Development of a Prognostic Model for Preoperative Stage I-III Breast Cancer Using Machine Learning with Integrated Cone-Beam Breast Computed Tomography Data in the Context of 3P Medicine

Background/Objectives: The lack of reliable prognostic predictors in breast cancer undermines the efficacy of its prediction, prevention, and personalized medicine (PPPM/3PM) approach. This study aimed to develop an integrated model based on cone-beam breast computed tomography (CBBCT) and hematological indicators to predict the prognosis of preoperative stage I-III breast cancer. Methods: A retrospective analysis was performed on 243 patients with pathologically confirmed stage I-III breast cancer. A new machine learning framework for feature selection integrated 10 machine learning algorithms and their 101 combinations. After feature selection, the patient risk score was calculated to construct a nomogram model for breast cancer prognosis. The nomogram model was evaluated using receiver operating characteristic (ROC) curve analysis and calibration curve. Univariate and multivariate logistic regression analyses verified the screened features and determined independent risk factors. Results: A machine learning computational framework based on 101 combinations selected 12 prognostic indicators of overall survival (OS) and 18 disease-free survivals (DFS) from 37 CBBCT and hematological features. The entire model achieved an AUC value of 0.837 in the training dataset and 0.813 in the validation dataset, which is superior to the clinical model without CBBCT indicators regarding OS prediction performance. Similarly, the AUC of the training and validation sets for DFS prediction was 0.996 and 0.732. Molecular typing, Enhancement curve types, and Morphology were independent risk factors associated with OS in the clinical prediction model. Calcification was an independent risk factor associated with DFS. We constructed a nomogram model combining the above features. Conclusions: Our study screened prognostic-related CBBCT and hematological features, and the nomogram showed satisfactory preoperative predictive efficacy for stage I-III breast cancer. It can be incorporated into the PPPM framework to help clinicians make more accurate treatment decisions.