Standard Imaging And Longitudinal Radiomics Of Multiparametric MRI To Predict The Risk Of Axillary Lymph Nodes After Neoadjuvant Chemotherapy In Breast Cancer

Purpose:The high false negative rate(FNR)of sentinel lymph nodes biopsy(SLNB)after neoadjuvant adjuvant chemotherapy(NAC)might result in unnecessary axillary lymph node dissection(ALND).This study aimed to a multicenter radiomic model based on standard imaing and Longitudinal radiomics of multiparametric MRI(RMLM)to predict the status of axillary lymph nodes after neoadjuvant chemotherapy in breast cancer.Methods:1.The study reviewed a total of 257 patients with cN+breast cancer who underwent both sentinel lymph node biopsy(SLNB)and axillary lymph node dissection(ALND)following neoadjuvant chemotherapy(NAC).A logistic regression model was developed based on these factors and the results of post-NAC AT and axillary ultrasound(AUS);2.1359 eligible breast cancer patients with pre-NAC and post-NAC MR images were enrolled om four hospitals in the primary cohort(PC,n=409),external validation cohorts(EVC1,n=330;EVC2,n=170;EVC3,n=340)and prospective validation cohort(n=110).Patients were monitored by MRI prior to and after completing NAC.Longitudinal radiomics analysis was performed pre-and post-treatment using dynamic contrastenhanced imaging,T2-weighted imaging,and diffusion-weighted imaging mapping using 3D Slicer software.Here,we developed and validated a machine learning model based on multiple spatial radiomic features and longitudinal radiomics analysis to identify the metastatic risk of SLNs and NSLNs.Results:1.Four clinical factors with p<0.1 in the univariate analysis,including ycT0(p<0.001),clinical stage before NAC(p=0.025),estrogen receptor(ER)expression(p=0.009),and HER2 status(p=0.001),were independent predictors of NSLN metastases.The clinical model based on the above four factors resulted in the area under the curve(AUC)of0.82(95%CI:0.76-0.88)in the training set and 0.83(95%CI:0.74-0.92)in the validation set The results of post-NAC AUS and AT were added to the clinical model to construct a clinical imaging model for the prediction of NSLN metastasis with AUC of 0.87(95%CI:0.81-0.93)in the training set and 0.89(95%CI:0.82-0.96)in the validation set;2.In the primary cohort,the RMLM performed better than human performance in predicting the status of SLNs(sensitivity:89.8%vs.71.1%;NPV:95.2%vs.853%;AUC:0.86 vs.0.69).The combination model that incorporated human performance and radiomic signatures resulted in a satisfactory sensitivity of 96.9%in the PC,85.4%-96.9%in the EVCs and 71.4%in the prospective validation cohort.In terms of predicting NSLN status,the RMLM alone could not outperform retrieval of more than 2 SLNs in sensitivity(0.92,0.82 in PC;0.70-0.78,0.78-0.90 in EVC1-3;0.75,0.57 in prospective validation cohort)andAUC(0.90,0.81 in PC;0.77-0.84,0.83-0.86 in EVC1-3;0.76,0.74 in prospective validation cohort).The combination model achieved excellent results for the sensitivity(0.98 in PC,0.91-1 in EVC1-3,1 in the prospective validation cohort),the AUC(0.85 in PC,0.80-0.86 in EVC1-3,0.83 in the prospective validation cohort),and the FNR(0.6%in PC 0-2.8%in EVC1-3,0 in the prospective validation cohort).Conclusion:1.The study incorporated the results of post-NAC AT and AUS with other clinal factors to develop a model to predict NSLN metastasis in patients with initial cN+before surgry;2.the RMLM had a high predictive ability to evaluate the status of ALNs and NSLNs,which could be helpful in determining appropriate axillary treatment.