Abstract:

Objective To build the machine learning models based on ultrasonographic features to predict the Luminal type of invasive breast cancer，and to screen out the important ultrasonographic characteristics that influence the Luminal type. Methods The ultrasonographic and immunohistochemical features of 529 patients with postoperative pathology⁃diagnosed invasive breast cancer were studied retrospectively. The patients were divided into two groups based on immunohistochemical features：luminal and non⁃luminal. Using SPSS Modeler 18.0 statis⁃ tical software，patients were randomly divided into training cohort and validation cohort，and five classifiers were used to build the models，including Logistic regression analysis，support vector machine （SVM），Bayesian network，random forest，and decision tree. The ROC curves were drawn according to the original propensity score of the models. The AUCs were used to evaluate the predictive ability of different models ，and DeLong test was used to compare the AUCs of the five models. The best model was used to screen out the most important variables for predicting the Luminal type. Results Logistic regression analysis and SVM outperformed other models in both the training and validation cohorts，and the difference was statistically significant（P < 0.05）. The three most important variables for predicting the Luminal type，according to the SVM model，were diameter，margin，and calcification. Conclusion Machine learning models based on ultrasonographic features can accurately predict the Luminal type of invasive breast cancer，and SVM outperforms Logistic regression analysis in terms of predictive ability. The three most important ultrasonographic features for predicting Luminal type are diameter，margin，and calcification. 

Key words:

 , machine learning； invasive breast cancer； Luminal type； ultrasound