Ki-67 Prediction in Breast Cancer: Integrating Radiomics From Automated Breast Volume Scanner and 2D Ultrasound Images via Machine Learning

Wei Wei,1,2,* Fei Xia,3,* Wang Zhou,1,* Wenwu Lu,1 Di Zhang,1 Qianqing Ma,2 Xiangyi Xu,1 Chaoxue Zhang1 1Department of Ultrasound, The First Affiliated Hospital of Anhui Medical University, Hefei, Anhui, People’s Republic of China; 2Department of Ultrasound, The First Affiliated Hospital of Wannan Medical College (Yijishan Hospital), Wuhu, People’s Republic of China; 3Department of Ultrasound, WuHu Hospital, East China Normal University (The Second People’s Hospital, WuHu), Wuhu, Anhui, People’s Republic of China*These authors contributed equally to this workCorrespondence: Chaoxue Zhang, Department of Ultrasound, The First Affiliated Hospital of Anhui Medical University, No. 218 Jixi Road, Shushan District, Hefei, Anhui, People’s Republic of China, Email [email protected]: This study aimed to develop and validate a predictive model using radiomics features from automatic breast volume scanner (ABVS) and 2D ultrasound images to preoperatively assess Ki-67 expression in breast cancer (BC), thereby supporting personalized clinical treatment planning.Methods: Data from 426 BC patients who met the inclusion criteria were retrospectively analyzed. Univariate and multivariate logistic regression analyses were performed on the clinical ultrasound characteristics to construct a clinical model. Radiomics features were extracted from both the tumor and the sub-regions based on ABVS and 2D images. The silhouette coefficient was used to evaluate clustering performance and determine the optimal number of clusters. Radiomics-based prediction models were developed using four machine learning classifiers: Logistic Regression, ExtraTree, XGBoost, and LightGBM. A combined model was further constructed by integrating radiomics and habitat radiomics features from ABVS and 2D images with relevant clinical factors. Model performance was evaluated using the Receiver Operating Characteristic (ROC) curve, calibration curve, and decision curve analysis (DCA).Results: In the validation set, the area under the ROC curve (AUC) values of the radiomics model (RadABVS + 2D), the habitat radiomics model (Hab ABVS + 2D), and the combined radiomics model (Rad-HabABVS + 2D) were 0.603, 0.664, and 0.850, respectively. By integrating independent clinical factors (US-ALNs, T-stage) with the Rad-HabABVS + 2D model, a comprehensive model (CMClinical + Rad-Hab) was constructed using LightGBM. According to the DeLong test, this model significantly outperformed others in terms of AUC (P < 0.05). The AUC values for the training and validation sets were 0.951 (95% CI: 0.928– 0.973) and 0.884 (95% CI: 0.832– 0.949), respectively. The calibration curves and DCA of CMClinical + Rad-Hab demonstrated excellent model calibration and clinical utility.Conclusion: The CMClinical + Rad-Hab model developed in this study enables accurate preoperative prediction of Ki-67 expression in BC patients, facilitating personalized and precise treatment strategies.Keywords: breast cancer, Ki-67, automated breast volume scanner, radiomics, ultrasound