Abstract:

Objective This study aims to investigate the value of a combined model integrating conventional ultrasound and radiomics for the preoperative prediction of the Luminal molecular subtype and Ki-67 expression status in patients with Luminal breast cancer. Methods A retrospective analysis was performed on ultrasound imaging data obtained from 375 patients diagnosed with breast cancer. Univariate and multivariate regression analyses were implemented first to identify independent predictive factors， and radiomic features were screened independently. Next， an integrated prediction model based on LightGBM was constructed by combining clinical ultrasound characteristics and radiomic features， to differentiate the Luminal molecular subtype of breast cancer and predict the Ki-67 expression status in patients with Luminal breast cancer. The performance of the established model was evaluated via receiver operating characteristic （ROC） curves and decision curve analysis （DCA）， and the SHapley Additive exPlanations （SHAP） method was applied to interpret the contribution of individual features to the model's prediction output. Results The combined model for Luminal molecular subtype discrimination achieved areas under the receiver operating characteristic curve （AUCs） of 0.836 and 0.776 in the training set and validation set， respectively. For the Ki-67 expression status prediction model constructed for patients with Luminal breast cancer， the AUC values reached 0.854 and 0.822 in the training cohort and validation cohort， respectively. Both models exhibited satisfactory predictive performance. Decision curve analysis demonstrated that the two models delivered substantial net clinical benefits across a broad range of threshold probabilities. SHAP （SHapley Additive exPlanation） analysis revealed that original_glcm_InverseVariance and wavelet.LH_firstorder_Median were the most impactful features for the two models respectively， confirming their critical contributions to the predictive performance of the models. Conclusion The stratified visualization model based on conventional ultrasound combined with radiomics shows potential clinical value for differentiating the Luminal molecular subtype and predicting Ki-67 expression status in breast cancer.

Key words: Luminal breast cancer, Ki-67, radiomics, shapley additive explanations