Application Of Qualitative Features And Quantitative Three-dimensional Kinetic Features Of Magnetic Resonance Imaging In The Assessment Of Axillary Lymph Node Status And Prognosis In Breast Cancer

Background:Breast cancer is the malignancy with the highest incidence and mortality rate in women,which seriously affects women’s life and health.Accurate preoperative assessment of the axillary lymph node status of breast cancer can guide the development of clinical axillary management strategies to ensure that patients can benefit from treatment,while minimizing postoperative complications and improving their quality of life.Objective:Part Ⅰ.To explore the predictive value of breast edema score(BES)for preoperative assessment of pathological axillary lymph node(p ALN)burden in patients with early-stage breast cancer.Part Ⅱ.To explore the predictive value of machine learning models constructed by combining preoperative clinicopathological factors and MRI qualitative features on lymphovascular invasion(LVI)and prognosis in patients with breast cancer.Part Ⅲ.Combining breast MRI qualitative features and quantitative three-dimensional(3D)volumetric kinetic features of DCE-MRI for comprehensive assessment of axillary lymph node burden in early-stage breast cancer,and evaluating its value for prognostic assessment.Part Ⅳ.To combine clinicopathological factors,the pre-and post-treatment MRI qualitative features and quantitative 3D volumetric kinetic features of DCE-MRI to assess the axillary lymph node status and prognosis after neoadjuvant treatment for breast cancer.Methods:Part Ⅰ.Clinicopathological information was collected from 1092 patients(677 in the training dataset;415 in the validation dataset)with c T1-T2 stage breast cancer.The primary endpoint of this study was p ALN burden(<3 vs.≥3).MRI qualitative features(BES and MRI-ALN status)were evaluated.Logistic regression models were used to analyze the relationship between BES and p ALN burden,and the area under the curve(AUC),net reclassification improvement(NRI)and integrated discrimination improvement(IDI)were used to evaluate the predictive value of BES on p ALN burden.Part Ⅱ.575 patients(386 in the training dataset;189 in the validation dataset)with preoperative MRI for breast cancer were included.Pathological LVI status was used as the endpoint.Based on the clinicopathological factors and MRI qualitative features,logistic regression(LR),e Xtreme Gradient Boosting(XGBoost),k-Nearest Neighbor(KNN)and support vector machine(SVM)methods were used to construct the prediction models,and the results were confirmed in the validation dataset.Kaplan-Meier curves,Log-rank test and Cox proportional hazard model were applied to analyze the relationship between prediction models and diseasefree survival(DFS)in patients with breast cancer.Part Ⅲ.1488 patients with c T1-T2 stage breast cancer were included for analysis,with 615 in the training dataset,578 in the external validation dataset 1,and 295 in the external validation dataset 2.The study endpoint and MRI qualitative features evaluation as Part I.The 3D volumetric kinetic features of breast cancer were quantified based on DCE-MRI images,and a Kinetic score was constructed.A prediction model was constructed based on the MRI qualitative characteristics and Kinetic score.The relationship between the prediction model and RFS was analyzed using Kaplan-Meier curves,Log-rank test and Cox proportional hazard model.Part Ⅳ.Clinicopathological information and pre-and post-treatment MRI images were collected from 270 breast cancer patients who received neoadjuvant therapy.The study endpoint was p ALN status after neoadjuvant treatment.A logistic regression model was used to construct predictive models based on independent clinicopathological factors,pre-and posttreatment MRI qualitative features and quantitative 3D volumetric kinetic features to analyze their relationship with p ALN status and RFS after neoadjuvant treatment for breast cancer.Results:Part Ⅰ.BES was an independent predictor of p ALN burden in both the training and validation datasets.Analyses with MRI-ALN status alone showed that BES significantly improved the predictive performance of p ALN burden(training set: AUC: 0.65 vs 0.71,P < 0.001;IDI =0.045,P < 0.001;continuous NRI = 0.159,P = 0.050;validation set: AUC: 0.64 vs 0.69,P =0.009;IDI = 0.050,P < 0.001;continuous NRI = 0.213,P = 0.047)).Part Ⅱ.The prediction model constructed using the XGBoost algorithm based on clinicopathological factors and MRI qualitative features had better predictive performance than LR,SVM,and KNN models,with an AUC of 0.832(95% confidence interval [CI]: 0.789,0.876)in the training set and 0.838(95% CI: 0.775,0.901)in the validation set.After adjusting for other clinicopathological factors,the LVI-score established by the XGBoost model was an independent predictor of RFS.A high LVI-score was significantly associated with poorer DFS compared with a low LVI score(P = 0.014).Part Ⅲ.The DCE-MRI-based quantitative kinetic score was an independent predictor of p ALN burden in c T1-T2 stage breast cancer in the training dataset and the 2 independent validation datasets.The prediction model combining MRI qualitative features and kinetic score had an AUC of 0.779(95% CI: 0.734,0.824)in the training dataset,0.728(95% CI: 0.682,0.773)in the external validation dataset 1,and 0.755(95% CI: 0.680,0.830)in the external validation dataset 2.In addition,the score that established by the prediction model was also an independent predictor of RFS in early-staged breast cancer(P < 0.05).Part Ⅳ.The integrated model constructed by clinicopathological indicators,pre-and posttreatment MRI qualitative features and quantitative 3D volumetric kinetic features of DCEMRI achieved an AUC of(0.881;95%CI,0.837-0.925).In addition,both longitudial kinetic score and the score established by the integrated predictive model were independent predictors of RFS after neoadjuvant therapy for breast cancer(P < 0.05).Conclusion:The combination of breast MRI qualitative features and DCE-MRI based 3D volumetric quantitative kinetic features is expected to provide a reference basis for the precise management of axillary lymph nodes in breast cancer with different treatment options and prognostic assessment,which might help the selection of axillary surgery options for breast cancer patients and assist clinical diagnosis and treatment.