Pathomics signature based on machine learning can predict the response to neoadjuvant chemotherapy in breast cancer patients

Abstract:

Objective To develop a novel marker for predicting the response to neoadjuvant chemotherapy（NAC） in patients with breast cancer（BC） using the pathomics approach.Methods A retrospective analysis was performed for 211 patients with non-specific invasive BC in The Affiliated Hospital of Southwest Medical University，among whom 155 were enrolled as training group and 56 were enrolled as validation group. CellProfiler software was used to extract high-dimensional pathomics signature from the digital pathological sections of patients，and then the Mann-Whitney U test，the Spearman correlation coefficient，and the least absolute shrinkage and selection operator（LASSO） algorithm were used for the stepwise screening of features. The optimal features after screening were used to develop pathomics signature（PS） by the support vector machine（SVM） method in the training set and validate in the independent validation set. PS and significant clinicopathological factors（P<0.05） identified in the univariate analysis were included in the multivariate logistic regression analysis for further validation.Results PS had an area under the ROC curve of 0.749（95%CI=0.672-0.827） in the training set and 0.737（95%CI=0.604-0.870） in the validation set. The multivariate logistic regression analysis showed that PS ［odds ratio（OR）=2.317］ and human epidermal growth factor receptor 2（OR=4.018） were independent predictive factors for response to NAC in BC patients.Conclusion PS can help clinicians accurately predict the response to NAC before treatment and improve the personalized treatment for BC patients.