Texture Analysis With 3.0T MRI For Early Prediction Of Pathologic Complete Response To Neoadjuvant Chemotherapy In Breast Cancer Patients

Objective:To investigate whether the use of breast dynamic enhancement MRI texture features to establish a mathematical model can predict breast cancer patients after neoadjuvant chemotherapy(NAC)pathological complete response(p CR).Method:Between November 2015 and February 2020,this retrospective study included 66 patients who underwent NAC and subsequently underwent breast cancer surgery(mean age 50.3 years;range 29-66years).After NAC,13 patients underwent breast-conserving surgery and 53 patients underwent total mastectomy.Before NAC,in the middle of NAC(determined according to the chemotherapy regimen,after 3 or 4 cycles of NAC treatment),the tumor was monitored with 3.0T MR imaging.By using texture analysis research software,texture analysis was performed on T2 WI,contrast-enhanced T1 WI,diffusion-weighted image(DWI)and apparent diffusion coefficient(ADC)maps before and during mid-treatment.A random forest method was used to build a prediction model,and p CR responders were classified with texture parameters.Using the 5-fold cross-validation method,the area under the ROC curve(AUC)was used to evaluate the performance of predicting p CR.Result:After neoadjuvant chemotherapy,18 of 66 patients(27.3%)achieved p CR.Whether the patient achieves p CR after neoadjuvant chemotherapy is related to the patient’s HER-2 status,whether to receive neoadjuvant targeted therapy,lymph node metastasis,the patient’s age,specific chemotherapy options,whether it is triple negative breast cancer,and the patient’s surgical method Irrelevant.Random forest classifier showed the lowest diagnostic performance of ADC on the pre-mid-term MRI changes of NAC(AUC = 0.50;95% confidence interval,range 0.43,0.61),and the highest diagnostic performance on T1 WI images enhanced by mid-term NAC(AUC = 0.82;95% confidence interval,range0.73-0.87).For the prediction of pathological complete response after neoadjuvant chemotherapy in breast cancer patients,comparing the AUC value of other images with the median AUC value of mid-term enhanced T1 WI,the diagnostic performance on the NAC mid-term enhanced T1 WI image is the best(confidence interval: 95%,P values <0.05).On the random forest model established by the mid-term enhanced T1 WI image,it is sorted according to the importance of the features,and the diagnostic performance of the RF models containing TOP3,6,9,12,15,and 17 features is not different from that of the RF model containing all features(confidenceinterval: 95%,P value> 0.05).On NAC mid-term enhanced T1 WI images,random forest classifier(AUC = 0.82,95% confidence interval,range: 0.73-0.87),k-nearest neighbor method(AUC = 0.79,confidence interval 95%,range0.71-0.84),naive bayesian model(AUC = 0.75,confidence interval 95%,range 0.64-0.83),adaboost classifier(AUC = 0.75,confidence interval 95%,range 0.66-0.82),decision tree classifier(AUC = 0.70,95% confidence interval,range 0.60-0.75).Random forest classifier shows better diagnostic performance.Conclusion:The texture parameters of T1 WI images enhanced in the mid-term of NAC treatment are important for the early prediction of pCR after NAC.