastography parameters was evaluated by using the operating characteristic curve,and then the Chi-square automatic interactive detection method was used for classification analysis to establish the prediction model.ResultsIn ultrasound elastography parameters,SWV lesions of malignant lesions were significantly higher than benign lesions,the best cutoff value was 4.525m/s,the area under the ROC curve was 0.856,the sensitivity was 80.9%,and the specificity was 90.1%.SWV lesions/SWV glands in malignant lesions were also higher than those in benign lesions,the best cutoff value was 0.944,the area under ROC curve was 0.837,the sensitivity was 82.7%,and the specificity was 83.5%.As for the VTI area ratio,it was closer to 1 in benign lesions,so the VTI area of benign lesions did not change much compared with that of two-dimensional lesions.The best cut-off value was 1.201,the sensitivity was 71.8%,and the specificity was 62.6%.The prediction model includes SWV lesions and VTI area ratio.SWV lesions/SWV glands do not make the classification tree continue to be divided and improve the algorithm accuracy,so they are automatically deleted.The depth of the classification tree was divided into two branches.First,when the SWV lesions ≤3.958m/s,the VTI area ratio should be considered.When the VTI area ratio>1.493,the terminal node 5 was obtained,and the malignant probability was 73.3%.When the VTI area ratio>1.304 and≤1.493,the terminal node 4 is obtained,and the probability of malignancy is 38.1%.When the VTI area ratio is less than or equal to 1.304,terminal node 3 is obtained,and the probability of malignancy is 3.1%.When the SWVlesions>3.958m/s,terminal node 2 was obtained,and the malignant probability was 88.1%.The area under ROC curve of the classification algorithm was 0.901,the sensitivity was 98.2%,and the specificity was 68.1%.Using this classification algorithm,62 benign lesions of 201 lesions could avoid puncture biopsy.ConclusionsThe classification algorithm combined with VTQ and VTI technology has high diagnostic performance,which reduces the unnecessary needle biopsy by 30.8%and the unnecessary waste of resources.The Application of Ultrasound Quantitative Parameter Prediction Model in Breast LesionsObjectiveThe prediction model was established by using different ultrasonic quantitative parameters combined with classification algorithm,and its application value in diagnosing four types of breast imaging report and data system(BI-RADS)was discussed.MethodsA total of 241 breast lesions diagnosed by ultrasound in the Affiliated Provincial Hospital of Anhui Medical University in category 4 of BI-RADS classification were selected from January 2016 to December 2019.When several lesions were present in the same patient,only the most suspicious or the largest lesions were included in the study,a total of 241 breast lesions.Each lesion was examined by Acoustic Radiation Force Imaging(ARFI)and color doppler ultrasound.The blood flow resistance index(RI),shear wave velocity(SWV lesion,SWV gland)and ratio(SWV lesion/SWV gland)of the lesion were measured.Using the pathological results(including 140 benign cases and 101 malignant cases)as the gold standard,the diagnostic performance of various quantitative parameters of ultrasound was evaluated by using the Receiver operating characteristic curve,and then the Chi-square automatic interactive detection method was used for classification analysis to establish a prediction model.ResultsThe prediction model included two quantitative parameters,SWVlesion and RI,and the depth was two branches.When SWV lesion ≤3.795m/s,RI should be considered.When RI<0.620,the terminal node 3 was obtained,and the malignant probability was 3.3%.When RI>0.79,the terminal node 4 was obtained,and the malignant probability was 43.3%.When the SWV lesion>3.795 m/s,then RI was considered.When RI<0.71,the terminal node 5 was obtained,and the malignant probability was 83.3%.When RI>0.71,node 6 was obtained,and all of them were malignant lesions.At the end node 3,the probability of malignancy was only 3.3%,and follow-up for the lesion could be considered.For this classification tree,the area under ROC curve was 0.938,which was significantly higher than 0.781(Z=4.72,P ≤0.001),0.858(Z=2.68,P=0.003)and 0.856(Z=2.81,P=0.002)of RI,SWV and SWV lesions/SWV glands.The sensitivity and specificity were 98.6%and 57.4%.Using this predictive model,58 of the 101 benign lesions were correctly classified as benign,thereby avoiding 58 unnecessary needle biopsies or surgeries.ConclusionThe prediction model established by using the classification algorithm of SWV lesions combined with RI quantitative data has high diagnostic performance,which can avoid subjective deviation of ultrasound to a large extent,objectively help clinical decision whether to carry out needle biopsy,and reduce unnecessary invasive examination.