Ultrasound-based Multiregional Elastography And Radiomics For Breast Cancer Diagnosis

Part Ⅰ The Significance of Multiregional Sound Touch Elastography in the Differential Diagnosis of Benign and Malignant Breast Lesions Objectives:To determine the best STE parameter,evaluate and compare the performance of Young’s modulus values based on sound touch elastography(STE)in the differential diagnosis of benign and malignant breast lesions.The diagnostic performance was further improved by grading with the 2013 breast imaging reporting and data system(BI-RADS).Method:1.This study retrospectively enrolled 187 patients with breast lesions with clinical data and pathological results admitted to the First Affiliated Hospital of Nanchang University from June 2019 to April 2022.2.Conventional ultrasound and STE images were collected by two experts with more than 5 years of experience in breast ultrasound and were reviewed by another two experts with more than 10 years of experience in breast ultrasound.Conventional ultrasound data,including location,size(maximum diameter),the distance between the lesion and nipple,and BI-RADS category,were recorded.STE ultrasound data,including Young’s modulus values(mean,maximum,minimum,and standard deviation)of the lesion and shell size 1 mm,2 mm,and 3 mm shells around the lesion,i.e.,Emean-A/1shell/2shell/3shell,Emax-A /1shell/2shell/3shell,Emin-A/1shell/2shell/3shell,and Esd-A/1shell/2shell/3shell.3.Young’s modulus values of breast lesions and shells were compared between the benign and malignant breast lesions.4.Receiver operator characteristic(ROC)curve was drawn according to the pathological result,and the optimal cut-off value was obtained according to the maximum Youden index.Sensitivity,specificity,positive predictive value,negative predictive value,accuracy,and area under the curve(AUC)were calculated.The STE parameter with the best diagnostic performance was determined by comparing the performance of all STE parameters.5.The best STE parameter combined with the BI-RADS category was used to establish an STE-BI-RADS category.The BI-RADS category was downgraded when the best STE parameter was below the cut-off value,and the BI-RADS category was upgraded when the best STE parameter was higher than the cut-off value.The diagnostic performance of the STE-BI-RADS category and BI-RADS category in differentiating benign and malignant breast lesions was compared.Results:1.187 patients with 187 breast lesions were enrolled.There were 89(47.1%)breast benign lesions and 98(51.9%)malignant lesions.Patients with malignant lesions was significantly older than patients with benign lesions [54(47-59)vs.44(31-49)years,p<0.001].2.In conventional ultrasound evaluation,the size(maximum diameter)of malignant breast masses was significantly larger than that of benign breast masses[25(17-31)vs.15(11-18)mm,p<0.001].There was no significant difference in the location and distance between the lesion and the nipple between benign and malignant lesions(p=0.088 and 0.560).The BI-RADS category of malignant lesions was significantly higher than that of benign lesions(p<0.001).There were 42 of category 3,35 of category 4a,9 of category 4b,2 of category 4c,and 1 of category 5in the benign lesions.There were 6 lesions of category 3,17 of category 4a,28 of category 4b,13 of category 4c and 34 of category 5 in the malignant lesions.3.Malignant lesions had a significantly higher Emean-A,Emax-A,and Esd-A and significantly lower Emin-A than benign breast lesions(p<0.001).Malignant lesions had a significantly higher Young’s modulus values of 1 mm,2 mm,and 3mm shell(Emean-1shell/2shell/3shell,Emax-1shell/2shell/3shell,Emin-1shell/2shell/3shel l,and Esd-1shell/2shell/3shell)than benign breast lesions(p<0.001).The Emean-A and Emax-A levels of benign and malignant masses in the large breast mass group were significantly higher than those in the small breast mass group(p<0.05).The Emax-1shell/2shell/3shell of benign masses in the large breast mass group were significantly higher than those in the small breast mass group.The Emean-1shell/2shell/3shell,Emax-1shell/2shell/3shell of malignant masses in the large breast mass group were significantly higher than those in the small breast mass group(p<0.05).4.AUCs of Emean-A,Emax-A,Emin-A,and Esd-A were 0.810(0.745-0.875),0.837(0.779-0.895),0.728(0.655-0.801)and 0.739(0.668-0.810),respectively.AUCs of Emean-1shell,Emax-1shell,Emin-1shell,and Esd-1shell were 0.832(0.771-0.893),0.862(0.809-0.915),0.736(0.663-0.809),and 0.739(0.669-0.810),respectively.AUCs of Emean-2shell,Emax-2shell,Emin-2shell,and Esd-2shell were0.853(0.797-0.910),0.873(0.823-0.923),0.752(0.680-0.823),and 0.767(0.700-0.834),respectively.AUCs of Emean-3shell,Emax-3shell,Emin-3shell,and Esd-3shell were 0.867(0.812-0.923),0.887(0.838-0.936),0.769(0.700-0.839),and0.784(0.720-0.849),respectively.Comparing AUCs of all STE parameters,the performance Emax-3shell was significantly higher than that of Young’s modulus valuesof lesions,Emin-1shell,Emin-1shell/2shell/3shell and Esd-1shell/2shell/3shell(p<0.05).The AUC,sensitivity,specificity,accuracy,and positive and negative predictive values of Emax-3shell were 0.887(0.838-0.936),91.8%,74.2%,83.4%,79.6%,and 89.2%,respectively.5.The AUC of the STE-BI-RADS category was significantly higher than that of the BI-RADS category [0.915(0.875-0.955)vs.0.867(0.817-0.916);p=0.002].The STE-BI-RADS category has a higher specificity(76.3% vs.47.2%),positive predictive value(80.4% vs.66.2%)and accuracy(82.3% vs.71.7%),similar sensitivity(93.9% vs.93.9%),negative predictive value(90.8% vs.87.5)and missed diagnosis rate(12.5% vs.12.5%),and lower unnecessary biopsy rate(24.5% vs.33.8%).Conclusion:Young’s modulus values of the breast lesion and shell around itself based on STE ultrasound have excellent performance in the differential diagnosis of benign and malignant breast lesions.Emax-3shell may be the best STE parameter for breast cancer diagnosis.STE ultrasound can improve the specificity,negative predictive value,and accuracy of the BI-RADS category and reduce the probability of unnecessary biopsy without lowering the sensitivity and negative predictive value in diagnosing breast cancer.Part Ⅱ Ultrasound-based Multiregional Radiomic Model in the Differential Diagnosis of Benign and Malignant Breast LesionsObjectives:To evaluate the value of ultrasound-based radiomics features of peritumoral and ipsilateral parenchymal regions in the differential diagnosis of benign and malignant breast lesions,and develop and validate a multiregional model combining intratumoral,peritumoral,and parenchymal radiomic features.Methods:1.526 patients with 594 breast masses were enrolled in this study,including a training cohort(n=339),an internal test cohort(n=146),and an external test cohort(n=106).2.Conventional ultrasound data were collected,including the location,size(maximum diameter),distance between the lesions and the nipple,and BI-RADS category.3.Intratumoral region of interest(ROI),3 mm peritumoral ROI,and parenchymal ROI 10-20 mm away from the lesion were manually segmented.464 intratumoral radiomic features,455 peritumoral features,and 455 parenchymal features were extracted.The recursive feature elimination(RFE)using the random forest as a based classifier and the Boruta method was used for feature selection.In the training cohort,the selected radiomic features were trained based on the random forest classifier to develop the intratumoral,peritumoral,and parenchymal models,intratumoral&peritumoral(In&Peri),intratumoral&parenchymal(In&P),and intratumoral&peritumoral&parenchymal(In&Peri&P;multiregional)models.4.Model performance was evaluated in the internal and external test cohorts.Discrimination of models was assessed using the area under the receiver operator characteristic(ROC)curve(AUC),sensitivity,specificity,and accuracy.Calibration of models was evaluated using the calibration curve and the Hosmer-Lemeshow test.5.The integrated discrimination improvement(IDI)test evaluated the performance improvement of radiomic models with added peritumoral and parenchymal features.6.In this study,the malignant probability predicted by the best radiomic model is called the radiomic signature score(Rad-score).The optimal cutoff value of the Rad-score was determined by maximizing the Youden index.We further constructed a combined(Rad-BI-RADS)category by downgrading and upgrading the BI-RADS according to the optimal Rad-score cutoff value: the BI-RADS category was downgraded when the Rad-score below the cutoff value,and the BI-RADS category was upgraded when the Rad-score higher the cutoff value.Results:1.Breast cancer accounted for 54.9%(186/339),54.1%(79/146),and 65.1%(69/106)of the training,internal test,and external test cohorts(p=0.143).2.The age,size,frequency of noninner quadrant location,and BI-RADS category of the malignant breast lesions were significantly higher than benign lesions(p<0.05).There was no significant difference in the distance between the lesion and the nipple between benign and malignant lesions(p=0.706).3.10 intratumoral radiomic features,9 peritumoral radiomic features,and 5peritumoral radiomic features were selected for training the intratumoral,peritumoral,and parenchymal models,respectively.19 intratumoral and peritumoral radiomic features,14 intratumoral and parenchymal radiomic features,and 23 intratumoral,peritumoral,and parenchymal radiomic features were selected to train the In&Peri,In&P,and In&Peri&P models.4.Based on the data of the internal and external test cohorts,in the intratumoral,peritumoral,parenchymal,In&Peri,In&P,and In&Peri&P models,the In&Peri&P model had the highest AUC(0.929 and 0.911)and accuracy(82% and 80%),the In&P model had the highest sensitivity(89% and 86%),and the In&Peri model had the highest specificity(88% and 78%).The intratumoral,In&Peri,and In&Peri&P models showed good calibration.5.AUC of the intratumoral model was indifferent to that of the peritumoral model(p=0.529 and 0.435)and significantly better than that of the parenchymal model(p<0.05)in the internal and external test cohorts.The In&Peri and In&P models significantly improved the AUC of the intratumoral model(p<0.05).The In&Peri&P model had the best differential diagnosis performance(p<0.05).The In&Peri&P model significantly improved AUC values of the intratumoral model,In&Peri,and In&P models(p<0.05).6.The malignant probability predicted by the In&Peri&P model is the Rad-score of this study.The Rad-BI-RADS category significantly improved the specificity(71%/75% vs.39%/46%)and accuracy(83%/86% vs.73%/76%)and maintained sensitivity(93%/90% vs.96%/97%)of the BI-RADS category in the internal and external test cohorts.Conclusions:Ultrasound-based radiomic features extracted from peritumoral and parenchymal regions may improve diagnostic performance for breast cancer over intratumoral features alone.The multiregional model may optimize radiomics in discriminating malignant from benign breast lesions,which suggests its potential to yield better diagnostic performance.