Accuracy of distinguishing benign, high-risk lesions and malignancies with inductive machine learning models in BIRADS 4 and BIRADS 5 lesions on breast MR examinations.

Rationale And Objectives: The aim of this study is to explore the utility of Inductive Decision Tree models (IDTs) in distinguishing between benign, malignant, and high-risk (B3) breast lesions.

Materials And Methods: We analyzed 124 histologically confirmed lesions in 114 patients who underwent breast MR with BI-RADS 4 or 5 assessment. The dataset comprised 10 imaging parameters and one clinical observation. Using the IDTs method (algorithm C5.0 boosted with AdaBoost algorithm) combined with the data balancing method SMOTE (Synthetic Minority Oversampling Technique) and a corresponding new method called LCC (Leveling of Cases per Class), we developed corresponding 3-class classification models (Benign, B3, or Malignant). The training set used for classification model development consists of 112 cases with 12 variables, and the model's performance was assessed using 10-fold Cross-Validation and Leave-One-Out methods (utilizing the training set) and the Use Test Set method (testing on an unknown (for the models) dataset of 12 cases with 12 variables).

Results: This preliminary study demonstrates the potential for IDTs to accurately distinguish between benign, B3 and Malignant lesions based on extracted data from breast MRI exams with a high classification accuracy (88.70 %), mean sensitivity of 97.18 % and specificity of 98.59 % achieved by the optimal classification model, derived from the combination of the IDTs method and the LCC data balancing method.