Abstract: Objective To investigate the predictive value of radiomics model based on cone⁃beam breast CT (CBBCT) images for pathological complete response (pCR) of neoadjuvant therapy in breast cancer. Methods CBBCT images from 106 female breast cancer patients who underwent neoadjuvant therapy in Guangxi Medical University Cancer Hospital from January 2022 to May 2023 were retrospectively analyzed. The patients were randomly divided into the training group and the validation group at a ratio of 8∶2. A total of 2, 264 radiomics features were extracted, and radiomics models were constructed through a cross⁃combination of feature selectors and machine⁃learning classifiers. The performance of the model was assessed using the receiver operating characteristic (ROC) curve, and decision curve analysis (DCA) was applied to evaluate the net benefits at different threshold probabilities between the training and the validation groups. Results The area under the curve (AUC) of L2 norm regularization⁃decision tree model demonstrated strong performance in the training group was  0.941(95%CI: 0.897-0.984), accuracy was 86.9%, specificity was 94.2%, and sensitivity was 75.0%. In the validation group, the model achieved an AUC of 0.732(95%CI: 0.518⁃0.947), accuracy of 72.7%, specificity of 85.7%, and sensitivity of 50.0%. Both the training and validation groups achieved the most net benefit. Conclusions The L2 norm regularization⁃decision tree prediction model based on CBBCT images demonstrates strong performance in predicting pCR of neoadjuvant therapy in breast cancer, which can provide valuable information for individualized treatment and timely adjustments to chemotherapy regimens.

Key words: Breast cancer, Cone?beam breast CT, Radiomics, Neoadjuvant therapy, Pathological response