Investigation Of Diagnosis,Prediction Of Metastasis And HER2 Status Of Breast Cancer On ABVS Ultrasound Radiomics

Purpose : Breast cancer diagnosis,axillary lymph node metastasis risk assessment,and Human epidermal growth factor receptor 2(HER2)status prediction are the key links of the decision-making of clinical diagnosis and treatment of breast cancer.This research aimed to evaluate the clinical utility of an automated breast volume scanning(ABVS)-based ultrasound radiomics model for the breast cancer diagnosis,risk assessment of axillary lymph node metastasis,and preoperative prediction of HER2 status.Subjects and methods: 1.This retrospective study recruited patients with breast nodules(n=200)who underwent preoperative ABVS examination.208 ultrasound radiomics features were extracted from the axial and coronal planes of ABVS images.Recursive feature elimination(RFE),random forest(RF),and Chi-square test(Chi-square)were used for feature selection.The classifiers of support vector machine(SVM),logistic regression(LR),and extreme gradient boosting(XGBoost)were utilized to identify breast benign and malignant nodules.The classification performance of the ultrasound radiomics models was evaluated by the area under the curve(AUC),sensitivity,specificity,accuracy,and precision.Five-fold cross-validation was conducted to improve the robustness of the model and avoid overfitting.2.A total of 276 patients with early invasive breast cancer(EIBC)who underwent preoperative ABVS examination in two medical centers were included in this study.179 patients with early invasive breast cancer in our center were divided into a training set(n=143)and validation set(n=36)according to a ratio of 8:2,with97 patients with early invasive breast cancer in the external center as an external test set.Intraclass correlation coefficient(ICC)was utilized to evaluate the consistency of ABVS radiomics features and ABVS ultrasound features.The ultrasound radiomics signature was developed with the least absolute shrinkage and selection operator(LASSO)algorithm.The significant predictors of axillary lymph nodes metastases(ALNM)were screened by univariate logistic regression.The significant predictors were screened using multivariate logistic regression to identify the independent predictors,and then the ultrasound radiomics nomogram was constructed.The receiver operating characteristic curve(ROC),calibration curve(CC),and decision curve(DC)were utilized to assess the predictive performance and clinical application value of the Nomogram.The external test set was used to test the clinical generalization of the nomogram.3.This retrospective study recruited 271 patients with invasive breast cancer who underwent preoperative ABVS in two medical centers.174 patients in our center were randomly assigned to a training set and a validation set(ratio 8:2),and 97 patients with invasive breast cancer in the external center were used as a test set.The results of postoperative pathological immunohistochemistry(immunohistochemistry,IHC)or fluorescence in situ hybridization(fluorescence in situ hybridization,FISH)in breast cancer patients were regarded as the gold standard for determining a positive HER2,a molecular biomarker for breast cancer.Models were constructed based on ultrasound radiomics features extracted from the tumor area,3mm peritumoral area,and 5mm peritumoral area of ABVS,and clinical features(clinical,ABVS,and serological features).Multiple classifiers were performed to optimize the single data source model,and the feature combination method and model weighted combination method were conducted to optimize the combination model.ROCs were utilized to evaluate the predictive performance of the ultrasound radiomics models.The external test set was used to test the generalization of the ultrasound radiomics models.Results: 1.For a single plane or a combination of planes,a combination model of RFE and SVM yielded the best performance when identifying benign and malignant breast lesions;a combination model of RFE and LR exhibited the next-best performance.The ultrasound radiomics models based on a combination of ABVS planes performed better than those based on a single ABVS plane.Regarding the axial plane and coronal plane,the ultrasound radiomics model using a combination of RFE and LR yielded the best discriminant performance: the average AUC was 0.86 ± 0.06(95% confidence interval [confidence interval,CI],0.76-0.96);the mean values of corresponding sensitivity,specificity,accuracy,and precision were 87.90%,68.20%,80.70%,82.90%,and respectively.2.In the training,validation,and test sets,the ultrasound radiomics nomogram model included ABVS ultrasound radiomics signature,ultrasound-evaluated axillary lymph nodes status,convergence signs,and erythrocyte distribution width(standard deviation [SD]),achieved better predictive efficiency,and the AUC(95% CI)were 0.78(0.71,0.86),0.77(0.61,0.94),and0.83(0.75,0.91),respectively;with the sensitivity,specificity and accuracy were83.80%,80.00%,85.30%;63.40%,80.00%,66.00%;72.20%,80.00%,74.20%;respectively.The CC demonstrated good consistency between the predictive probabilities of the Nomogram and observation.The DC confirmed that the Nomogram had higher patient benefits.3.The weighted combination models in predicting HER2 status achieved better performance in the validation set.For the validation set,the single data source model,the feature combination model,and the model weighted combination model achieved the highest AUC(95%CI)of0.80(0.66,0.95),0.74(0.56,0.92),and 0.83(0.69,0.96),respectively;with the sensitivity and specificity were 100.00%,62.50%;81.80%,66.70%;90.90%,75.00%;respectively.For the external test set,the optimal model weighted combination model attained the AUC(95%CI)of 0.69(0.58,0.80);with the sensitivity and specificity were69.44%,51.61%;respectively.Conclusions: 1.A breast lesion identification method based on multi-planar ABVS images,which could be accurately and conveniently applied to the noninvasive classification of benign and malignant breast lesions preoperatively and provide a ultrasound radiomics basis for precise diagnosis and treatment of breast cancer patients.2.An ABVS-based radiomics nomogram was developed to predict the risk probability of ALNM in EIBC patients,which could be used as a preoperative and non-invasive predictive tool for ALNM and optimize clinical decision-making of patients with EIBC.3.A model weighted combination model based on ABVS-based intratumoral and peritumoral ultrasound radiomics features,clinical features,ABVS ultrasonic features,and serological features,which could be used to predict HER2 status in preoperative breast cancer patients simply and noninvasively,and propose ultrasound radiomics basis for precise targeted therapy in patients with breast cancer.