| Breast cancer is the most common malignancy among women all over the world.There has been great progress in the treatment and survival of breast cancer.The occurrence of axillary lymph node metastasis in breast cancer patients affects the prognosis and surgical methods as well as the follow-up treatment of breast cancer patients.With the continuous exploration of the treatment methods of breast cancer,neoadjuvant therapy has been widely used in breast cancer,which makes the efficacy of neoadjuvant therapy become an important indicator of the prognosis of patients.For breast cancer patients who do not achieve PCR,prognostic indicators related to prognosis are particularly important.Part one Prediction of lymph node status in patients with breast cancer based on multi-modal deep learningObjective: This paper used deep learning to extract features of clinical pathological data and whole slide imaging(WSI),and developed predictive models to predict early breast cancer lymph node metastasis.Methods: Collected the clinicopathological data(age,tumor side,tumor location,tumor size,histological type,histological grade,gland formation,nuclear atypia,mitosis counts,tumor infiltrating lymphocytes(Tils),tumor interstitial changes,the status of ER,PR and HER2,molecular typing,postoperative lymph node metastasis,etc.)and corresponding digital pathological images(Whole H&E slice digital scan image)of 4038 female patients with invasive breast cancer who had not undergone clinical treatment in the Fourth Hospital of Hebei Medical University.Extracted clinical pathological data features and image features(nuclear atypia,mitosis counts,Tils,etc.).Combined clinical pathological data with digital pathological image features,and used multi-modal learning models(MMLM)to train these two modal deep learning networks jointly,so that obtained the final model prediction results.The final model was tested by the test set and external data,and verified by the area under the ROC curve.The ROC curve was compared with the Delong test.Results: The deep learning model performance which predict lymphatic metastasis and further predict no metastasis,isolated tumor cells(ITC),micro-metastasis and macro-metastasis,was verified by test set.1.For the two categories of lymphatic metastasis(with or without metastasis),the AUC value of prediction based on clinical pathological indicators was 0.770;the AUC value based on WSI was 0.709;the prediction based on MMLM was0.809.In general,MMLM showed good prediction performance.2.Predicted the four categories of lymphatic metastasis status(no metastasis,ITC,micro-metastasis and macro-metastasis)by MMLM,subsequently compared the prediction performance based on clinical pathological indicators,WSI,and MMLM.The AUC values of no lymph metastasis were respectively 0.770,0.709,0.809;the AUC values of ITC were respectively 0.619,0.531,0.634;the AUC values of micro-metastasis were respectively 0.636,0.617,0.691;the AUC values of macro-metastasis were respectively 0.748,0.691,0.758.The comparison of AUC values among the three prediction methods revealed that the MMLM developed by combination the clinical pathological characteristics with WSI characteristics showed a more accurate prediction effect in the prediction of lymphatic metastasis status.Conclusion: The lymphatic metastasis prediction model of early breast cancer was developed,by means of multi-modal learning which combined clinical pathological information and digital pathological image features.The testing with all cases and all of molecular types revealed very high prediction performance and quite stable prediction results.Part two Development and validation of a novel model for predicting prognosis of non-p CR patients after neoadjuvant therapy for breast cancerObjective: This study aimed to construct a prediction model with more accurate and reliable prediction results by combining multiple clinicopathological factors,so as to provide a more accurate decision-making basis for subsequent clinical treatment.Methods: In this study,1009 cases of invasive breast cancer and surgically resected after neoadjuvant therapy from 2010 to 2017,and all patients did not achieve p CR.All indicators in this trial were interpreted in a double-blind manner by two pathologists with at least ten years of experience,including histological grading,Tils,ER,PR,HER2,and Ki67.The prediction model used R language to calculate the calibration degree and ROC curve of the prediction model in the training set and validation set.Results: Through univariate survival analysis,the results showed high histological grade(P=0.037),last clinical stage(P<0.001),HER2 negative(P=0.044),high residual cancer burden(RCB)grading(P<0.001),low Tils(P<0.001),lymphatic metastasis(P=0.049),low MP grade(P=0.013)were related to poor OS in non-PCR patients after neoadjuvant.All above data were analyzed by a multivariate analysis,and the results indicated that last clinical stage,HER2 negative,high RCB grading and low Tils were independently correlated with poor OS in non-PCR patients with breast cancer after neoadjuvant therapy.Among all cases in the training set,the prediction model predicted that the AUC value of 3-year survival rate was 0.95 and the AUC value of 5-year survival rate was 0.79;the AUC values of the prediction of RCB grading for 3-year and 5-year survival rate were 0.70 and 0.67 respectively,which proved that the prediction model showed better prediction performance than the RCB grading on the OS of non-PCR patients with breast cancer after neoadjuvant therapy.And the same results were acquired in HR,HER2+ and TN classifications: in HR,the AUC values of 3-year and 5-year survival rate calculated by predictive model were 0.97 and 0.83 respectively;in HER2+ were 0.99 and 0.86 respectively;in TN were 0.95 and 0.82 respectively.Predicted by the RCB classification,the AUC values of 3-year and 5-year survival rate were 0.79 and 0.75 respectively in HR;in HER2+were 0.77 and 0.64 respectively;in TN were 0.87 and 0.76 respectively.Conclusion: We combined 4 breast cancer prognosis-related factors through R software,developing a more accurate prediction model than RCB to evaluate the prognosis of non-p CR patients after neoadjuvant treatment of breast cancer,and to provide decision-making basis for further clinical treatment. |
PDF Full Text Request