[Objective]By integrating multi-parameter magnetic resonance imaging(MRI)radiomics features,an interpretable machine learning model was constructed to accurately predict pre-operative traditional Chinese medicine(TCM)syndromes in breast cancer patients,providing objective and quantitative basis for clinical TCM syndrome types differentiation.[Methods]A total of 315 patients pathologically diagnosed breast cancer from the First Affiliated Hospital of Zhejiang Chinese Medical University between June 2019 to October 2023 were retrospectively included.The preoperative TCM syndromes of the patients were classified into liver depression with phlegm coagulation,chongren dysregulation and positive deficiency blazing toxin type.These patients were randomly allocated to the training(221 cases)and validation sets(94 cases)in a ratio of 7:3.Radiomics features in dynamic contrast-enhanced imaging(DCE),diffusion-weighted imaging(DWI)and apparent diffusion coefficient(ADC)images of each patient were extracted by using Kruskal-Wallis rank sum test,Spearman correlation analysis and least absolute shrinkage and selection operator(LASSO)polynomial Logistic regression screening for radiomics features.Then,light gradient boosting machine(LightGBM)was used to construct the syndrome prediction model.[Results]A total of 2 832 radiomics features were extracted from the DCE,DWI and ADC images of each patient.After a series of feature selection methods,the total numbers of features for DCE,DWI,ADC and three-sequence combination were 3,3,3 and 7 respectively.The machine learning model integrating multi-sequence MRI image information had the best performance,with micro-averaged area under the curve(AUC)of 0.852 and 0.838 for the training and validation sets,and macro-averaged AUC of 0.833 and 0.810 for the training and validation sets respectively.Shapley additive explanation(SHAP)analysis showed that inverse difference moment normalization(Idmn)of the ADC sequence,average gray level(Mean)of the ADC sequence and inverse variance of the DCE sequence were the three features that contributed most to the prediction results.[Conclusion]The interpretable machine learning model constructed in this study can accurately predict pre-operative TCM syndromes in breast cancer patients.SHAP interpretability analysis provides quantitative contribution value information,which can help clinicians understand the model prediction process and provide a new perspective for the application of traditional Chinese medicine in the treatment of breast cancer.

[Objective]By integrating multi-parameter magnetic resonance imaging(MRI)radiomics features,an interpretable machine learning model was constructed to accurately predict pre-operative traditional Chinese medicine(TCM)syndromes in breast cancer patients,providing objective and quantitative basis for clinical TCM syndrome types differentiation.[Methods]A total of 315 patients pathologically diagnosed breast cancer from the First Affiliated Hospital of Zhejiang Chinese Medical University between June 2019 to October 2023 were retrospectively included.The preoperative TCM syndromes of the patients were classified into liver depression with phlegm coagulation,chongren dysregulation and positive deficiency blazing toxin type.These patients were randomly allocated to the training(221 cases)and validation sets(94 cases)in a ratio of 7:3.Radiomics features in dynamic contrast-enhanced imaging(DCE),diffusion-weighted imaging(DWI)and apparent diffusion coefficient(ADC)images of each patient were extracted by using Kruskal-Wallis rank sum test,Spearman correlation analysis and least absolute shrinkage and selection operator(LASSO)polynomial Logistic regression screening for radiomics features.Then,light gradient boosting machine(LightGBM)was used to construct the syndrome prediction model.[Results]A total of 2 832 radiomics features were extracted from the DCE,DWI and ADC images of each patient.After a series of feature selection methods,the total numbers of features for DCE,DWI,ADC and three-sequence combination were 3,3,3 and 7 respectively.The machine learning model integrating multi-sequence MRI image information had the best performance,with micro-averaged area under the curve(AUC)of 0.852 and 0.838 for the training and validation sets,and macro-averaged AUC of 0.833 and 0.810 for the training and validation sets respectively.Shapley additive explanation(SHAP)analysis showed that inverse difference moment normalization(Idmn)of the ADC sequence,average gray level(Mean)of the ADC sequence and inverse variance of the DCE sequence were the three features that contributed most to the prediction results.[Conclusion]The interpretable machine learning model constructed in this study can accurately predict pre-operative TCM syndromes in breast cancer patients.SHAP interpretability analysis provides quantitative contribution value information,which can help clinicians understand the model prediction process and provide a new perspective for the application of traditional Chinese medicine in the treatment of breast cancer.