Exploring Radiomic Feature Groups Contributions in Recurrence Prediction of Breast Cancer: A Comparative Analysis of Multiple Machine Learning Models

The primary reason for the use of magnetic resonance imaging (MRI) in breast pre-treatment planning is its ability to reveal relevant information, which aids medical physicists in determining the most appropriate treatment strategy. The advancement of radiation therapy technologies and procedures has led to the development of more complex treatment options. As a result, the likelihood of treatment failure increases, thereby presenting heightened difficulties for medical physicists. Within the field of medical physics, radiomics assumes a crucial function in augmenting the process of treatment planning, monitoring treatment response, and predicting treatment outcomes. Consequently, this aids clinicians in customizing treatment strategies with the aim of enhancing patient outcomes. The objective of this study is to examine the role of different sets of radiomic features in the prediction of breast cancer recurrence in a patient cohort that exhibits significant heterogeneity. This cohort includes patients with various molecular subtypes, cancer stages, grades, and imaging parameters. In order to achieve this goal, a total of six machine learning algorithms were utilized. These algorithms include Logistic Regression, Support Vector Machine (SVM), Adaptive Boosting (AdaBoost), Extreme Gradient Boosting (XGBoost), Bagging, and Balanced Bagging. Prior to applying these algorithms, the most significant radiomic features were selected using the Recursive Feature Elimination (RFE) technique. The algorithms were assessed in terms of their predictive performance through the utilization of precision, recall, F1-score, and receiver operating characteristic (ROC) curves. The results indicated that the feature groups “Combining Tumor and FGT Enhancement” and “FGT Enhancement” were particularly prominent among the features that were ranked highly. In contrast, the groups labeled as “Tumor Enhancement Variation” and “Tumor Enhancement” exhibited a higher frequency among the chosen features, indicating a more pronounced correlation between characteristics related to tumors and the predictive outcome. All classifiers exhibited outstanding predictive abilities when employing the Synthetic Minority Over-sampling Technique (SMOTE), indicating that radiomic features may play a crucial role as prognostic factors in decision-making, even in a diverse patient population. This study makes a valuable contribution to the growing body of research on the utilization of artificial intelligence in the field of medical physics, with a specific focus on radiation therapy and medical imaging.